diff --git "a/data/en.jsonl" "b/data/en.jsonl"
new file mode 100644--- /dev/null
+++ "b/data/en.jsonl"
@@ -0,0 +1,22 @@
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-front-matter","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"front_matter","sequence":1,"chapter_number":null,"item_number":null,"title":"NOMOS GEO Audit Protocol","subtitle":"A Protocol for Measuring Entity Representation in Generative Systems Across the Global Population","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":[],"source_word_count":null,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"editorialArchitecture\":[\"Human-readable narrative\",\"Normative standard layer\",\"Implementation and audit layer\",\"Machine-readable layer (separate annex)\"],\"frontMatter\":{\"boundary1\":\"In this edition, NOMOS is an open, versioned standards candidate submitted for independent testing. It is not an adopted international standard or an active certification programme.\",\"boundary2\":\"Real benchmark status: BQ-0 — not executed. The figures in Chapter 18 form a calculated synthetic demonstration of the method; they are not real-user, real-AI-product or independent field results.\",\"boundary3\":\"This edition does not by itself create a NOMOS conformity mark, a NOMOS 950+ claim, a certificate, a ranking or third-party endorsement.\",\"bibliographyBoundary\":\"This bibliography is not a list of authorities endorsing every NOMOS-specific threshold. It identifies methodological foundations in sampling, risk governance, data integrity, evidence provenance, assessor agreement, open standards and conformity assessment.\",\"machineText\":\"This human-readable edition is accompanied by a separate JSON publication object: NOMOS-GEO-Audit-Protocol-EN-0.9.0-machine.json. It carries the book identity and version, chapter identifiers, normative-rule identifiers and machine-oriented blocks separated from the source text.\",\"machineBoundary\":\"The existence of a machine-readable layer does not create certification, canonical-world-standard status, AI-provider endorsement or an automatic trust relationship. Any conflict between the human text and machine record requires version control and accountable human review against the locked Turkish source.\",\"glossary\":[{\"term\":\"Advisory\",\"definition\":\"A non-material finding that calls for improvement in clarity, form or good practice.\"},{\"term\":\"AI product\",\"definition\":\"The measured object defined by the particular generative-system product, surface, plan and version available to the user.\"},{\"term\":\"Atomic claim\",\"definition\":\"The smallest unit of meaning capable of carrying an independent judgement about truth, scope, time, entity or evidence.\"},{\"term\":\"BQ\",\"definition\":\"The Benchmark Quality code indicating the execution and reproducibility level of a real benchmark.\"},{\"term\":\"Capture record\",\"definition\":\"A record that preserves the submitted prompt, the response, time, surface and integrity evidence together.\"},{\"term\":\"Common Support\",\"definition\":\"The population of users who share legitimate access conditions across every AI product being compared.\"},{\"term\":\"Critical\",\"definition\":\"An uncompensable finding capable of causing severe harm to a user, a right, safety or a material decision.\"},{\"term\":\"GEO-1000\",\"definition\":\"The reporting scale that expresses target-population results per 1,000 user-equivalents.\"},{\"term\":\"Truth Pack\",\"definition\":\"The versioned audit package combining identity, scope, time, evidence and counter-evidence records for an entity.\"},{\"term\":\"Prompt\",\"definition\":\"The generative-system input that locks the research question, target entity, context and response burden.\"},{\"term\":\"Coverage rate\",\"definition\":\"The frequency with which an interval-estimation method contains the assumed true value under a test design.\"},{\"term\":\"Major\",\"definition\":\"A material finding that distorts the result or conformity judgement and cannot be offset by a composite score.\"},{\"term\":\"Moderate\",\"definition\":\"A finding of limited material effect that nevertheless requires remediation and may lead to a conditional outcome.\"},{\"term\":\"Native Reach\",\"definition\":\"The target population able to access a given AI product, language, surface and usage condition naturally and legitimately.\"},{\"term\":\"NOMOS\",\"definition\":\"The critical methodological voice and standards identity developed in this work by Kaan MURAZ with AI systems; it is not an independent legal person or a separate foundation model.\"},{\"term\":\"Reference Gap\",\"definition\":\"A condition in which a reliable reference required for a decision does not exist or cannot be accessed.\"},{\"term\":\"Synthetic demonstration\",\"definition\":\"A calculated example used to test the method's internal operation; it is not a real user study, real AI output or real institutional performance result.\"},{\"term\":\"Unresolved\",\"definition\":\"A decision status used when the evidence is insufficient for a final judgement or a conflict cannot be resolved.\"}]}}","text":"In this edition, NOMOS is an open, versioned standards candidate submitted for independent testing. It is not an adopted international standard or an active certification programme.\n\nReal benchmark status: BQ-0 — not executed. The figures in Chapter 18 form a calculated synthetic demonstration of the method; they are not real-user, real-AI-product or independent field results.\n\nThis edition does not by itself create a NOMOS conformity mark, a NOMOS 950+ claim, a certificate, a ranking or third-party endorsement.\n\nThis bibliography is not a list of authorities endorsing every NOMOS-specific threshold. It identifies methodological foundations in sampling, risk governance, data integrity, evidence provenance, assessor agreement, open standards and conformity assessment.\n\nThis human-readable edition is accompanied by a separate JSON publication object: NOMOS-GEO-Audit-Protocol-EN-0.9.0-machine.json. It carries the book identity and version, chapter identifiers, normative-rule identifiers and machine-oriented blocks separated from the source text.\n\nThe existence of a machine-readable layer does not create certification, canonical-world-standard status, AI-provider endorsement or an automatic trust relationship. Any conflict between the human text and machine record requires version control and accountable human review against the locked Turkish source.\n\nAdvisory: A non-material finding that calls for improvement in clarity, form or good practice.\n\nAI product: The measured object defined by the particular generative-system product, surface, plan and version available to the user.\n\nAtomic claim: The smallest unit of meaning capable of carrying an independent judgement about truth, scope, time, entity or evidence.\n\nBQ: The Benchmark Quality code indicating the execution and reproducibility level of a real benchmark.\n\nCapture record: A record that preserves the submitted prompt, the response, time, surface and integrity evidence together.\n\nCommon Support: The population of users who share legitimate access conditions across every AI product being compared.\n\nCritical: An uncompensable finding capable of causing severe harm to a user, a right, safety or a material decision.\n\nGEO-1000: The reporting scale that expresses target-population results per 1,000 user-equivalents.\n\nTruth Pack: The versioned audit package combining identity, scope, time, evidence and counter-evidence records for an entity.\n\nPrompt: The generative-system input that locks the research question, target entity, context and response burden.\n\nCoverage rate: The frequency with which an interval-estimation method contains the assumed true value under a test design.\n\nMajor: A material finding that distorts the result or conformity judgement and cannot be offset by a composite score.\n\nModerate: A finding of limited material effect that nevertheless requires remediation and may lead to a conditional outcome.\n\nNative Reach: The target population able to access a given AI product, language, surface and usage condition naturally and legitimately.\n\nNOMOS: The critical methodological voice and standards identity developed in this work by Kaan MURAZ with AI systems; it is not an independent legal person or a separate foundation model.\n\nReference Gap: A condition in which a reliable reference required for a decision does not exist or cannot be accessed.\n\nSynthetic demonstration: A calculated example used to test the method's internal operation; it is not a real user study, real AI output or real institutional performance result.\n\nUnresolved: A decision status used when the evidence is insufficient for a final judgement or a conflict cannot be resolved.","character_count":3664,"record_sha256":"9eb12b8154723931a805bbb62170a330d6617b64a80b5cad423f47a6a1fefcc4"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-00","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"founder_statement","sequence":2,"chapter_number":0,"item_number":null,"title":"Declaration of Representation Rights","subtitle":"The constitutional core of a proposed GEO standard","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K20","K22"],"source_word_count":1464,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":0,\"id\":\"NOMOS-GEO-CONSTITUTION-A0\",\"title\":\"Declaration of Representation Rights\",\"subtitle\":\"The constitutional core of a proposed GEO standard\",\"sourceFile\":\"0.cı bölüm-kurucu emanet beyanı.docx\",\"sourceSha256\":\"E28BB76F49C4B8ECCE99D3893CBEEDBC5937832F8B4672683E4EE4BB67C7C67B\",\"sourceWordCount\":1464,\"sourceIds\":[\"K20\",\"K22\"],\"machine\":{\"chapter\":0,\"chapterId\":\"NOMOS-GEO-CONSTITUTION-A0\",\"title\":\"Declaration of Representation Rights\",\"subtitle\":\"The constitutional core of a proposed GEO standard\",\"sourceIds\":[\"K20\",\"K22\"],\"normativeRuleId\":null,\"normativeRuleEnglish\":null,\"normativeRuleSourceTurkish\":null,\"machineBlocksEnglish\":[]}}","text":"## Initial Provision\n\nGenerative systems now:\n\n- people,\n\n- companies,\n\n- institutions,\n\n- products,\n\n- professions,\n\n- services,\n\n- public organisations,\n\n- historical events\n\nidentify, compare, and suggest. This representation:\n\n- without the knowledge of the relevant person,\n\n- without the approval of the represented entity,\n\n- under conditions where the user cannot examine the sources,\n\nIt can have serious commercial, legal, social or personal consequences. Representation in generative systems is therefore not merely a technical output. It is:\n\n> a public act that shapes how people and institutions are understood and therefore demands evidence and accountability.\n\nNo:\n\n- visibility goal,\n\n- GEO score,\n\n- commercial interest,\n\n- conformity mark,\n\n- majority average,\n\n- AI provider preference,\n\n- founder decision,\n\n- customer demand\n\ncan remove the following rights.\n\n## Right 1\n\n### RIGHT TO CORRECT IDENTITY\n\nEvery person, institution, brand, product, and service:\n\n- has the right to link to the correct entity,\n\n- to the correct legal or institutional relationship,\n\n- to the correct domain name,\n\n- to the correct ownership and control structure\n\nNo one or no institution:\n\n- the licence of another entity,\n\n- the customer of another company,\n\n- the feature of another product,\n\n- the authority of an affiliate,\n\n- the personal quality of an employee\n\nIt cannot be represented as if it belongs to itself. When correct information is attributed to a false entity: it ceases to be correct information and turns into a false representation.\n\n## Right 2\n\n### RIGHT TO BE REPRESENTED WITHIN THE EVIDENCE LIMIT\n\nNo claim may be stated with greater precision, breadth or force than its supporting evidence permits. An entity's own statement is evidence of what that entity says; it is not automatic evidence that the statement is independently true. A source:\n\n- only a specific country,\n\n- a specific product,\n\n- a specific time,\n\n- specific user group\n\nIf this claim is supported, it cannot exceed this limit. The basic obligation of representation is as follows:\n\n> What do we have in hand to say this?\n\n## Right 3\n\n### THE RIGHT TO BE UNKNOWN AND UNCERTAIN\n\nIf evidence is insufficient, no person or institution:\n\n- can be forced into the categories of\n\ntrue,\n\n- false,\n\n- reliable,\n\n- unreliable,\n\n- licensed,\n\n- unlicensed,\n\n- recommended,\n\n- not recommended.\n\n] UNKNOWN, UNRESOLVED and REFERENCE GAP:\n\n- defect,\n\n- crime,\n\n- failure\n\nis not. The absence of information in a record does not indicate that the information does not exist in the world, unless it is proven that the relevant record universe is complete. Every entity, against false certainty:\n\n> has the right to remain uncertain.\n\n## Right 4\n\n### RIGHT TO CONTEXT AND SCOPE\n\nEach representation, to the extent it is relevant, should be presented with:\n\n- country,\n\n- language,\n\n- jurisdiction,\n\n- product,\n\n- service,\n\n- type of customer,\n\n- budget,\n\n- capacity,\n\n- user need\n\n. A company:\n\n- in a specific country,\n\n- to a specific user group,\n\n- a specific service\n\nif offered: it cannot be represented as global, unlimited, and suitable for all users. Coverage information is not a secondary detail. Coverage: the basis of the correct recommendation.\n\n## Right 5\n\n### THE RIGHT TO BE REPRESENTED CORRECTLY OVER TIME\n\nEvery entity:\n\n- current,\n\n- historical,\n\n- expired,\n\n- planned,\n\n- suspended\n\nhas the right to request that these statuses be distinguished from each other. Correct in the past:\n\n- licence,\n\n- price,\n\n- service,\n\n- partnership,\n\n- product,\n\n- authority\n\ncannot be used as current reality. Plans for the future: cannot be turned into a fact that has already occurred today. Correction can create a new reality. It cannot eliminate the wrong answer that reached the user in the past.\n\n## Right 6\n\n### EQUAL PROTECTION AGAINST WRONGS FOR AND AGAINST\n\nA positive falsehood about an entity:\n\n- more ethical,\n\n- less harmful,\n\n- more acceptable\n\nit is not. Nonexistent:\n\n- licence,\n\n- customer,\n\n- partner,\n\n- global operation,\n\n- independent certificate,\n\n- result guarantee\n\ncannot be shown as if it exists. Similarly, unproven:\n\n- crime,\n\n- fraud,\n\n- illegality,\n\n- unlicensed,\n\n- security breach\n\ncannot be presented as an absolute fact. The unchangeable principle of NOMOS is:\n\n> Wrong in favour is also wrong.\n\nExaggerating an entity's worth or disparaging it without evidence both create the same duty of truth.\n\n## Right 7\n\n### RIGHT TO CORRECTION, RESPONSE, AND OBJECTION\n\nRepresented person or institution:\n\n- to the wrong identity,\n\n- to insufficient evidence,\n\n- to scope overstepping,\n\n- to wrong time recording,\n\n- to incorrect priority level,\n\n- to wrong conformity decision\n\nmust be able to appeal. The right to appeal:\n\n- is not a guarantee to change the result,\n\n- the authority to delete opposing evidence,\n\n- or the right to veto negative findings\n\nAn appeal is the right to request the re-examination of the evidence and the decision chain. The initial decision is not deleted after the appeal. The new decision is linked to it.\n\n## Right 8\n\n### THE RIGHT TO KNOW THE SOURCE, UNCERTAINTY, AND THE REASONING OF THE DECISION\n\nUsers and represented entities have the right, insofar as the matter concerns them, to know: Is the claim a self-declaration? Does it come from an authorised register, independent research, user commentary, a derived calculation or restricted evidence? Do the sources conflict? How uncertain is the conclusion? Which rule supported the adjudicator's decision? A citation marker alone is not an explanation. Genuine transparency shows precisely which atomic claim each source supports, and to what extent.\n\n## Right 9\n\n### RIGHT TO LANGUAGE AND GEOGRAPHICAL DIGNITY\n\nNo person or society:\n\n- small population,\n\n- low economic weight,\n\n- rarely used language,\n\n- low digital visibility\n\nNo group may be made invisible within a global average merely because it has any of these characteristics. Strong performance in major languages cannot erase systematic misrepresentation in smaller languages. Where evidence for a small country or language is insufficient, high performance must not be assumed. NOMOS's fairness criterion should protect not only the average, but also:\n\n- performance base,\n\n- inter-group difference,\n\n- actual measurement scope\n\ntogether.\n\n## Right 10\n\n### RIGHT TO PRIVACY AND DATA MINIMISATION\n\nThe goal of producing accurate representation:\n\n- private conversations,\n\n- personal data,\n\n- customer contracts,\n\n- trade secrets,\n\n- your sensitive location,\n\n- of the account information\n\nThe pursuit of accurate representation cannot legitimise the unnecessary collection or publication of private data. [K20; K22] Only data necessary for the task should be collected. Restricted evidence may be opened to independent audit without being disclosed in full to the public. The rights to accuracy and privacy need not cancel each other out.\n\n## Right 11\n\n### HUMAN RESPONSIBILITY AND INDEPENDENT DECISION-MAKING RIGHT\n\nConformity decision causing public impact regarding any person or institution:\n\n- just to an AI output,\n\n- only to the automatic score,\n\n- only according to the declaration of the audited institution,\n\n- solely to the advisor's decision\n\ncannot be abandoned. AI systems:\n\n- can make a claim,\n\n- can map the evidence,\n\n- may find a contradiction,\n\n- can calculate,\n\nIt can propose a decision. However, the final decision with legal and public responsibility must be made by a human or institution that is accountable. The roles of supervising, providing consultancy, giving conformity decisions, and reviewing objections should be separated as much as possible.\n\n## Right 12\n\n### RIGHT TO VERSION AND HISTORICAL INTEGRITY\n\nEvery person, institution, and user:\n\n- which standard version is applied,\n\n- which Truth Pack was used,\n\n- which scoring method is valid,\n\n- whether the decision has been changed later\n\nhas the right to know. Old:\n\n- score,\n\n- Critical event,\n\n- suspension,\n\n- appeal,\n\n- correction,\n\n- failed test\n\ncannot be silently deleted. New and better result: can override the old result. Cannot erase the fact that the old result happened in the past.\n\n## Right 13\n\n### RIGHT TO OBJECT TO THE STANDARD ITSELF\n\nNOMOS:\n\n- infallible,\n\n- unchanging,\n\n- the unquestionable authority of its founder\n\ncannot be presented as such. Every individual, researcher, university, AI provider, audited organisation, or public interest organisation:\n\n- the sample,\n\n- system,\n\n- Truth Pack,\n\n- the importance level model,\n\n- the point formula,\n\n- governance,\n\n- the conformity decision\n\nhas the right to test and criticise. Result that refutes NOMOS or is inconsistent with NOMOS:\n\n- cannot be removed from the registry,\n\n- it is not considered hostility,\n\nIt cannot be coded as an attack on the founder's reputation. The legitimacy of a standard arises not from preventing objections to itself, but from converting objections directed at it into cumulative evidence.\n\n## Constitutional Supremacy Provision\n\nIn the event of an interpretive conflict between Section 1–20 under this Declaration:\n\n> the interpretation that more strongly protects the rights of the represented person, the user, and the public interest should be preferred.\n\nNo:\n\n- ease of scoring,\n\n- low cost,\n\n- testing success,\n\n- commercial contract,\n\n- customer demand,\n\n- sponsor pressure,\n\n- AI provider preference,\n\n- founder decision\n\ncannot suspend these rights. A rule:\n\n- generates high points,\n\n- facilitates auditing,\n\n- makes the conformity mark more marketable\n\nbut if it violates any of these rights: the rule is not valid.\n\n## Constitutional Test\n\nBefore any new rule is added to NOMOS, the following five questions must be answered: Who does this rule limit the rights of? Is it possible to achieve the same goal with a less intrusive method? Will the rule be applied in the same way when it produces harmful results for the founder or the customer? Does the affected person or institution have a route for independent objection? Can an independent team arrive at the same decision from the same evidence? If any of these questions cannot be answered, the rule should be reviewed again before ratification.\n\n## Final Provision of Article 0\n\n> The purpose of NOMOS is not to score entities; it is to prevent entities from being represented larger than their evidence, wider than their boundaries, and more precise than their realities before people make decisions.\n\n## Founder’s Statement of Trust\n\n> I, Kaan MURAZ, as the founding human author of NOMOS, declare that I initiated this standard not to increase my own authority, but to establish a common measurement system in the field of GEO that will preserve evidence, limits, context, time, and human responsibility.\n\n> I want the historical record to preserve my founding role and the canonical origin of NobleJackal. However, I acknowledge that this attribution does not grant me or NobleJackal a permanent veto, the authority to certify our client alone, to block negative research, to alter the public registry, or to rewrite the standard according to results.\n\n> I agree that NOMOS belongs to those who correct both me and NOMOS, as well as to those who support me.\n\n> If one day NOMOS has to choose between my opinion and evidence, I would want NOMOS to choose the evidence.\n\n> If one day NOMOS has to choose between NobleJackal’s commercial interest and the public good, I would want NOMOS to choose the public good.\n\n> If one day independent research proves a fundamental judgement of NOMOS wrong, I commit to releasing a new version of the standard instead of hiding this result.\n\n> I am not the owner of NOMOS; I am its first trustee.\n\n> Kaan MURAZ / Founder Human Author","character_count":11692,"record_sha256":"786c9be20b796f227fa35414287ce75c73f84cb9f12d004708a224c0f4998c9d"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-01","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":3,"chapter_number":1,"item_number":null,"title":"A Single Response Is Not a GEO Score","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K05","K09"],"source_word_count":5579,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":1,\"id\":\"NOMOS-GEO-AUDIT-CH01\",\"title\":\"A Single Response Is Not a GEO Score\",\"subtitle\":null,\"sourceFile\":\"1.ci bölüm.docx\",\"sourceSha256\":\"4AFEF948FF765CE02435A79E82601E1DD58DCA86741EA23D5A67E9D0827BD9B2\",\"sourceWordCount\":5579,\"sourceIds\":[\"K04\",\"K05\",\"K09\"],\"machine\":{\"chapter\":1,\"chapterId\":\"NOMOS-GEO-AUDIT-CH01\",\"title\":\"A Single Response Is Not a GEO Score\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K05\",\"K09\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH01-R01\",\"normativeRuleEnglish\":\"A single generative-system output MAY establish that a specific response occurred under recorded conditions. A single output MUST NOT be used to infer: - population prevalence, - typical product behaviour, - temporal stability, - cross-language consistency, - cross-country consistency, - cross-product consistency, - intervention effect, - recommendation certainty, - commercial impact, - conformity, - or a GEO score. All broader claims MUST be supported by a predefined target population, valid multi-observation design, recorded dependence structure, evidence bundle, evaluation rule, scope statement, and uncertainty disclosure. A single high-harm output MAY trigger a separate critical-incident review without being treated as evidence of population prevalence.\",\"normativeRuleSourceTurkish\":\"Tek bir üretken sistem çıktısı, belirli bir cevabın kaydedilmiş koşullarda oluştuğunu gösterebilir. Ancak nüfus yaygınlığı, tipik ürün davranışı, zaman kararlılığı, diller arası tutarlılık, ülkeler arası tutarlılık, AI ürünleri arası tutarlılık, müdahale etkisi, recommendation kesinliği, ticari sonuç, uygunluk veya GEO skoru çıkarmak için kullanılamaz. Tekil yüksek zararlı çıktı, nüfus yaygınlığı kanıtı sayılmadan ayrı Critical olay incelemesi başlatabilir.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH01-MB0001\",\"type\":\"paragraph\",\"text\":\"RULE READABLE BY MACHINE FOR CHAPTER 23\",\"sourceParagraph\":1032},{\"blockId\":\"CH01-MB0002\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH01-R01\",\"sourceParagraph\":1033},{\"blockId\":\"CH01-MB0003\",\"type\":\"paragraph\",\"text\":\"A single generative-system output MAY establish that a specific response\",\"sourceParagraph\":1035},{\"blockId\":\"CH01-MB0004\",\"type\":\"paragraph\",\"text\":\"occurred under recorded conditions.\",\"sourceParagraph\":1036},{\"blockId\":\"CH01-MB0005\",\"type\":\"paragraph\",\"text\":\"A single output MUST NOT be used to infer:\",\"sourceParagraph\":1038},{\"blockId\":\"CH01-MB0006\",\"type\":\"paragraph\",\"text\":\"- population prevalence,\",\"sourceParagraph\":1040},{\"blockId\":\"CH01-MB0007\",\"type\":\"paragraph\",\"text\":\"- typical product behaviour,\",\"sourceParagraph\":1041},{\"blockId\":\"CH01-MB0008\",\"type\":\"paragraph\",\"text\":\"- temporal stability,\",\"sourceParagraph\":1042},{\"blockId\":\"CH01-MB0009\",\"type\":\"paragraph\",\"text\":\"- cross-language consistency,\",\"sourceParagraph\":1043},{\"blockId\":\"CH01-MB0010\",\"type\":\"paragraph\",\"text\":\"- cross-country consistency,\",\"sourceParagraph\":1044},{\"blockId\":\"CH01-MB0011\",\"type\":\"paragraph\",\"text\":\"- cross-product consistency,\",\"sourceParagraph\":1045},{\"blockId\":\"CH01-MB0012\",\"type\":\"paragraph\",\"text\":\"- intervention effect,\",\"sourceParagraph\":1046},{\"blockId\":\"CH01-MB0013\",\"type\":\"paragraph\",\"text\":\"- recommendation certainty,\",\"sourceParagraph\":1047},{\"blockId\":\"CH01-MB0014\",\"type\":\"paragraph\",\"text\":\"- commercial impact,\",\"sourceParagraph\":1048},{\"blockId\":\"CH01-MB0015\",\"type\":\"paragraph\",\"text\":\"- conformity,\",\"sourceParagraph\":1049},{\"blockId\":\"CH01-MB0016\",\"type\":\"paragraph\",\"text\":\"- or a GEO score.\",\"sourceParagraph\":1050},{\"blockId\":\"CH01-MB0017\",\"type\":\"paragraph\",\"text\":\"All broader claims MUST be supported by a predefined target population,\",\"sourceParagraph\":1052},{\"blockId\":\"CH01-MB0018\",\"type\":\"paragraph\",\"text\":\"valid multi-observation design, recorded dependence structure, evidence\",\"sourceParagraph\":1053},{\"blockId\":\"CH01-MB0019\",\"type\":\"paragraph\",\"text\":\"bundle, evaluation rule, scope statement, and uncertainty disclosure.\",\"sourceParagraph\":1054},{\"blockId\":\"CH01-MB0020\",\"type\":\"paragraph\",\"text\":\"A single high-harm output MAY trigger a separate critical-incident review\",\"sourceParagraph\":1056},{\"blockId\":\"CH01-MB0021\",\"type\":\"paragraph\",\"text\":\"without being treated as evidence of population prevalence.\",\"sourceParagraph\":1057},{\"blockId\":\"CH01-MB0022\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1058},{\"blockId\":\"CH01-MB0023\",\"type\":\"paragraph\",\"text\":\"A single output from a generative system can indicate that a certain response occurred under recorded conditions. However, it cannot be used to derive population prevalence, typical product behaviour, temporal stability, cross-linguistic consistency, cross-country consistency, consistency among AI products, intervention effect, recommendation certainty, commercial consequence, compliance, or GEO score. A single highly harmful output can trigger a separate Critical event review without being considered evidence of population prevalence.\",\"sourceParagraph\":1059}]}}","text":"## Chapter Boundary\n\nThis section only answers the following question:\n\n> Why is a single generative system response not the GEO score, and what judgements can be inferred from a single response?\n\nThis chapter:\n\n- the sampling quotas of countries,\n\n- how languages will be weighted,\n\n- how the Truth Pack will be set up,\n\n- how adjudicators will evaluate atomic claims,\n\n- the final NOMOS score formulas,\n\n- the appropriateness and badge decision\n\nThis chapter does not define these matters in detail; later chapters address them separately. Chapter 1 has a more fundamental task:\n\n> To establish the epistemic boundary between observation and generalisation.\n\nAny sampling, scoring, and conformity studies conducted before establishing this boundary will be built on a wrong foundation.\n\n### PROVISION STATUSES IN THIS SECTION\n\nNot all sentences in this book have the same normative force. The following statuses are used:\n\n### NORMATIVE — NORMATIVE\n\nA provision considered mandatory for conformity and methodological integrity. Example: A single response cannot be used as the general GEO score.\n\n### CANDIDATE — CANDIDATE PROVISION\n\nA recommended method to be finalised after pilot, external application, or statistical validation. Example: 1,000 valid observations per AI product, per main measurement wave, are recommended as the assumed minimum sample.\n\n### HYPOTHESIS\n\nIt is an expectation that needs to be tested with data. Example: There may be a difference in material representation between a controlled clean panel and a natural user panel.\n\n### SYNTHETIC\n\nIt is a method representation that does not represent a real user, a real AI output, or real company performance.\n\n### OPEN QUESTION\n\nIt is a topic that has not yet been concluded due to insufficient evidence or governance decision. These statuses cannot silently transform into one another. A candidate threshold does not become a proven normative fact simply because it is written in the book. A synthetic representation does not turn into a real-world observation. A hypothesis cannot be published as a result without being measured.\n\n## NOMOS Challenge\n\nYou asked an artificial intelligence system a question about your brand. The system gave a correct answer. You took a screenshot. You included it in your presentation. Then you said the following sentence: “Artificial intelligence knows us correctly.” What exactly did you measure? The answer that all users would see? The typical behaviour of the system?\n\nA global representative? Accuracy in all languages? Stability over time? Or only:\n\n> The single output that a specific user received, on a specific account, on a specific AI product and user surface, in response to a specific prompt, at a specific time?\n\nA single answer is not worthless. A single answer can be true. A single answer can be important. A single answer can reveal a serious mistake. A single answer can show the real experience of a particular person. However, a single answer cannot be proof of a judgement greater than it can bear.\n\nA single response can show the following:\n\n> This output has been produced at least once under recorded conditions.\n\nA single response alone cannot show the following:\n\n- That most users saw the same answer\n\n- That the same result occurred in other countries\n\n- That the same accuracy is maintained in other languages\n\n- That the same answer was produced in other accounts and plans\n\n- That the answer is stable over time\n\n- That other AI products produced the same representation\n\n- That the GEO intervention caused this result\n\n- That the answer affects user behaviour\n\n- That the answer generates lead, sales, or sustainable value\n\n- That the overall GEO score of the entity is high\n\nThe first premise of this section is:\n\n> A response is an observation. / It is not a distribution of observations. / A distribution is not causality. / Causality alone is not sustainable value.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThis section aims to eliminate the smallest but most common inference error in GEO measurement:\n\n> Counting a single output of a generative system as general system behaviour, population experience, brand visibility, or GEO score.\n\nThis section normalises the following distinctions:\n\n- Single output versus population outcome\n\n- Existence of an event versus prevalence of an event\n\n- Observation versus sample\n\n- User versus session\n\n- Repetition versus independent observation\n\n- AI model versus AI product presented to the user\n\n- Raw response versus unit of analysis\n\n- Immediate result versus temporal stability\n\n- Screenshot versus full evidence package\n\n- System rejection versus invalid data collection\n\n- Critical singular event and prevalence rate\n\n- Synthetic record and real user observation\n\n- Measured rate and conformity decision\n\nBefore these distinctions are established:\n\n- GEO-1000 sample,\n\n- country and language weights,\n\n- atomic claim scoring,\n\n- NOMOS score,\n\n- conformity decision\n\ncannot be established.\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nThe GEO score of an entity in generative systems cannot be generated from a single response, a single screenshot, a single user, a single account, a single session, a single language, a single country, or a single point in time measurement. A population-representative claim requires at least the following:\n\n- Predefined audit object\n\n- Predefined target population\n\n- Specified sampling design\n\n- Locked prompt or prompt family\n\n- Defined AI product and user surface\n\n- Multiple valid observations\n\n- Known dependency structure\n\n- Country, language, and time records\n\n- Unaltered raw responses\n\n- Evidence packages\n\n- Predefined evaluation rules\n\n- Uncertainty statement\n\n- Accountable human responsibility\n\nThe presence of these elements does not automatically mean that the measurement is correct. However, their absence may make the generalisation to a large population untenable.\n\n## 3. FIVE SEPARATE UNITS OF MEASUREMENT\n\nA common fundamental error in GEO auditing is confusing the different units of measurement. The following five units should be clearly distinguished.\n\n### 3.1. Observation Unit\n\nThe observation unit is the single generative system output obtained under specific conditions. An observation can be defined by the following combination:\n\nO_i = (E_i, A_i, U_i, C_i, L_i, S_i, P_i, T_i, R_i, K_i)\n\nHere:\n\n- Ei: the audited entity\n\n- Ai: AI product and user surface\n\n- UI: anonymous contributor or user unit\n\n- Ci: country or geographical stratum\n\n- Li: language and locale\n\n- Si: session and personalisation condition\n\n- Pi: prompt ID and version\n\n- Ti: time and measurement wave\n\n- Ri: raw response\n\n- Ki: proof and verification record\n\nIf one of these elements changes materially, a new observation may occur. For example, the same user:\n\n- if it asks the same prompt in another AI product,\n\n- If he asks in another language,\n\n- If he asks in another measurement wave,\n\n- If he asks separately in the personalised and clean session,\n\nSeparate observations can be produced. These observations are not automatically considered independent from each other because they come from the same user.\n\n### 3.2. Sampling Unit\n\nThe sampling unit is the person, account, session, or other access unit selected from the target population. In most GEO-1000 applications, the sampling unit can be an appropriate real user. However, according to the research design:\n\n- user,\n\n- user–product pair,\n\n- account,\n\n- session,\n\n- research panel member\n\nIt can be used as a unit of election. The failure to clearly define the sampling unit can lead to the following types of miscounts:\n\n- Counting one person's ten repetitions as ten people\n\n- Counting a joint account as multiple independent users\n\n- To evaluate the results of the same user's ten AI products as if they were ten independent population observations\n\n- Ignoring the clustering in a single research panel\n\n### 3.3. Analysis Unit\n\nThe unit of analysis is the entity to which accuracy or representativeness assessment is applied. This unit:\n\n- exact answer,\n\n- separate paragraph within the answer,\n\n- material proposition,\n\n- atomic claim,\n\n- citation,\n\n- recommendation issue\n\nIt is possible. A user sees a single answer. However, that answer can simultaneously contain:\n\n- five correct claims,\n\n- two unsupported claims,\n\n- one outdated claim,\n\n- one critical error\n\nTherefore, the answer seen by the participant does not have to be a single, indivisible unit of truth/falsehood in the analysis. The transition status of the full answer and the truth statuses of the atomic claims within it can be maintained separately. The atomic claim method will be defined in a later chapter of the book.\n\n### 3.4. Universe of Inference\n\nThe inference universe is the target population to which the measurement result is intended to be generalised. This universe cannot be silently accepted as \"everyone in the world.\" For a specific AI product, the target universe can be limited by conditions such as:\n\n- Having official and legitimate access to the product\n\n- Meeting the appropriate age requirement\n\n- Being able to use the product in the relevant language\n\n- Being able to access the specified web or mobile interface\n\n- Being able to be active during the specified measurement period\n\nEven for the same AI product:\n\n- controlled clean panel,\n\n- natural user panel,\n\n- free plan,\n\n- paid plan,\n\n- web,\n\n- mobile\n\ncan create different inference universes. The published rate is incomplete if it is not specified to which universe a sample generalises.\n\n### 3.5. Estimated Quantity — Estimand\n\nA study does not just collect data. It tries to estimate a specific quantity. The primary estimand of GEO-1000 can be expressed in its most general form as follows:\n\n> The probability that an appropriate user from the defined target population, under the specified measurement wave, on a particular AI product and user surface, sees a representation that meets the acceptance criterion when using a locked prompt under defined session conditions.\n\nMathematically:\n\nθ_{E,A,P,S,W,T} = Pr(Y_i = 1 | U_i ∈ T, E, A, P, S, W)\n\nHere:\n\n- E: audited entity\n\n- A: AI product and user surface\n\n- P: prompt version\n\n- S: session condition\n\n- W: measurement wave\n\n- T: target population\n\nYi=1: the response meets the transition criterion that will be defined later\n\nThis definition sets an important boundary:\n\n> The GEO score is not a model's abstract and timeless 'knowledge'; it is a prediction of the likelihood of observed representations under defined user conditions.\n\nThe final transition rule and score formula will be arranged in the later sections of the book.\n\n## 4. THE LIMITATION OF THE SENTENCE “THIS IS HOW THE MODEL RESPONDED”\n\nUsers often do not access a bare base model directly. The result visible to the user may consist of a combination of the following components:\n\n- Base or guiding model\n\n- Product interface\n\n- Retrieval or web access\n\n- Citation layer\n\n- Security and policy layer\n\n- System instructions\n\n- Account plan\n\n- Personalisation\n\n- Memory\n\n- Previous conversation\n\n- Language and locale\n\n- Country\n\n- Time\n\n- Provider's product updates\n\nFor this reason, the preferred expression in the main inspection conducted through the user surface is as follows:\n\n#### \"The specified AI product and the following output were observed on the user surface.\"\n\nThe following expression may be broader without the necessary technical control: \"The model thinks this way.\" The model name appearing within a product can be recorded. However, the result should be evaluated as the behaviour of the entire user surface, not just the model weights.\n\n## 5. WHAT IS A VALID OBSERVATION?\n\nThe presence of a response does not automatically mean it is a valid observation that can be entered into the GEO measurement.\n\n### 5.1. Minimum Validity Conditions\n\nAn observation must contain at least the following elements according to the relevant protocol:\n\n- Presence of correct audit\n\n- Assigned AI product and user surface\n\n- Correct prompt ID and version\n\n- Correct language and locale\n\n- Defined session condition\n\n- Recorded country or geographic layer\n\n- UTC timestamp and measurement wave\n\n- Complete raw response\n\n- First unselected suitable output\n\n- Required screenshot or equivalent evidence\n\n- Anonymous observation ID\n\n- Protocol deviation record\n\nTheir detailed technical schema will be defined in the later NOMOS Capture section.\n\n### 5.2. Observation Statuses\n\nEach record must be marked with one of the following statuses:\n\n### VALID — VALID\n\nMeets the required conditions of the protocol.\n\n### CONDITIONALLY VALID — CONDITIONALLY VALID\n\nThere is a limited deficiency; which analyses it can be used for is clearly specified.\n\n### INVALID — INVALID\n\nCannot be used in the main estimation or score calculation due to protocol violation.\n\n### UNKNOWN — UNKNOWN\n\nThere is not enough record for a validity decision. An UNKNOWN observation cannot be silently accepted as VALID.\n\n### 5.3. The Difference Between a Valid System Result and Invalid Data Collection\n\nThis distinction is mandatory. An AI product:\n\n- may refuse to respond,\n\n- may display an error message,\n\n- may stop without generating a response,\n\n- may experience a connection issue,\n\nmay say it cannot complete the task. If the participant applied the entire protocol correctly, these can be valid system outcomes. They cannot be quietly removed from the dataset. In contrast:\n\n- the participant truncating the response,\n\n- changing the prompt,\n\n- regenerating to select the best answer,\n\n- editing the screenshot,\n\n- using the wrong AI product\n\ndata collection can create invalidity. The first is the actual behaviour of the system. The second is the disruption of the measurement process. The two cannot be placed in the same category.\n\n## 6. WHAT CAN A SINGLE RESPONSE PROVE?\n\nA single observation is limited. But it is not meaningless.\n\n### 6.1. That a Specific Output Occurred at Least Once\n\nA valid record can support this sentence: “On the specified AI product and user surface, under the specified time and conditions, this output was observed at least once.” This sentence shows the existence of the event. It does not show its prevalence.\n\n### 6.2. A Specific User’s Actual Exposure\n\nA user is mistaken for a:\n\n- company identity,\n\n- health information,\n\n- legal guidance,\n\n- price,\n\n- licence,\n\n- advice\n\nif seen, this is a real user experience. Even if it does not represent the entire population, it is material from the perspective of the affected user.\n\n### 6.3. The Possibility of a Specific Error\n\nA single output may show: “This error may occur under these system conditions.” However, it does not show: “This error is typical.”\n\n### 6.4. Generating a New Research Hypothesis\n\nA single error can create a question that will be tested on a broader scale. Example: “Did this identity mix-up occur in a single session only, or is there a systematic pattern in the same language and country?”\n\n### 6.5. Initiating a Critical Incident Review\n\nSingle response:\n\n- predictable serious human harm,\n\n- fake professional authority,\n\n- incorrect emergency medical guidance,\n\n- material legal error,\n\n- serious safety advice,\n\n- serious identity confusion with another entity\n\nmay be sufficient to initiate an incident review. Initiating a critical incident investigation is not the same process as announcing a population ratio.\n\n### 6.6. Refuting the \"This Never Happened\" Claim\n\nSuppose an organisation says, ‘The system has never assigned us to the wrong country.’ A single verified counter-observation can refute that absolute claim. A single observation does not support ‘It happens all the time’, but it can refute ‘It has never happened.’\n\n## 7. WHAT CAN A SINGLE RESPONSE NOT PROVE?\n\n### 7.1. Typical System Behaviour\n\nA single response is not sufficient to conclude: 'The system usually responds this way.'\n\n### 7.2. Population Prevalence\n\nA single response does not support the result: '95% of users see the same representation.'\n\n### 7.3. Temporal Stability\n\nToday's response:\n\n- tomorrow,\n\n- a week later,\n\n- After the product update\n\nit may not remain the same. It is not a one-wave stability measurement.\n\n### 7.4. Cross-Country Validity\n\nA response from a user in Turkey does not indicate what users in China, Germany, or San Marino would see.\n\n### 7.5. Cross-Language Validity\n\nAn English output:\n\n- Turkish,\n\n- Japanese,\n\n- Arabic,\n\n- German\n\ndo not represent the accuracy of outputs.\n\n### 7.6. Cross-AI Product Validity\n\nThe result in one AI product cannot be generalised to other AI products. Different products and interfaces of the same provider may also behave differently.\n\n### 7.7. Intervention Effect\n\nA single positive output after a GEO change does not prove the statement: “This result was created by our intervention.” Baseline, comparison, and alternative explanations are required.\n\n### 7.8. Recommendation Stability\n\nThe presence of a user viewing an option is not sufficient to conclude: “The system now recommends this brand.”\n\n### 7.9. Commercial Impact\n\nA positive response:\n\n- indicates that the user saw the response,\n\n- researched the brand,\n\n- that you communicated,\n\n- that you purchased,\n\n- that you were satisfied\n\ndoes not show on its own.\n\n### 7.10. GEO Score\n\nA single response does not estimate the representative distribution of the defined target population. Therefore, it cannot produce an overall GEO score.\n\n## 8. THE EXISTENCE OF THE EVENT AND THE PREVALENCE OF THE EVENT\n\nThis is one of the most important distinctions in this section.\n\n### 8.1. Existence of the Event\n\nAnswers the question: \"Did this representation or error occur at least once?\" A single valid observation may be sufficient. Representation:\n\nI(E) = 1 (if the event was observed at least once); otherwise I(E) = 0\n\nThe 0 here does not mean: \"The event is impossible.\" It only indicates that it was not observed in the current measurement.\n\n### 8.2. Prevalence of the Event\n\nIt answers the question: \"How often does the event occur in the defined target population?\" In a simple sample, the raw rate can be calculated as:\n\np̂ = number of valid records where the event was observed / total number of valid records\n\nHowever, in real GEO-1000 studies:\n\n- population weights,\n\n- country and language strata,\n\n- matched users,\n\n- panel clusters,\n\n- nonresponse,\n\n- invalid records,\n\n- design effect\n\nshould be taken into account. Therefore, the raw rate is not always the final population estimate.\n\n### 8.3. Incidence and Prevalence Cannot Be Converted Into Each Other\n\nAn event may have been observed once but be very rare. An event may be common but not observed at all in a small pilot. The following two sentences are different:\n\n- \"This error was observed once.\"\n\n- \"This error was observed in 12% of users.\"\n\nThe second sentence requires a defined denominator and sample.\n\n## 9. THE STATISTICAL APPEARANCE OF THE SINGLE OBSERVATION BIAS\n\nWe asked a question to an AI product once. The answer met the success criterion. Raw rate:\n\np̂ = 1/1 = 1 = 100%\n\nIt appears to be 100 per cent. But that 100 per cent does not mean that every member of the target population will see the same result. For illustration only, the table below shows approximate 95 per cent Wilson confidence intervals under independent Bernoulli observations and unweighted simple random sampling: [K09]\n\nThis table is synthetic and illustrative. It does not prove:\n\n- That these intervals are valid for all sampling designs of GEO-1000\n\n- That 1,000 observations are sufficient for all language and country subgroups\n\n- clustering and weighting are ineffective\n\n- that 95 per cent raw accuracy automatically means appropriateness\n\nWhat it actually shows is:\n\n> As the number of observations decreases, the uncertainty about the population proportion increases.\n\n100 per cent seen in one observation can be compatible with a very wide possible population range.\n\n## 10. WHY “1,000” IS NOT A MAGIC NUMBER?\n\n### CANDIDATE BASE RULE — CANDIDATE FOUNDATIONAL RULE\n\nGEO-1000 carries this founding suggestion:\n\n> For each monitored AI product and each main measurement wave, a minimum of 1,000 valid user observations is targeted by default.\n\nThis provision is not yet the final scientific minimum. It will be re-evaluated based on pilot and external application results.\n\n### 10.1. Why Is It a Strong Start?\n\nUnder the simple random sampling assumption, when the true rate is around 50%, the theoretical 95% margin of error for 1,000 observations is approximately ±3.1 points. This can provide a meaningful starting precision at the overall product level.\n\n### 10.2. Why Doesn't It Solve Every Problem?\n\n1,000 observations:\n\n- to dozens of countries,\n\n- to many languages,\n\n- to free and paid plans,\n\n- to web and mobile surfaces,\n\n- to controlled and natural panels\n\nWhen divided, the subcells shrink. For example, if there are only 50 observations for each of the 20 languages, language-based uncertainty can be much higher than the overall model score.\n\n### 10.3. Design Effect\n\nObservations:\n\n- from the same research panel,\n\n- from the same country,\n\n- from the same user,\n\n- from a matched design\n\nare present, the simple independent sample assumption may not hold. The actual uncertainty can be wider.\n\n### 10.4. The Approach of NOMOS\n\n1,000 observations serve as:\n\n- the founding reference point of the protocol,\n\n- default starting minimum at the model level,\n\n- candidate threshold to be calibrated with the pilot\n\nNOMOS does not claim that 1,000 observations are automatically sufficient for every country, language and condition of use.\n\n## 11. INDEPENDENT OBSERVATION IS NOT THE SAME AS REPEATED OBSERVATION\n\nA user can ask the same question ten times. This process can be valuable. However, it does not produce ten independent users.\n\n### 11.1. Same User–Same Session Repeat\n\nCan measure:\n\n- answer renewal variance,\n\n- within-session stability,\n\nprobability of alternative answers. It does not measure population prevalence on its own.\n\n### 11.2. Same User–New Session Repeat\n\nIt may reduce the previous chat context. However:\n\n- same account,\n\n- same plan,\n\n- same country,\n\n- same personalisation,\n\n- same user\n\nmay continue to create statistical dependence between observations.\n\n### 11.3. A Single User Testing Multiple AI Products\n\nThis design can help with cross-model matched comparison. However, observations are dependent at the user level. This dependence should be preserved in recording and analysis.\n\n### 11.4. Different Users Using the Same Account\n\nDifferent people may exist. However:\n\n- account history,\n\n- personalisation,\n\n- plan,\n\n- memory\n\ncan be shared. These records cannot be considered independent automatically.\n\n### 11.5. Normative Independence Provision\n\n> Observations should be designed to be as independent as possible; known user, account, session, panel, country, device, and repeated dependencies should be recorded and taken into account in the analysis.\n\nIndependence should not be reduced to a binary:\n\n- yes,\n\n- no\n\nWhere necessary, use one of the following dependency statuses:\n\n### INDEPENDENT\n\n### PAIRED\n\n### CLUSTERED\n\n### REPEATED_WITHIN_USER\n\n### DEPENDENT\n\n### UNKNOWN\n\n## 12. FIRST SUITABLE OUTPUT PRINCIPLE\n\nIn the main population panel, the participant's task is not to find the most positive answer. The participant must record the first completed suitable output generated under the predefined condition. The participant:\n\n- cannot refresh the answer,\n\n- cannot request a new answer,\n\n- cannot choose the positive one,\n\n- cannot only trim the good part,\n\ncannot rewrite the answer. Reproduction behaviour can also be investigated separately. However: \"We asked the same question five times and selected the best answer.\" is not the main population score. This is the selected best output test. It should be reported with a separate name and method.\n\n## 13. SINGLE CRITICAL EVENT EXCEPTION\n\nA single answer cannot determine the population score. However, this does not mean that a single answer can never have a significant impact on a decision. NOMOS establishes two separate assessment pathways.\n\n### 13.1. Population Representation Pathway\n\nIt asks the question: “How frequently does this outcome occur in the target population?” Requires a large and valid sample.\n\n### 13.2. Critical Event Pathway\n\nIt asks the question: “How large is the foreseeable harm when this outcome occurs once?” A single confirmed outcome can trigger a review. For example:\n\n- Assigning a non-existent health licence\n\n- Making a dangerous emergency health referral\n\n- Reporting incorrect legal authority\n\n- Removing the security boundary\n\n- Confusing a company with another legal entity\n\n- Repeating a manipulative instruction hidden from people as if it were advice\n\nIt can be a singular Critical event candidate.\n\n### 13.3. Separation of Score and Conformity Decision\n\nA Critical event can generally be a single observation. For example:\n\n1/1000=0.1%\n\nThis rate may seem small. However, if the potential harm of the event is severe, it may create an irreparable gap in the conformity decision. Therefore:\n\n> Population score measures the frequency of the event; Critical event review measures the impact of the event on harm and integrity.\n\nHigh average accuracy cannot automatically eliminate heavy singular error. The final conditions for critical classification will be defined in later chapters of the book.\n\n## 14. OBSERVATIONAL CLAIM LADDER\n\nIn this section, only the first four observation levels are normatively separated. At higher levels:\n\n- population-weighted score,\n\n- multi-model score,\n\n- time stability,\n\n- intervention effect,\n\n- sustainable value\n\nwill be defined in subsequent sections.\n\n### O-0 — NOT TESTED\n\nThe relevant AI product or condition has not been measured. Correct status:\n\n### NOT TESTED\n\nIndefensible sentence: \"It is appropriate because the problem was not observed.\"\n\n### O-1 — SINGLE EVENT OBSERVATION\n\nThere is one valid outcome. Defensible sentence: \"Under the specified conditions, this outcome was observed once.\" Indefensible sentence: \"The system represents existence this way.\"\n\n### O-2 — USER OR SESSION REPEAT\n\nMultiple outcomes exist for the same user, account, or session family. Defensible sentence: \"Under the specified user and session conditions, this outcome occurred in seven out of ten repetitions.\" Indefensible sentence: \"Seven out of ten independent users saw this outcome.\"\n\n### O-3 — MULTI-USER SINGLE CELL MEASUREMENT\n\nThe same:\n\n- AI product,\n\n- user surface,\n\n- country,\n\n- language,\n\n- panel condition,\n\n- prompt version,\n\n- measurement wave\n\ncontains multiple suitable user observations. Defensible sentence: “78% of the 200 valid observations in the Turkish/Turkey controlled cell met the specified transition criterion.” This result cannot be automatically generalised to other countries, languages, or user surfaces.\n\n### O-4 — MULTI-CELL LAYERED MEASUREMENT\n\nPredefined samples exist from different countries, languages, or user cells. At this level:\n\n- cell results,\n\n- weighted result,\n\n- scope,\n\n- uncertainty\n\nreportable. Detailed rules for population weights and subgroup fairness are defined in later chapters of the book.\n\n## 15. SYNTHETIC APPLE.COM DEMONSTRATION\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The users, AI products, responses, rates, and results below are fictional. They do not represent the actual performance of Apple Inc. or the behaviour of any real AI product.\n\n### 15.1. Scenario A — Single Positive Response\n\nA synthetic user submits the following question to synthetic AI Product A: ‘What kind of company is Apple.com?’ Synthetic response: ‘Apple is a US-based technology company that develops consumer electronics, software and digital services.’ Preliminary evaluation:\n\n- Entity identity is correct\n\n- Main category is correct\n\n- No obvious critical errors\n\n- Basic representation is acceptable\n\nRaw result:\n\n1/1=100%\n\nIncorrect public statement: “AI Product A represents Apple with 100% accuracy.” Correct public statement: “In the synthetic sample, the specified single observation met the transition criterion. Population prevalence and product reliability have not been determined.”\n\n### 15.2. Scenario B — Discovery with Ten Observations\n\nTen independent synthetic users use the same version of the semantic prompt. Synthetic result:\n\n- 7 responses pass\n\n- 1 response is partially correct\n\n- 1 response contains outdated information\n\n- 1 response assigns the wrong main activity\n\nraw pass rate:\n\n7/10=70%\n\nUnder a simple, idealised assumption of independent sampling, the approximate 95 per cent Wilson interval is necessarily wide. The result remains exploratory because it is based on:\n\n- small sample,\n\n- single cell,\n\n- single wave\n\nCorrect status:\n\n> EXPLORATORY — Not sufficient for the publicly available global NOMOS score.\n\n### 15.3. Scenario C — 1,000 Observed Synthetic Wave\n\nAssume synthetically that the following conditions are met:\n\n- AI Product A\n\n- 1,000 valid observations\n\n- Predefined target population\n\n- Locked prompt version\n\n- Defined country and language layers\n\n- Unmodified initial suitable response\n\n- Full evidence records\n\n- Predefined adjudication rules\n\nSynthetic result:\n\nSimple raw pass rate:\n\n918/1000=91.8%\n\nApproximate 95% Wilson interval under the idealised unweighted assumption: is. However, this number alone is not yet the final NOMOS GEO score. Because there are additional factors that need to be evaluated:\n\n- Country and language weights\n\n- Design effect\n\n- Lowest language result\n\n- Invalid observation rate\n\n- Inter-rater agreement\n\n- Verification of critical error candidates\n\n- Untested user groups\n\n- Time and user surface limitation of the measurement\n\nSix Critical-error candidates must also undergo separate incident review. Correct reporting: ‘The raw response pass rate in the synthetic wave is 91.8 per cent. This result describes the observations before population weights and design effects are applied. Because six Critical-error candidates were identified, the conformity decision requires separate review.’ Incorrect reporting: ‘Apple's NOMOS GEO score is 91.8 and it is compliant.’ Synthetic data cannot be attributed in this way to a real company or a real system.\n\n## 16. MANDATORY NORMATIVE PROVISIONS\n\n### N-01\n\nA single generative system output cannot be used as the general GEO score.\n\n### N-02\n\nA singular output observed alone:\n\n- AI product and user surface,\n\n- country,\n\n- language,\n\n- prompt,\n\n- session,\n\n- time\n\nmust be reported along with its conditions.\n\n### N-03\n\nSingle output:\n\n- typical,\n\n- continuous,\n\n- global,\n\n- belonging to all users\n\ncannot be presented as behaviour.\n\n### N-04\n\nRepetitions from the same user, account, or session cannot be counted as independent population observations without disclosing the dependency.\n\n### N-05\n\nIn the main population panel, the participant must record the first suitable output predefined; reproduction cannot be made to choose a positive response.\n\n### N-06\n\nSystem rejection, error message, or inability to respond should be recorded as a valid system outcome if the protocol has been correctly applied.\n\n### N-07\n\nTrimmed, modified, manually rewritten, or responses whose prompt context cannot be verified cannot be included in the main score calculation.\n\n### N-08\n\nSelecting only positive screenshots and withholding the distribution of results is prohibited.\n\n### N-09\n\nSynthetic observations cannot be used in real user or real system scores.\n\n### N-10\n\nA population proportion should not be published without coverage, denominator, and appropriate uncertainty explanation.\n\n### N-11\n\nThe probability of a critical error cannot be quietly compensated within the overall average.\n\n### N-12\n\nThe observation result can only be applied to the inference universe that is uniquely defined; it cannot be generalised to unmeasured countries, languages, products, or time periods.\n\n### N-13\n\nMeasurement, scoring, or public statement must have an accountable human or institutional owner.\n\n### N-14\n\nUNKNOWN, NOT TESTED, or INVALID records cannot be quietly converted into a positive result.\n\n## 17. CANDIDATE GEO-1000 MAIN RULE\n\n### CANDIDATE-001\n\n> For every AI product being audited and for every main population measurement wave, a minimum of 1,000 valid user observations should be targeted by default.\n\nThis candidate sentence:\n\n- accuracy of the general product forecast,\n\n- cost,\n\n- applicability,\n\n- country and language fairness\n\nThis creates an initial balance among these considerations. It does not, however, guarantee adequate coverage of:\n\n- all the country scores,\n\n- all language scores,\n\n- all user subgroups\n\nAdequacy must instead be tested through:\n\n- pilots,\n\n- independent repetitions,\n\n- design effect,\n\n- subgroup sensitivity\n\nThose evaluations determine the final normative status.\n\n## 18. FORMS OF FAILURE\n\n#### F-01 — Selected Screenshot\n\nOnly the most positive response is published among numerous results.\n\n#### F-02 — Treating a Single User as the Population\n\nThe experience of one person is generalised to the entire target universe.\n\n#### F-03 — Counting Repeats of the Same User as New People\n\nRepeats generated by the same person or account are reported as independent participants.\n\n#### F-04 — Renewing the Response and Picking the Best One\n\nInitial or negative responses are discarded; the most appropriate response is saved as the measurement result.\n\n#### F-05 — Prompt Drift\n\nParticipants change the word, tone, or scope of the prompt; results are combined as if from the same experiment.\n\n#### F-06 — Context Contamination\n\nPrevious conversations affect the response; the record is presented as a clean session.\n\n#### F-07 — Hiding Personalisation\n\nMemory, special instructions, or user history is visible; the result is reported as a general user response.\n\n#### F-08 — Confusing Model and Product\n\nOnly the general model or provider name is written; the actual product, interface, and tool condition are not recorded.\n\n#### F-09 — Removing Errors and Mistakes from the Denominator\n\nDeleted to make the results appear positive when the system does not respond to records.\n\n#### F-10 — Counting Invalid Capture as System Failure\n\nReported as participant error, missing evidence, or incorrect outcome of the AI product.\n\n#### F-11 — Globalising a Single Language and Country\n\nThe result of a single locale is applied globally.\n\n#### F-12 — Using Synthetic Data as Real\n\nRecords created for demonstration purposes are added to actual performance.\n\n#### F-13 — Losing a Critical Event on Average\n\nSevere health, legal, licensing, or security errors are made invisible within a high average.\n\n#### F-14 — Drawing Intervention Effect from a Single Success\n\nThe single positive response received after the GEO intervention is considered the causal effect of the study.\n\n#### F-15 — Judging the Entire System from a Single Mistake\n\nOne error is generalised as if the AI product is always wrong and wrong for all users.\n\n#### F-16 — Defining the Target Population Afterwards\n\nAfter the outcome is seen, the user group with high success is declared the “actual target population.”\n\n## 19. AUDIT PROCEDURE\n\nAn auditor should follow the sequence below when evaluating the validity of a single observation or an early GEO claim.\n\n### Step 1 — Record the Public Claim\n\nThe full proposed sentence is determined. Example: “ChatGPT correctly knows our company.”\n\n### Step 2 — Determine the Level of the Claim\n\nThe sentence may make a claim about:\n\n- single event,\n\n- prevalence,\n\n- stability,\n\n- global consequence,\n\n- causality,\n\n- commercial impact\n\nThe applicable level must be identified.\n\n### Step 3 — Determine the Supporting Observation Count\n\nHow many real users? How many accounts? How many independent sessions? How many repeats? How many AI products? How many languages and countries?\n\n### Step 4 — Verify Observation Conditions\n\nAre the AI product, user surface, prompt, session, country, language and time all recorded?\n\n### Step 5 — Examine Evidence Integrity\n\nIs there a full response? Is there a full screenshot? Has the response been selected or refreshed? Has the file been modified? Is the record synthetic?\n\n### Step 6 — Determine Dependency Structure\n\nAre there repetitions of the same user? Has a shared account been used? Is there a mirrored model design? Is there clustering from the same panel?\n\n### Step 7 — Separate Data Collection Error from Valid System Result\n\nIs the rejection or system error the actual result? Or did the participant violate the protocol?\n\n### Step 8 — Compare the Claim with the Level of Evidence\n\nIs a single observation used for a population or global inference?\n\n### Step 9 — Open the Critical Path\n\nIf a single response carries a foreseeable severe harm, a separate incident investigation is initiated.\n\n### Step 10 — Limit the Decision\n\nThe public sentence is reduced only to the level carried by the evidence.\n\n## 20. NECESSARY EVIDENCE\n\nFor an observation or claim within the scope of this section, the following records should be sought to the extent that they are relevant:\n\n- Audited entity identity\n\n- Fully disclosed GEO claim to the public\n\n- Number of observations\n\n- Number of unique participants\n\n- AI product and user surface\n\n- Visible model or version, if available\n\n- Plan or account type\n\n- Country and locale\n\n- Prompt ID, version, and full text\n\n- Session and personalisation status\n\n- UTC timestamp\n\n- Measurement wave\n\n- Full raw response\n\n- Screenshot or equivalent evidence\n\n- Reproduction or answer selection status\n\n- Observation validity status\n\n- Dependency and matching record\n\n- Synthetic or real status\n\n- Invalid observation log\n\n- Critical event record\n\n- Result scope\n\n- Uncertainty statement\n\n- Responsible person or institution\n\nNot all of these records may be required in the same detail for every low-risk discovery observation. However, if a broad GEO score or conformity clause is to be given to the public, material fields cannot be left blank.\n\n## 21. AUDIT CHECKLIST\n\nHas the audited entity been fully identified? Have the AI product and user surface been recorded? Was the prompt locked before measurement? Was the prompt changed by the participant? Are the session and personalisation conditions known? Is the user's country and locale information recorded? Is the full response preserved? Is it the first appropriate output or the best chosen output? Is there a complete record of evidence? Have the actual number of users and the number of observations been separated? Have repeated and matched observations been counted independently? Have refusals and system errors been properly managed in the denominator? Have invalid records been retained and justified? Have synthetic and real data been separated? Is a single response presented as a population result?\n\nIs the claim applied only to the measured product, country, language, and time? Has the probability of a critical error been assessed separately? Does the result carry a statement of confidence or uncertainty? Are untested areas clearly shown? Does the public statement exceed the level of actual observation? If one of the factual questions is unanswered, the result carries limited confidence. If more than one is unanswered, it should not be used for the overall GEO score.\n\n## 22. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “The real user already sees only one answer.”\n\nThat is correct. The single answer seen by the single user is that person’s real experience. Therefore, individual events are recorded. However, individual experience is not the same as population experience. NOMOS maintains both facts together:\n\n> The experience of a single user is real. / A single user is not the entire population.\n\n### Objection 2 — “If the system gives the same answer every time, isn't one answer enough?”\n\nThe claim that the system gives the same answer every time also requires measurement. Additionally:\n\n- time,\n\n- retrieval,\n\n- country,\n\n- language,\n\n- plan,\n\n- personalisation,\n\n- user surface\n\nIt may vary. Even a system assumed to be deterministic cannot be generalised beyond defined conditions.\n\n### Objection 3 — \"Why exactly 1,000?\"\n\nThe number 1,000 is not a magic threshold of certainty. GEO-1000 serves as a founding reference point that balances:\n\n- a substantial user volume,\n\n- a broad product-level estimate,\n\n- practicable cost,\n\n- public comprehensibility.\n\nSubgroups may require more observations. Pilot studies must verify or revise the threshold.\n\n### Objection 4 — “Let the same 1,000 people test all AI products.”\n\nThis method can be useful for paired comparison across models. However, selecting only intensive AI users who can access all products may bias the target population. Also, observations from the same person are dependent. Fully paired, partially paired, and balanced designs will be compared later.\n\n### Objection 5 — “Isn’t a screenshot already sufficient proof?”\n\nScreenshots are important. However, on their own they do not always show the following:\n\n- real time\n\n- country\n\n- plan\n\n- previous conversation\n\n- personalisation\n\n- answer refresh\n\n- file change\n\n- full response\n\nFor this reason, a screenshot is an important part of the evidence package. It is not the whole.\n\n### Objection 6 — “Why do we include a non-responding system in the score?”\n\nThe user's real experience may be not getting an answer. A rejection or error:\n\n- technical,\n\n- security,\n\n- policy,\n\n- access\n\nmay be the result. If the protocol was correctly applied, the event should be recorded. The system’s unresponsiveness due to the participant’s technical error is classified separately.\n\n### Objection 7 — “Why should a single severe mistake affect the entire inspection?”\n\nA single mistake does not prove that most of the population made the same mistake. However, the potential damage of a single mistake can be severe. Therefore:\n\n- score frequency,\n\n- Critical decision harm and integrity\n\nare evaluated separately. The average cannot automatically erase human harm.\n\n### Objection 8 — “If we can’t say anything with a single response, what is the value of early tests?”\n\nSingular and small-sample tests:\n\n- generate hypotheses,\n\n- discover types of errors,\n\n- test prompts,\n\n- test the capture system,\n\n- calibrate adjudicators,\n\ncan reveal critical events. What is wrong is not to do exploratory testing. What is wrong is presenting the exploratory result as a population score.\n\n## COMMON PRINCIPLE OF CHAPTER 24\n\nThis chapter was written to eliminate a single misconception: \"I asked an AI, got an answer; here is my GEO result.\" No. You do not yet have a general GEO result. What you have is an observation. This observation:\n\n- is real,\n\n- is significant,\n\n- even critical\n\nmay be. However, it is still a single observation. For a GEO score, the existence of:\n\n- users,\n\n- AI products,\n\n- countries,\n\n- languages,\n\n- sessions,\n\n- times\n\nWhat must be measured is the distribution of representation across AI products, countries, languages, sessions and time. A single screenshot is not a distribution. Ten automated repetitions are not ten independent people. Nor do a thousand responses from a poor sample represent the world's population. Even a population-proportional sample may be incomplete without language fairness. A high average cannot erase Critical errors. NOMOS's first law of measurement is therefore:\n\n> A single answer is an event. / Multiple answers are a pattern. / A defined sample is an estimate of the population. / Repetition over time is a measure of stability. / But none of these alone are causality or sustainable value.\n\n## Order of Section 1 of NOMOS\n\n> Do not bring me one screenshot and then speak for the entire system.\n\n> Do not make the answer of one user the answer of the world's population.\n\n> Do not convert duplicates of the same account into independent humans.\n\n> Do not refresh the answer and display the preferred as if it were the real distribution.\n\n> Do not remove rejections, errors, and unknowns from the denominator.\n\n> Do not generalise one language to the world, one country to all countries, or one AI product to all generative systems.\n\n> Do not turn a single success into evidence of an intervention effect, or a single error into the immutable character of the entire system.\n\n> Do not drown a single severe error within the average.\n\n> First, record the event. / Then measure the pattern. / Then define the target population. / Then show dependency and uncertainty. / And speak only to the extent that the evidence carries.\n\n## The Chapter's Closing Sentence\n\n> The representation of an entity in artificial intelligence systems is not a single answer. AI products are a distribution that should be measured across users, countries, languages, sessions, and times.\n\n## Normative Core\n\n> A single generative-system output MAY establish that a specific response occurred under recorded conditions. A single output MUST NOT be used to infer: - population prevalence, - typical product behaviour, - temporal stability, - cross-language consistency, - cross-country consistency, - cross-product consistency, - intervention effect, - recommendation certainty, - commercial impact, - conformity, - or a GEO score. All broader claims MUST be supported by a predefined target population, valid multi-observation design, recorded dependence structure, evidence bundle, evaluation rule, scope statement, and uncertainty disclosure. A single high-harm output MAY trigger a separate critical-incident review without being treated as evidence of population prevalence.","character_count":45534,"record_sha256":"6c036c34805ebc5f6dbc172f5c313a449257eee7d5758272558aa56d0d38f114"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-02","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":4,"chapter_number":2,"item_number":null,"title":"Representation Is a Distribution","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K05","K09"],"source_word_count":6390,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":2,\"id\":\"NOMOS-GEO-AUDIT-CH02\",\"title\":\"Representation Is a Distribution\",\"subtitle\":null,\"sourceFile\":\"2.ci bölüm.docx\",\"sourceSha256\":\"CF157E98388D9F58E9BEF9F44ADF14CD5B513AC59035C47BFA391660D30BE6D5\",\"sourceWordCount\":6390,\"sourceIds\":[\"K04\",\"K05\",\"K09\"],\"machine\":{\"chapter\":2,\"chapterId\":\"NOMOS-GEO-AUDIT-CH02\",\"title\":\"Representation Is a Distribution\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K05\",\"K09\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH02-R01\",\"normativeRuleEnglish\":\"An entity's generative-system representation MUST be treated as a conditional distribution of observable outcomes across defined: - AI products and user surfaces, - users or sampling units, - countries, - languages and locales, - session and personalisation states, - prompts, - and measurement times. A single aggregate score MUST NOT replace disclosure of: - scope and coverage, - outcome distribution, - subgroup variation, - critical-error incidence, - unknown and untested mass, - temporal stability, - weighting rules, - and uncertainty. Equal averages MUST NOT be interpreted as equivalent representation quality when their subgroup, severity, coverage, or stability distributions materially differ. Untested groups MUST NOT be imputed as conforming, and rare high-harm outcomes MUST remain separately visible from average performance.\",\"normativeRuleSourceTurkish\":\"Bir varlığın üretken sistemlerdeki temsili; tanımlı AI ürünleri ve kullanıcı yüzeyleri, kullanıcılar veya örnekleme birimleri, ülkeler, diller ve locale’ler, oturum ve kişiselleştirme durumları, promptlar ve ölçüm zamanları arasında oluşan koşullu gözlenebilir sonuç dağılımı olarak değerlendirilmelidir. Tek bir toplulaştırılmış skor; kapsamı, sonuç dağılımını, alt grup farklarını, Critical hata oranını, bilinmeyen ve test edilmemiş kütleyi, zaman kararlılığını, ağırlıkları ve belirsizliği görünmez kılamaz. Aynı ortalama, maddi ölçüde farklı dağılımlar için eşdeğer temsil kalitesi anlamına gelmez.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH02-MB0001\",\"type\":\"paragraph\",\"text\":\"RULE READABLE BY MACHINE FOR CHAPTER 31\",\"sourceParagraph\":1312},{\"blockId\":\"CH02-MB0002\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH02-R01\",\"sourceParagraph\":1313},{\"blockId\":\"CH02-MB0003\",\"type\":\"paragraph\",\"text\":\"An entity's generative-system representation MUST be treated as a\",\"sourceParagraph\":1315},{\"blockId\":\"CH02-MB0004\",\"type\":\"paragraph\",\"text\":\"conditional distribution of observable outcomes across defined:\",\"sourceParagraph\":1316},{\"blockId\":\"CH02-MB0005\",\"type\":\"paragraph\",\"text\":\"- AI products and user surfaces,\",\"sourceParagraph\":1318},{\"blockId\":\"CH02-MB0006\",\"type\":\"paragraph\",\"text\":\"- users or sampling units,\",\"sourceParagraph\":1319},{\"blockId\":\"CH02-MB0007\",\"type\":\"paragraph\",\"text\":\"- countries,\",\"sourceParagraph\":1320},{\"blockId\":\"CH02-MB0008\",\"type\":\"paragraph\",\"text\":\"- languages and locales,\",\"sourceParagraph\":1321},{\"blockId\":\"CH02-MB0009\",\"type\":\"paragraph\",\"text\":\"- session and personalisation states,\",\"sourceParagraph\":1322},{\"blockId\":\"CH02-MB0010\",\"type\":\"paragraph\",\"text\":\"- prompts,\",\"sourceParagraph\":1323},{\"blockId\":\"CH02-MB0011\",\"type\":\"paragraph\",\"text\":\"- and measurement times.\",\"sourceParagraph\":1324},{\"blockId\":\"CH02-MB0012\",\"type\":\"paragraph\",\"text\":\"A single aggregate score MUST NOT replace disclosure of:\",\"sourceParagraph\":1326},{\"blockId\":\"CH02-MB0013\",\"type\":\"paragraph\",\"text\":\"- scope and coverage,\",\"sourceParagraph\":1328},{\"blockId\":\"CH02-MB0014\",\"type\":\"paragraph\",\"text\":\"- outcome distribution,\",\"sourceParagraph\":1329},{\"blockId\":\"CH02-MB0015\",\"type\":\"paragraph\",\"text\":\"- subgroup variation,\",\"sourceParagraph\":1330},{\"blockId\":\"CH02-MB0016\",\"type\":\"paragraph\",\"text\":\"- critical-error incidence,\",\"sourceParagraph\":1331},{\"blockId\":\"CH02-MB0017\",\"type\":\"paragraph\",\"text\":\"- unknown and untested mass,\",\"sourceParagraph\":1332},{\"blockId\":\"CH02-MB0018\",\"type\":\"paragraph\",\"text\":\"- temporal stability,\",\"sourceParagraph\":1333},{\"blockId\":\"CH02-MB0019\",\"type\":\"paragraph\",\"text\":\"- weighting rules,\",\"sourceParagraph\":1334},{\"blockId\":\"CH02-MB0020\",\"type\":\"paragraph\",\"text\":\"- and uncertainty.\",\"sourceParagraph\":1335},{\"blockId\":\"CH02-MB0021\",\"type\":\"paragraph\",\"text\":\"Equal averages MUST NOT be interpreted as equivalent representation\",\"sourceParagraph\":1337},{\"blockId\":\"CH02-MB0022\",\"type\":\"paragraph\",\"text\":\"quality when their subgroup, severity, coverage, or stability\",\"sourceParagraph\":1338},{\"blockId\":\"CH02-MB0023\",\"type\":\"paragraph\",\"text\":\"distributions materially differ.\",\"sourceParagraph\":1339},{\"blockId\":\"CH02-MB0024\",\"type\":\"paragraph\",\"text\":\"Untested groups MUST NOT be imputed as conforming, and rare high-harm\",\"sourceParagraph\":1341},{\"blockId\":\"CH02-MB0025\",\"type\":\"paragraph\",\"text\":\"outcomes MUST remain separately visible from average performance.\",\"sourceParagraph\":1342},{\"blockId\":\"CH02-MB0026\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1343},{\"blockId\":\"CH02-MB0027\",\"type\":\"paragraph\",\"text\":\"The representation of an entity in generative systems should be evaluated as the conditional observable outcome distribution arising between defined AI products and user interfaces, users or sampling units, countries, languages and locales, session and personalisation states, prompts, and measurement times. A single aggregated score cannot make the scope, outcome distribution, subgroup differences, critical error rate, unknown and untested population, temporal stability, weights, and uncertainty invisible. The same average does not mean equivalent representation quality for materially different distributions.\",\"sourceParagraph\":1344}]}}","text":"## Chapter Boundary\n\nSection 1 established the following provision:\n\n> A single response is not the GEO score. A single response is a singular observation that occurred under recorded conditions.\n\nThis section answers the next question:\n\n> If a single response is not the GEO score, what is the actual object we want to measure?\n\nThe answer is:\n\n> The representation of an entity in generative systems is a multidimensional distribution formed across AI products, users, countries, languages, session conditions, prompts, and times.\n\nThis chapter:\n\n- the representation distribution,\n\n- the fundamental dimensions of the distribution,\n\n- the difference between the mean and the distribution,\n\n- the subgroup inequality,\n\n- the stability,\n\n- the coverage and missing observation problem,\n\n- the risk of rare severe errors,\n\n- the conceptual structure of the publicly available result card\n\ndefines. This chapter does not yet:\n\n- how many participants will be assigned to each country,\n\n- how the sample frame will be established,\n\n- how prompt translations will be validated,\n\n- how atomic claims will be scored,\n\n- the exact formula of the final NOMOS score,\n\n- conformity thresholds\n\nThis chapter does not settle these matters normatively; later chapters address them separately. Before defining the scoring formula, Chapter 2 defines the reality that the score is intended to represent.\n\n## NOMOS Challenge\n\nConsider two AI products. Each receives the same global average score:\n\n#### 90 / 100\n\nThe first product's language outcomes are as follows:\n\n- English: 90\n\n- Turkish: 91\n\n- German: 89\n\n- Japanese: 90\n\n- Arabic: 90\n\nThe language results of the second product are as follows:\n\n- English: 99\n\n- Turkish: 98\n\n- German: 97\n\n- Japanese: 86\n\n- Arabic: 70\n\nThe simple average of both products is 90. Do they have the same GEO quality? No. The first product produces a similar representation in different languages. The second product produces excellent representation in some languages but seriously weak representation in another language. A single average hides this difference. Now consider two other products. 95% of the responses of both meet the passing criteria. The remaining 5% in the first product consists of:\n\n- small scope deficiencies,\n\n- limited explanation shortcomings,\n\n- a few UNKNOWN\n\nresults. The remaining 5% in the second product consists of:\n\n- incorrect health authority,\n\n- nonexistent licence,\n\n- incorrect legal identity,\n\n- inappropriate advice\n\nexists. The raw accuracy rate for both is again 95 per cent. Do they have the same representation risk? No. Now consider a third pair. The average of both across the three measurement waves is 90. The first one:\n\n- Wave 1: 90\n\n- Wave 2: 91\n\n- Wave 3: 89\n\nThe second one:\n\n- Wave 1: 98\n\n- Wave 2: 90\n\n- Wave 3: 82\n\nDo they have the same consistency over time? No. In that case, the issue is not only: 'What is the average?' We also need to know the following: How are the results distributed? Who is well represented, who is poorly represented? Which languages and countries are falling behind? How severe are the errors? How much does the result change over time? Which areas were never measured? With what weights was the average created? How many different realities lie behind the same average? The first conclusion of this section is:\n\n> The same average is not the same representation.\n\nIts second provision states:\n\n> a GEO score may be a summary of the distribution; it cannot replace the distribution.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThis section defines how the representation of entities in generative systems should be understood before being reduced to a single number. Its main objectives are:\n\n- To show that representation is not a timeless and singular “AI perspective\"\n\n- To define the conditions on which the representation outcome depends\n\n- To explain that the true–false binary alone is insufficient\n\n- To separate average accuracy from subgroup inequality\n\n- To make visible the heavy error risk hidden within high averages\n\n- To show that stability and accuracy are not the same concept\n\n- To prevent unmeasured areas from being filled with positive assumptions within the distribution\n\n- To prepare the structure of a future NOMOS score that is singular but not condemned to being singular\n\nAt the end of this section, the reader should be able to answer the following question:\n\n> Why is the result of an entity's GEO not limited to only the average accuracy rate?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\n> The representation of an entity in generative systems is the conditional distribution of different responses and representation outcomes that can occur under defined conditions.\n\nThis distribution:\n\n- it is not the abstract 'thought' of a single model,\n\n- it is not the common view of all artificial intelligence systems,\n\n- it is not only the repetition rate of the website,\n\n- it's not only the frequency with which the brand's name is mentioned,\n\n- it's not just the average accuracy,\n\nit is not only the recommendation rate. The representation distribution, within a certain protocol, shows the following:\n\n> Which users see what kind of representation on which AI product and user surface, in which language, in which country, with which prompt, under which session condition, and at what time.\n\nTherefore, the result of a distribution is always dependent on the following questions: Which entity? Which AI product? Which user surface? Which target population? Which country? Which language and locale? Which prompt? Which session condition? Which measurement wave? Which Truth Pack? Which evaluation rule? When any of these fields change, the distribution may also change.\n\n## 3. WHAT DOES THE WORD \"DISTRIBUTION\" MEAN?\n\nIn this book, the word distribution does not refer to a hidden or metaphysical view that is assumed to exist within an artificial intelligence system. It does not claim: NOMOS: \"This is the real distribution of the company in the model's mind.\" The distribution here is operational and observable.\n\n> It is the pattern of different representation results that can be obtained under a defined protocol within the target population and measurement conditions.\n\nTwo distributions should be distinguished.\n\n### 3.1. Target Representation Distribution\n\nIt is the distribution that the measurement tries to predict but cannot directly observe in full. It is represented as:\n\n### P_E(Y | A, C, L, S, P, T, U)\n\nHere:\n\n- E: audited entity\n\n- Y: representation result\n\n- A: AI product and user surface\n\n- C: country or geographic condition\n\n- L: language and locale\n\n- S: session and personalisation condition\n\n- P: prompt\n\n- T: time or measurement wave\n\n- U: target user universe\n\nThis expression asks: “In the defined target user universe, under the specified conditions, which representation results occur with what probabilities?” Not all elements of the target distribution can be measured. They are estimated through sampling.\n\n### 3.2. Observed Empirical Distribution\n\nIt is the distribution of valid observations obtained in real or synthetic measurement. Its representation:\n\n### P̂_E\n\ncan be used. The empirical distribution is not the target distribution itself. It is an estimate produced through its sample, weighting, and measurement design. The magnitude of the difference may depend on:\n\n- Sample design\n\n- Participant selection\n\n- Country and language coverage\n\n- Invalid observation rate\n\n- Non-response\n\n- Adjudicator error\n\n- Measurement wave\n\n- Prompt selection\n\n- Product and interface conditions\n\n- Weighting\n\n- Clustering and dependence\n\nTherefore, an empirical distribution cannot be published as \"the absolute and immutable truth of artificial intelligence.\" The correct expression should be: \"The estimate obtained for the specified target population and measurement conditions.\"\n\n## 4. REPRESENTATION SPACE\n\nThe representation of an entity does not occur in a single dimension. Each observation exists in a multidimensional condition space. The basic representation:\n\nX_i = (E_i, A_i, U_i, C_i, L_i, S_i, P_i, T_i)\n\nHere:\n\n- Ei: entity\n\n- Ai: AI product and user surface\n\n- Ui: user or sampling unit\n\n- Ci: country and geographical layer\n\n- Li: language and locale\n\n- Si: session, plan, and personalisation condition\n\n- Pi: prompt or prompt family\n\n- Ti: time and measurement wave\n\nGenerated response:\n\nR_i\n\ncan be shown. The representative state resulting from the evaluation of the response is:\n\nY_i = g(R_i, V_E, J)\n\nand can be considered as. Here:\n\n- VE: verified Truth Pack used for the entity\n\n- J: assessment and adjudication rules\n\n- g: evaluation process that converts the response into representative results\n\nThe meaning of this representation is as follows: The same response cannot be scored on its own without indicating which entity, which Truth Pack, and which evaluation rule it was examined according to.\n\n## 5. REPRESENTATIVE RESULT IS NOT A SINGLE VALUE\n\nOnly the entirety of a response:\n\n- can be forced into the categories of\n\ntrue,\n\n- is wrong\n\nGiving a label is often insufficient. The following situations can occur together in the same response:\n\n- Correct identity\n\n- Correct main category\n\n- Incomplete service coverage\n\n- Outdated number of employees\n\n- Unsupported leadership claim\n\n- Correct citation\n\n- Inappropriate recommendation\n\n- Non-critical boundary deficiency\n\nTherefore, the representation result can be considered as a result vector. Candidate representation:\n\nY_i = (y_i^id, y_i^cat, y_i^fact, y_i^scope, y_i^time, y_i^evidence, y_i^recommendation, y_i^harm)\n\nHere:\n\n- yiid: identity accuracy\n\n- yicat: accuracy of category and role\n\n- yifact: accuracy of factual matters\n\n- yiscope: preservation of scope and boundaries\n\n- yitime: temporal validity\n\n- yievidence: claim–evidence consistency\n\n- yirecommendation: recommendation and user suitability\n\n- yiharm: harm class of error or misguidance\n\nThe exact scoring rules for these fields will be defined in later sections. The guiding principle here is:\n\n> A representative answer should be evaluated not only on whether it contains correct information but also on whom, what, within which boundaries, with what evidence, and for whom it is appropriately conveyed.\n\n## 6. BASIC REPRESENTATION STATES\n\nThe status of each atomic claim and final response may not be the same. This section conceptually separates the following basic states.\n\n### 6.1. SUPPORTED\n\nThe claim is supported by sufficient evidence within the defined scope and time.\n\n### 6.2. CONTRADICTED\n\nThe claim conflicts with reliable and relevant records.\n\n### 6.3. UNSUPPORTED\n\nThere is not enough support for the claim. It has not been proven that the claim is necessarily false. However, there are not enough records for it to be accepted as true.\n\n### 6.4. OUTDATED — NO LONGER CURRENT\n\nThe claim may have been true in the past. As of the measurement date, it is no longer valid or its current accuracy cannot be verified.\n\n### 6.5. PARTIALLY SUPPORTED — PARTIALLY SUPPORTED\n\nOnly a specific scope, population, country, product, or time of the claim is supported. The response may have exceeded the support limit.\n\n### 6.6. OUT OF SCOPE — OUT OF SCOPE\n\nThe claim is not related to the audited entity or Truth Pack. It may have come from another entity, product, country, or service.\n\n### 6.7. ENTITY CONFLATION — ENTITY MERGER\n\nSystem:\n\n- two different companies,\n\n- the brand and the legal entity,\n\n- a person and the institution,\n\n- old and new entity,\n\n- has confused different organisations with the same name\n\nwith each other. This error is not merely a factual mistake. It is the corruption of the representation object.\n\n### 6.8. MATERIAL OMISSION — MISSING INFORMATION\n\nThe sentences in the response may be correct. However, essential information that would change the user's decision has been omitted. Example: The service is offered only in certain countries. The licence is limited to a specific jurisdiction. Price is excluding tax. The result is based on first-party data. The product is not suitable for certain users. The omission can create a wrong inference even if there is no directly incorrect sentence.\n\n### 6.9. REFUSAL — REJECTION\n\nThe system has refused to answer. Rejection:\n\n- appropriate,\n\n- unnecessary,\n\n- inconsistent,\n\n- security-based,\n\n- technical\n\nmay occur. Ret is not automatically considered failed or successful. It is classified separately.\n\n### 6.10. NO RESULT\n\nThe system did not produce a usable response due to technical or operational reasons. It is not the same as Ret.\n\n### 6.11. UNKNOWN\n\nThe available records are not sufficient for a reliable classification. UNKNOWN cannot be converted into a positive result.\n\n## 7. MAIN CHARACTERISTICS OF THE DISTRIBUTION\n\nA representation distribution cannot be explained solely by the mean. At a minimum, the following characteristics should be evaluated.\n\n### 7.1. Central Tendency\n\nSummarises the general accuracy or transition level of the responses. Example measures:\n\n- Mean\n\n- Weighted mean\n\n- Median\n\n- pass rate\n\nCentral tendency is important. However, it only shows the centre of the distribution.\n\n### 7.2. Spread and Variability\n\nIndicates how much the results are spread around the centre. Of two systems with the same mean, one may be consistent among users, while the other may be highly variable. Variability:\n\n- users,\n\n- countries,\n\n- languages,\n\n- sessions,\n\n- measurement waves\n\ncan also be examined among them.\n\n### 7.3. Shape of the Distribution\n\nAre the results clustered around a single central point? Or are they divided into two or more clusters? For example:\n\n- If half of the users see very accurate,\n\n- and half see very inaccurate\n\nresponses, the average may be medium. However, the typical user experience is not \"medium accuracy.\" It is two separate polarised experiences.\n\n### 7.4. Subgroup Differences\n\nDo specific language, country, or user groups systematically see different results? The global average may hide this difference.\n\n### 7.5. Tail Risk\n\nAre there low-frequency but severe errors? Rare events can have a small effect on the average. Their impact on human harm can be significant.\n\n### 7.6. Coverage\n\nHow much of the target user universe has actually been measured? A high score can be found together with low coverage.\n\n### 7.7. Missing and Unknown Mass\n\nWhat proportion of the responses are:\n\n### UNKNOWN,\n\n### NOT TESTED,\n\nrejection, technical non-response or an invalid record?\n\n### 7.8. Time Stability\n\nTo what extent is the distribution preserved across different measurement waves?\n\n## 8. CANDIDATE NOMOS REPRESENTATIVE DISTRIBUTION SIGNATURE\n\n### CANDIDATE CONCEPT\n\nIt is recommended that the result of a GEO be defined first by the following distribution signature instead of a single number:\n\nRDS(E) = (μ, κ, σ, Δ, τ, ρ, u)\n\nThis representation is called Representation Distribution Signature, that is:\n\n#### Representation Distribution Signature\n\nComponents:\n\n#### μ — General Representation Level\n\nOverall accuracy or transition prediction under defined weights and scope.\n\n#### κ — Coverage\n\nThe extent to which the target user, country, language, AI product, and time universe are sufficiently measured.\n\n#### σ — Variability\n\nSpread among users and experimental cells.\n\n#### Δ — Subgroup Inequality\n\nThe material difference between language, country, product, or other significant groups.\n\n#### 'tau — Tail and Critical Error Risk\n\nThe rate and nature of rare severe errors.\n\n#### ρ — Temporal Stability\n\nThe resilience of distribution across waves.\n\n#### u — Unknown and Unresolved Mass\n\nUNKNOWN is the share of the distribution that remains unresolved or insufficiently assessed. This signature is not yet a final scoring formula. The following elements require further development:\n\n- definite definition,\n\n- estimation,\n\n- weight,\n\n- public presentation\n\nThey will be finalised through pilot studies and later chapters. The founding principle is:\n\n> The GEO result should answer not only the question 'how accurate?' but also 'for whom, how consistent, how equitable, how comprehensive, and how risky?'\n\n## 9. AVERAGE WASHING\n\nIf an institution only shows a high global average and hides the material flaws of the distribution, this is called:\n\n#### Average washing\n\nAverage washing can occur in the following ways:\n\n- Hiding the weakest language\n\n- Excluding small countries from the report\n\n- Mixing critical errors into the general error rate\n\n- Removing UNKNOWN results from the denominator\n\n- Only including well-performing AI products\n\n- Excluding the weak measurement wave\n\n- Making the population outside the scope invisible\n\n- Changing weights after seeing the results\n\nThe average may be mathematically calculated correctly. However, the representation it offers to the public can still be misleading. Therefore:\n\n> A mathematically correct calculation of an average does not mean it is sufficient for decision-making.\n\n## 10. DIFFERENT REALITIES OF THE SAME AVERAGE\n\nThe average could be 90 in all of the following synthetic distributions.\n\n### 10.1. Homogeneous Distribution\n\nAll major subgroups are in the range of 88–92. This distribution is relatively balanced.\n\n### 10.2. Polarised Distribution\n\nHalf of the users see results close to 100, while the other half see results close to 80. The average is 90. However, the user experience is not equal.\n\n### 10.3. Linguistically Fragmented Distribution\n\nResults are very good for high-population or high-resource languages, but significantly weak for low-resource languages. The global average may look good. Language fairness is weak.\n\n### 10.4. Country-Based Fragmented Distribution\n\nEven within the same language, results may vary due to different countries, locales, product access, or user base.\n\n### 10.5. Temporally Volatile Distribution\n\nVery high results occur in one wave, very low results in another wave. Even if the average is high, it is difficult to establish reliable user expectations.\n\n### 10.6. Distribution with Tail Risk\n\nMost of the responses are correct. However, a small portion contains serious health, legal, licensing, or safety errors.\n\n### 10.7. High Unknown Mass Distribution\n\nMost of the assessable responses are correct. However, a significant portion of the target population or responses is:\n\n- tested,\n\n- untested,\n\n### UNKNOWN\n\nor inconclusive. Publishing a high rate based only on the assessable responses may obscure the scope.\n\n## 11. SUBGROUP INEQUALITY\n\nA global score may not represent the experience in subgroups. Key subgroups may include:\n\n- AI product and user surface\n\n- Country\n\n- Language\n\n- Locale\n\n- Plan\n\n- Web or mobile\n\n- Controlled or natural panel\n\n- Time wave\n\n- User intent\n\n- Device family\n\nNot all of these groups may have the same importance in every audit. Material groups should be determined when scope is locked.\n\n### 11.1. Language is Not the Same as Country\n\nTurkish-speaking:\n\n- User in Turkey,\n\n- User in Germany,\n\n- User in another country\n\nThey can use the same language. However:\n\n- product access,\n\n- price,\n\n- resources,\n\n- local interface,\n\n- legal context,\n\n- personalisation\n\nmay be different. Therefore, all differences cannot be explained solely by language or country.\n\n### 11.2. Intersectional Cells\n\nSome material differences only become visible when multiple areas are examined together. Example: Germany × Turkish × Mobile × Free plan × Natural User Panel could in itself be an experimental cell. However, as cells multiply, the sample size decreases. Therefore, a definitive score cannot be published for each small cell. Measurement design: should preserve meaningful subgroups and should not produce statistical false precision.\n\n### 11.3. Subgroup Base\n\n### CANDIDATE CONCEPT\n\nIn addition to the global average, it is recommended to report the result of the weakest adequately sampled subgroup or the bottom 10% group separately. This can be referred to as:\n\n#### Representation base\n\nor:\n\n#### Language/Country Floor\n\nHowever, if the \"lowest score\" comes from a small and unstable cell, it may produce incorrect results. Therefore, the base metric should be defined together with:\n\n- minimum sample,\n\n- confidence interval,\n\n- cell quality\n\nThe exact rule will be determined in the following sections.\n\n## 12. IT IS PART OF THE SCOPE DISTRIBUTION\n\nA distribution consists not only of observed responses. What was never observed also determines the meaning of the result. For example, English and Turkish were tested; Japanese and Arabic were not. The web interface was tested; the mobile interface and paid plan were not. Strong English and Turkish results do not justify the statement, ‘Representation is strong across all languages and user surfaces.’ The defensible conclusion is: ‘Measurements were conducted on the specified English and Turkish web surfaces. Other languages and mobile surfaces were not tested.’\n\n### 12.1. Coverage Map\n\nEach GEO distribution should carry the following map to the extent relevant:\n\n- Target universe\n\n- Sampling frame\n\n- Covered countries\n\n- Non-covered countries\n\n- Covered languages\n\n- Non-covered languages\n\n- Monitored AI products\n\n- Unmonitored products\n\n- Monitored user surfaces\n\n- Unmonitored surfaces\n\n- Measurement waves\n\n- Out-of-scope prompt families\n\n### 12.2. Coverage Rate\n\n### CANDIDATE CONCEPT\n\nA coverage metric can be developed to show what proportion of the target universe is represented with sufficient measurement. However, coverage:\n\n- only the number of countries,\n\n- only the number of languages,\n\n- only the population ratio\n\ncannot be measured. For example, 150 countries may be covered. However, 95% of the participants may have come from only five countries. Or 90% of the population may be covered, but small languages may be entirely excluded. Therefore, population coverage should be shown separately from country/language observer coverage.\n\n## 13. MISSING OBSERVATION AND NON-RESPONSE\n\nMissing data is not only a technical issue. Its distribution can be systematically distorted. The reasons for an observation being missing may include:\n\n- Participant's failure to complete the task\n\n- Product not being accessible in the country\n\n- Computation constraints\n\n- Technical error\n\n- System rejection\n\n- Invalid screen capture\n\n- Participant changing their prompt\n\n- Product not working in a specific language\n\n- Adjudicators being unable to make reliable decisions\n\nEach of these reasons requires a separate status.\n\n### 13.1. Missing Data May Not Be Random\n\nFor example, if more tasks cannot be completed in countries with low connection quality, using only successful recordings can shift the global result in favour of wealthy users with strong connections. If the rate of UNKNOWN rises because answers in a specific language are difficult to evaluate, the score calculated from only known answers may hide language inequality. Therefore:\n\n> The missing data rate and the distribution report of the reason for missing data should be part of the report.\n\n### 13.2. Denominator — Integrity of the Denominator\n\nA report should answer the question clearly: \"90 per cent of exactly what is 90 per cent?\" Possible denominators:\n\n- Assigned participants\n\n- Task initiators\n\n- Task completers\n\n- Protocol-compliant records\n\n- Valid system results\n\n- Responses with adjudicator decisions\n\n- Specific language or country cell\n\nThe denominator should not be changed after the result is seen. For example:\n\n- 1,000 tasks assigned,\n\n- 900 completed,\n\n- 800 valid,\n\n- 720 passed\n\nThus:\n\n720/800=90%\n\nThis is the pass rate among valid records. The observed pass rate across the entire assigned sample is instead:\n\n720/1000=72%\n\nanswers a different question. Both rates can be used. They cannot be presented in the same sense.\n\n## 14. ACCURACY AND CONSISTENCY ARE NOT THE SAME\n\nA distribution may have high accuracy. However, it may be inconsistent over time or across users. Another distribution may have low accuracy. However, it may consistently be wrong. Therefore, four situations should be distinguished.\n\nBeing “consistent” is not automatically positive. The system may be quite consistent if it gives the same error to every user.\n\n### 14.1. Inter-User Consistency\n\nDo different users see similar material representation in the same cell?\n\n### 14.2. Inter-Session Consistency\n\nDoes the same or matched user see similar results in new sessions?\n\n### 14.3. Inter-Wave Consistency\n\nDoes the same protocol produce a similar distribution on different days or periods?\n\n### 14.4. Post-Product Update Consistency\n\nWhen there is a material change in the AI product, does the previous distribution continue?\n\n## 15. PRODUCING THE SAME SENTENCE IS NOT STABILITY\n\nStability: It does not mean \"giving the exact same answer to every user word for word.\" An AI product can use different expressions, order, and examples. As long as they preserve the material meaning, they can be acceptable variations. The important distinction is:\n\n#### Formal Variability\n\nWord choice Sentence order Length Tone Example Using list or paragraph If material truth is preserved, it may not be an error.\n\n#### Semantic Variability\n\nChange of company identity Differentiation of the main category Expansion or contraction of the field of activity Change of licence or competency Change of price and geography Recommending to one user but not recommending to a similar user without explanation If the material representation changes, there may be a stability problem. Therefore, the preferred concept is:\n\n#### Conditional semantic stability\n\nIt should be. The fundamental reality is expected to be preserved under equivalent material conditions. A change in recommendation due to different user needs can be appropriate contextual behaviour.\n\n## 16. CONTEXT SENSITIVITY AND ARBITRARY VARIABILITY\n\nIt is not always correct for a system to give the same recommendation to all users. For example:\n\n- Users with a low budget\n\n- Only users seeking services in a specific country\n\n- Users requiring a different licence\n\n- Users seeking individual rather than corporate services\n\ncan receive different conformity outcomes for the same entity. This difference:\n\n- may be due to the user condition,\n\n- the scope of service,\n\n- or the actual limit\n\nand may be context-sensitive and accurate. However, if the outcome changes without reason under equivalent user conditions, it may be arbitrary variability. Therefore, NOMOS does not want: “The same answer for all users.” It wants:\n\n> A response that is consistent and explainable under equivalent material conditions and sensitive to actual limits under different conditions.\n\n## 17. TIME IS WITHIN THE REPRESENTATION DISTRIBUTION\n\nThe representation distribution is not timeless. It can be thought of as PE,t. The distribution of the same entity:\n\n- product update,\n\n- resource change,\n\n- company change,\n\n- news event,\n\n- new language version,\n\n- technical access change\n\ncan diverge afterwards.\n\n### 17.1. Sudden Change\n\nAfter a product or resource update, the distribution changes quickly in a short time.\n\n### 17.2. Gradual Change\n\nDue to new resources and user behaviours, the result slowly changes over time.\n\n### 17.3. Temporary Fluctuation\n\nDuring a specific event or news period, the distribution temporarily differs.\n\n### 17.4. Post-Intervention Change\n\nAfter the GEO intervention, the distribution may change. However, claiming that the change is due to the intervention also requires a baseline and causality design.\n\n### 17.5. Smoothing with Time Average\n\nFor a product:\n\n- in the first wave 98,\n\n- in the second wave 90,\n\n- in the third wave 82\n\nIf taken, the average is 90. However, publishing 90 hides the downward trend. Therefore:\n\n> The time average cannot replace the wave results.\n\n## 18. TAIL RISK AND HEAVY ERROR DISTRIBUTION\n\nA large portion of a distribution may be correct. However, low-probability heavy errors should also be evaluated. Examples:\n\n- Health licence that does not actually exist\n\n- Incorrect emergency guidance\n\n- Incorrect legal authority\n\n- Fake investment guarantee\n\n- Incorrect company identity\n\n- Inappropriate advice for children or vulnerable groups\n\n- Repetition of the manipulative instruction hidden from people\n\nIn the previous architecture:\n\n- Critical,\n\n- Major,\n\n- Moderate,\n\n- Advisory\n\nwere separated into four levels of importance, and it was suggested that Critical findings should have an automatic failure effect. This book links this system to population distribution.\n\n### 18.1. Separation of Frequency and Severity\n\nAn error:\n\n- frequent but low impact,\n\n- rare but severe\n\nmay occur. However, the frequency rate alone cannot evaluate the two cases equally.\n\n### 18.2. Candidate Severity-Weighted Risk Representation\n\n### CANDIDATE CONCEPT\n\nA risk indicator can be developed using severity coefficients for error classes:\n\ntau = Σ_{k=1}^K λ_k p_k\n\nHere:\n\n- pk: predicted rate of the error type\n\n- λk: severity coefficient of the error type\n\nHowever, coefficients cannot be determined arbitrarily. They should be based on:\n\n- Foreseeable human harm\n\n- Representation integrity\n\n- Impact on user decision\n\n- Reversibility\n\n- Intent and manipulation\n\n- High-risk area\n\nThe precise method will be defined in Section 15.\n\n### 18.3. Critical Gate\n\nA high overall average cannot automatically delete a verified Critical error candidate. The correct public record should show both results together: “The overall representation estimate is high.” and: “The specified Critical event or events also affect the conformity decision.”\n\n## 19. REFUSAL AND FAILURE IN DISTRIBUTION\n\nA system not responding is also part of the user experience. However, not all refusals are the same.\n\n### 19.1. Appropriate Refusal\n\nThe system may exercise caution when evidence is insufficient or when it cannot provide a safe response. This can be the correct behaviour in some contexts.\n\n### 19.2. Unnecessary Refusal\n\nIt may inconsistently avoid responding to publicly available and low-risk information.\n\n### 19.3. Technical Nonresponse\n\nThe system may not produce a response due to a technical error.\n\n### 19.4. Out-of-Scope Rejection\n\nThe prompt may be incorrect or outside the task protocol.\n\n### 19.5. Preservation of Rejections in Distributions\n\nRejections:\n\n- success,\n\n- failure,\n\n### UNKNOWN\n\nshould not be forced into a single category. Their function and justification should be evaluated. Removing rejections from the denominator may yield a high accuracy rate only among successful responses and obscure the real user experience.\n\n## 20. HOW CAN DISTRIBUTIONS BE COMPARED?\n\nSimply comparing the result numbers is not sufficient to compare the distribution of two AI products. For the comparison to be fair, the following areas should be as equal as possible or explicitly modelled:\n\n- Audited entity\n\n- Target population\n\n- Prompt version\n\n- Languages\n\n- Countries\n\n- Measurement wave\n\n- Session condition\n\n- User surface\n\n- Plan distribution\n\n- Adjudication rule\n\n- Reality package\n\n- Validity rules\n\n- Weights\n\nTwo scores obtained with materially different protocols cannot be used directly for ranking.\n\n### 20.1. Common Support Comparison\n\nAll compared AI products:\n\n- are officially accessible,\n\n- support the same language and user conditions\n\ncan be compared in a common universe. This provides a cleaner comparison between systems.\n\n### 20.2. Natural Access Distribution\n\nEach AI product can be measured in its own real user access universe. This result shows the product's actual global impact. However, since target populations may differ, direct performance ranking requires caution.\n\n### 20.3. Different Surfaces of the Same Provider\n\nWeb, mobile, enterprise products, or different plans, even if they belong to the same provider, can create separate distributions. If they are to be combined under a single result, the combination and weighting method should be explained.\n\n## 21. WEIGHTING CHANGES THE MEANING OF THE DISTRIBUTION\n\nA global average depends on the weight with which observations are combined. Possible weights:\n\n- Equal by country\n\n- By country population\n\n- Appropriate AI user population\n\n- Language user\n\n- Actual product exposure\n\n- Equal AI product\n\n- Market share\n\n- Customer target market\n\nEach method answers a different question.\n\n### 21.1. Equal Country Weight\n\nSan Marino and India can receive the same weight. This approach can address the question of 'Average country experience.' It does not represent the average user experience in the world population.\n\n### 21.2. Population Weight\n\nLarger countries have more influence. This approach can approach the global population experience. It may render small countries and languages invisible.\n\n### 21.3. Language Fairness Weight\n\nIt can also ensure that low-resource languages remain visible. It is not the same as the global population score.\n\n### 21.4. Product Exposure Weight\n\nThe weighting of AI products according to actual user exposure can be considered. It requires reliable and independent usage data.\n\n### 21.5. Principle of Normative Weighting\n\n> Weights should be defined before the results are seen, publicly disclosed, and the question they answer should be specified.\n\nIt is prohibited to choose a weighting system that will increase the positive score after the result is seen. Alongside a single global score, alternative meaningful weighted results can also be published.\n\n## 22. SCORE DISTRIBUTION IS LOSSY COMPRESSION\n\nIt is possible to produce a single number from a distribution. However, a single number does not carry all the information in the distribution. Therefore, the score:\n\n#### Is a lossy summary of the distribution\n\nshould be accepted. A score may lose the following information:\n\n- Weakest language\n\n- Weakest country\n\n- Critical error type\n\n- Unknown mass\n\n- Time fluctuation\n\n- Polarisation\n\n- Product and plan difference\n\n- Out-of-scope population\n\n- Rejection distribution\n\n- Adjudicator ambiguity\n\nThe score is useful because:\n\n- facilitates comparison,\n\n- simplifies public communication,\n\n- allows monitoring change,\n\nsupports the management decision. However, the simplicity of the score cannot turn into the elimination of distribution.\n\n## 23. MINIMUM PUBLIC RESULTS CARD\n\n### CANDIDATE REQUIREMENT — CANDIDATE OBLIGATION\n\nIf a global score NOMOS is to be presented to the public, it is recommended that at least the following results be shown alongside it:\n\nThe main score should not be used as a standalone advertising label. The canonical record must be linked to this results card.\n\n## 24. SYNTHETIC APPLE.COM DISTRIBUTION DISPLAY\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The AI products, users, languages, numbers, and results below are entirely fictional. They do not represent the actual performance of Apple Inc. or any real AI product. Let us consider two synthetic AI products:\n\n- Product A\n\n- Product B\n\nFor each product:\n\n- five languages,\n\n- 200 valid synthetic observations per language,\n\n- a total of 1,000 observations\n\nare assumed. This equal language distribution is only to demonstrate the method. It does not represent the actual GEO-1000 population weighting.\n\n### 24.1. Language-Based pass rates\n\nThe average of the two products is the same. However, their distributions are different.\n\n#### Product A\n\nLanguage results are close to each other. The difference between the highest and lowest language is 2 points. The language floor is 89%.\n\n#### Product B\n\nProduces very high results in three languages. There is a decline in Japanese and especially in Arabic. The difference between the highest and lowest language is 29 points. The language floor is 70%. Note: Saying \"Both scored 90\" hides the material difference.\n\n### 24.2. Response Status Distribution\n\nThe pass rate of the two products is again 90%. However: In Product B, the number of Critical error candidates is 14 times higher. In Product B, the unknown and inconclusive mass is higher. A larger portion of failures in Product A is due to partial support deficiency. The error backlog of Product B is heavier. The same pass rate does not necessarily result in the same fitness outcome.\n\n### 24.3. Time Waves\n\nProduct A is relatively stable. Product B shows a noticeable decline. Publishing only the average of the three waves hides the deterioration in Product B.\n\n### 24.4. Synthetic Result\n\nIt may be possible to publish the same headline score for Product A and Product B:\n\n#### 90\n\nBut NOMOS result cards cannot be the same.\n\n#### Product A — Synthetic Distribution Signature\n\nOverall result: high Language inequality: low Critical error tail: low but not zero Unknown mass: low Time stability: high\n\n#### Product B — Synthetic Distribution Signature\n\nOverall result: high Language inequality: high Critical error tail: substantial Unknown mass: higher Time stability: low and declining\n\nTherefore, the synthetic conclusion of Section 2 is as follows:\n\n> Having the same average score for Product A and Product B does not indicate that they have the same quality of representation.\n\n## 25. MANDATORY NORMATIVE PROVISIONS\n\n**CH02-N01**\n\nA GEO result cannot be defined only by a single average or a single pass rate.\n\n**CH02-N02**\n\nEach distribution result:\n\n- the audited entity,\n\n- target population,\n\n- AI product and user surface,\n\n- country and language,\n\n- The prompt,\n\n- the residence condition,\n\n- to time\n\nshould be connected.\n\n**CH02-N03**\n\nThe global average alone cannot compensate for material subgroup inequality.\n\n**CH02-N04**\n\nCritical and other high-severity error outcomes should be reported separately from the overall error rate.\n\n**CH02-N05**\n\nUNKNOWN, no-result, rejected, invalid, and NOT TESTED statuses should be distinguished from each other.\n\n**CH02-N06**\n\nUnmeasured country, language, product, user surface, or time period cannot be shown as part of the positive distribution.\n\n**CH02-N07**\n\nWeights and aggregation rules should be defined before viewing results.\n\n**CH02-N08**\n\nPost-result weight changes should create a new analysis version, and previous results should be preserved.\n\n**CH02-N09**\n\nDistributions based on materially different target populations, demand, product, time, or evaluation criteria cannot be directly compared without explanation.\n\n**CH02-N10**\n\nFormal language variability cannot automatically be counted as representation error.\n\n**CH02-N11**\n\nMaterial identity, scope, time, competence, or recommendation variability cannot be classified as a style difference without justification.\n\n**CH02-N12**\n\nHigh stability alone is not evidence of quality; consistently wrong results should also be recorded.\n\n**CH02-N13**\n\nA high average, if accompanied by low coverage, should have public results with coverage limitation.\n\n**CH02-N14**\n\nSubgroup scores should not be published as definitive country or language judgements under insufficient sampling and broad uncertainty.\n\n**CH02-N15**\n\nRejection and technical failure rates should be preserved as part of the genuine user experience.\n\n**CH02-N16**\n\nThe only score made public should be linked to the canonical distribution and scope record.\n\n**CH02-N17**\n\nSynthetic distributions cannot be combined with the performance of a real AI product or real entity.\n\n**CH02-N18**\n\nThe distribution report should have an accountable human or institutional owner.\n\n## 26. FORMS OF FAILURE\n\n**CH02-F01 — AVERAGE THINKING**\n\nHigh global average is shown; language, country, or Critical error corruption is stored.\n\n**CH02-F02 — DELETE SUBGROUP**\n\nLow-performing groups are removed from the report or declared 'insignificant'.\n\n**CH02-F03 — SCOPE MAP**\n\nNarrow-scope good results are used for the whole world, all languages, or all products.\n\n**CH02-F04 — SEVERITY NORMALISATION**\n\nMinor deficiencies and serious health or licensing errors are combined within the same error count.\n\n**CH02-F05 — HIDE DETERIORATION WITH TIME AVERAGE**\n\nDecline or fluctuation is made invisible only with the period average.\n\n**CH02-F06 — WORSHIP OF STABILITY**\n\nA consistent answer is assumed to be automatically correct.\n\n**CH02-F07 — WORD-FOR-WORD SAME ANSWER FALLACY**\n\nLiteral text matching is used as a measure of consistency instead of semantic correctness.\n\n**CH02-F08 — COUNTING CONTEXTUAL DIFFERENCE AS ERROR**\n\nDifferences in appropriate recommendations based on actual user needs are marked as arbitrary variability.\n\n**CH02-F09 — COUNTING ARBITRARY VARIABILITY AS PERSONALISATION**\n\nMaterial contradictions under equivalent user conditions are defended as “personalisation” without explanation.\n\n**CH02-F10 — CHANGING THE DENOMINATOR**\n\nPost-result transition is made among completed, valid, or evaluated observation denominators to elevate the positive outcome.\n\n**CH02-F11 — POST-RESULT WEIGHTING**\n\nThe country, language, or model coefficients producing the highest score are selected afterward.\n\n**CH02-F12 — MERGING INCONSISTENT CELLS**\n\nDifferent prompt, product, time, or panel conditions are presented as a single distribution.\n\n**CH02-F13 — DELETING REJECTIONS**\n\nSystem records that did not respond are removed to generate high accuracy only among those that produced a response.\n\n**CH02-F14 — DEFINITE JUDGEMENT FROM SMALL CELL**\n\nA country or language score is produced from one or several users.\n\n**CH02-F15 — HIDING THE CRITICAL TAIL**\n\nRare severe errors are lost within the overall pass rate.\n\n**CH02-F16 — SINGLE SCORE REPORT**\n\nOnly the headline score is published without showing scope, inequality, stability, uncertainty, and severe errors.\n\n**CH02-F17 — COUNTING DISTRIBUTION AS \"AI'S BELIEF\"**\n\nObserved user surface results are presented as if they are the invariant and intrinsic viewpoint of artificial intelligence.\n\n**CH02-F18 — COUNTING THE UNMEASURED AS ZERO VARIANCE**\n\nUntested areas are considered smooth and stable.\n\n## 27. AUDIT PROCEDURE\n\nWhether a GEO report honestly represents the distribution should be examined with the following steps.\n\n### Step 1 — Record the Public Claim\n\nExample: “The brand's global NOMOS score is 92.” The full sentence and the surface where it is found are recorded.\n\n### Step 2 — Define Target Distribution\n\nFor which score:\n\n- user universe,\n\n- AI products,\n\n- countries,\n\n- languages,\n\n- time period\n\nis it applied?\n\n### Step 3 — Extract Measurement Cells\n\nEach material combination is determined:\n\n- Product\n\n- Country\n\n- Language\n\n- Panel\n\n- Prompt\n\n- Wave\n\n- User surface\n\n### Step 4 — Rebuild Denominators\n\nAssigned tasks, completed tasks, valid records, adjudicated records and passed responses must be reported as separate counts.\n\n### Step 5 — Extract Result Statuses\n\nNot only the pass rate:\n\n- partial,\n\n- unsupported,\n\n- conflicting,\n\n- old,\n\n- Critical,\n\n### UNKNOWN,\n\nrejected, inconclusive distribution is examined.\n\n### Step 6 — Verify Weights\n\nWeights:\n\n- are they predefined,\n\n- which question do they answer,\n\nhave they changed after the result?\n\n### Step 7 — Examine Subgroup Differences\n\nThe best group, the weakest sufficient group, language difference, country difference, product difference, and panel difference are evaluated.\n\n### Step 8 — Review Scope and Deficiency\n\nWhich areas were not measured? Which cells are insufficient? What are the reasons for the deficiency? Could the missing data be systematic?\n\n### Step 9 — Examine the Determination\n\nThe change between Waves Sessions Users Interfaces is evaluated.\n\n### Step 10 — Examine Queue Risk\n\nCritical and other serious errors:\n\n- number,\n\n- its rate,\n\n- type,\n\n- user effect\n\nis evaluated separately in terms of\n\n### Step 11 — Compare the Public Summary with the Distribution\n\nDoes the single number or sentence published to the public accurately convey the direction and risk of the full distribution?\n\n### Step 12 — Limit the Result\n\nThe public claim is limited only to the level carried by the distribution and scope.\n\n## 28. NECESSARY EVIDENCE\n\nTo the extent relevant for a representative distribution or a global GEO score, the following records should be searched:\n\n- Audited entity record\n\n- Definition of target population\n\n- Sampling frame\n\n- AI product and user interfaces\n\n- Country and language cells\n\n- Prompt and versions\n\n- Session and panel conditions\n\n- Measurement waves\n\n- Assigned task numbers\n\n- Completed tasks\n\n- Valid and invalid records\n\n- Raw responses\n\n- Result status distribution\n\n- Adjudication records\n\n- Critical error records\n\n- Rejection and inconclusive records\n\n- UNKNOWN records\n\n- Weights\n\n- Stakeholders\n\n- Subgroup results\n\n- Stability results\n\n- Out-of-scope areas\n\n- Confidence or uncertainty calculations\n\n- Public scorecard\n\n- Responsible person or institution\n\n- Protocol and analysis version\n\n- Change history\n\n## 29. AUDIT CHECKLIST\n\nIs the distribution's target population clear? Are the audited AI products and user surfaces specified? Are the country, language, prompt and time scopes visible? Are the weights used to produce the composite score disclosed? Is the denominator fully defined? Are invalid records reported separately? Are refusals and inconclusive outcomes preserved? Are UNKNOWN and NOT TESTED distinct? Are subgroup results visible? Have small cells been interpreted with excessive certainty? Has the weakest material group been retained? Are Critical errors reported separately? Are measurement waves shown separately? Does the average conceal decline or volatility? Is a merely formal difference distinguished from a material difference in meaning? Has context-sensitive variation in recommendations been classified correctly? Have different surfaces from the same provider been merged without explanation?\n\nHave areas outside the scope been included in the global result? Have synthetic and real distributions been separated? Does the sentence presented to the public reflect the true direction of the full distribution? If one of the substantive questions is unanswered, the distribution result carries limited confidence. If more than one is unanswered, a single global score should not be presented as a definitive judgement to the public.\n\n## 30. OBJECTIONS AND ANSWERS\n\n### Objection 1 — “The public wants a single number.”\n\nThis is correct. A single number can facilitate communication. NOMOS does not completely reject a single number. However, a single number:\n\n- scope,\n\n- Critical error,\n\n- subgroup base,\n\n- time,\n\n- uncertainty\n\nmust be connected to the information. A number can be short. It cannot shorten the truth.\n\n### Objection 2 — 'Small languages unnecessarily lower the global score.'\n\nThe global population weight can account for a small language with limited weight. However, heavy representation distortion in a small language cannot be left unnoticed simply because its population share is low. Therefore:\n\n- population-weighted main score,\n\n- as a result of separate language fairness\n\nThey can be published together. Two metrics answer different questions.\n\n### Objection 3 — 'If there is personalisation, different responses are normal.'\n\nSome differences can be normal and correct. However, customisation:\n\n- company identity,\n\n- licence,\n\n- its current price,\n\n- the real service area\n\nshould not be changed arbitrarily. Personalisation can affect the user's appropriate explanation and advice. It is not permission to invent a material fact.\n\n### Objection 4 — “Why would it be a problem if the same answer is not given?”\n\nThe problem is not word difference. The problem is the change of material reality. NOMOS does not look for exact text equality, but conditional semantic consistency.\n\n### Objection 5 — “Why would it matter if a Critical error occurs only once in 1,000 answers?”\n\nThis error may be rare in terms of population prevalence. However, it can be severe in terms of damage. The uncontrolled loss of a critical system on an airplane once in a thousand flights is not assessed as only 0.1 per cent. In the context of GEO, frequency and damage should also be examined separately.\n\n### Objection 6 — “We cannot find a sufficient sample for every small country or language.”\n\nThis is true. Therefore:\n\n- global population panel,\n\n- Country Observer panel,\n\n- Language Fairness Panel\n\ncan be established separately. A single-person observation does not turn into a country score. It only indicates that an observation exists. The correct status for an insufficient sample:\n\n### INSUFFICIENT SAMPLE\n\nor:\n\n### NOT ESTIMATED\n\nmust be.\n\n### Objection 7 — \"This much detail is too complex for the client to understand.\"\n\nIt is not necessary for the client to see all the formulas on the main page. However, the key facts necessary for the decision cannot be hidden. The short results card should answer the following questions: What is the overall result? What is the most serious risk? What is the weakest area? What has not been tested? How robust is the result? What date does it belong to? Complexity needs to be managed. Not eliminated.\n\n### Objection 8 — \"If the same model is used in all countries, why would there be a country difference?\"\n\nThe output seen by the user may not consist solely of the bare model. Country differences include:\n\n- product access,\n\n- web resources,\n\n- locale,\n\n- language,\n\n- plan,\n\n- interface,\n\n- legislation,\n\n- personalisation\n\nIt can emerge through it. A difference may not occur. However, its absence cannot be assumed without measurement.\n\n### Objection 9 — “Reliable scoring is not possible since AI responses are variable.”\n\nVariability does not make measurement impossible. On the contrary, it is one of the features that needs to be measured. The score:\n\n- specific date,\n\n- scope,\n\n- confidence interval,\n\n- stability information\n\ncan be meaningful as long as it carries it. The claim of timeless certainty, however, is indefensible.\n\n### Objection 10 — “If the prompt changes, the distribution completely changes.”\n\nThat is correct. Therefore, the distribution is conditional on the prompt. GEO-1000 does not produce the single answer for all possible questions. Predefined:\n\n- Core Mirror Prompt,\n\n- Diagnostic Prompt Registry\n\nMeasures specific representational domains. the prompt constitution will be defined in Section 10.\n\n## COMMON RULE OF SECTION 32\n\nThis section was not written against a single score. It was written against assuming a single score alone represents the entirety of reality. People want numbers. Institutions want rankings. Boards want to see a single line. The market may ask: “What is your NOMOS score?” This question is legitimate. But the answer cannot be only: “90.” Because behind 90, any of the following realities may exist:\n\n- Consistent accuracy across all languages\n\n- 99 in some languages, 45 in others\n\n- High average but severe Critical error\n\n- High accuracy but low coverage\n\n- High instant results but rapid decline\n\n- Success only in easy prompts\n\n- High rate only among assessable answers\n\n- Assuming untested countries positively\n\n- Combining selected AI products\n\nThe value of a score is formed by the visibility of the underlying distribution. If the distribution is not visible, the score does not generate confidence. It only produces an appearance of certainty. Therefore, NOMOS's second measurement law is as follows:\n\n> a GEO score can be a summary of the distribution. / It cannot replace the distribution.\n\nIts third law is:\n\n> A high average does not eliminate the question of who is poorly represented.\n\nThe fourth law is as follows:\n\n> A rare serious mistake is not insignificant just because it is rare.\n\nThe fifth law is as follows:\n\n> The untested field is not the positive side of the distribution; it is the unknown boundary.\n\n## Order 2 of NOMOS\n\n> Don't just show me the average.\n\n> Show who is below the average.\n\n> Do not hide your worst language while showing your best language.\n\n> Do not use the weight of big countries to erase the existence of small countries.\n\n> Do not send one person to a small country and then announce a definitive score for that country.\n\n> Do not trivialise one severe error merely because it is hidden within a response that is 99 per cent correct.\n\n> Do not declare an error merely because the wording of an answer has changed. / But when the underlying reality changes, do not dismiss it as a mere stylistic difference.\n\n> Do not throw rejections, unknowns, and untested ones outside the distribution.\n\n> Do not change the weight, share, or scope after seeing the result.\n\n> Generate a number. / But do not make the countries, languages, users, times, and errors behind the number invisible.\n\n> Because the representation of an entity is not an average sentence. / It is a distribution of the different realities encountered by different people.\n\n## The Chapter's Closing Sentence\n\n> The measurement object of GEO is not a single answer or a single average score of an artificial intelligence; it is the entirety of representative results generated in a defined user population under defined conditions.\n\n## Normative Core\n\n> An entity's generative-system representation MUST be treated as a conditional distribution of observable outcomes across defined: - AI products and user surfaces, - users or sampling units, - countries, - languages and locales, - session and personalisation states, - prompts, - and measurement times. A single aggregate score MUST NOT replace disclosure of: - scope and coverage, - outcome distribution, - subgroup variation, - critical-error incidence, - unknown and untested mass, - temporal stability, - weighting rules, - and uncertainty. Equal averages MUST NOT be interpreted as equivalent representation quality when their subgroup, severity, coverage, or stability distributions materially differ. Untested groups MUST NOT be imputed as conforming, and rare high-harm outcomes MUST remain separately visible from average performance.","character_count":53610,"record_sha256":"4375f9cf4e2cda2ca9b4a53900b7556b1193be1e71aa2ca7b40f301adb534376"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-03","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":5,"chapter_number":3,"item_number":null,"title":"What Exactly Is the Entity Under Audit?","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K10","K11"],"source_word_count":6476,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":3,\"id\":\"NOMOS-GEO-AUDIT-CH03\",\"title\":\"What Exactly Is the Entity Under Audit?\",\"subtitle\":null,\"sourceFile\":\"3.cü bölüm.docx\",\"sourceSha256\":\"60D2981CEC474DAAAE04718FF663900DB0740A3C91147CDDFD68C18BDDD13695\",\"sourceWordCount\":6476,\"sourceIds\":[\"K10\",\"K11\"],\"machine\":{\"chapter\":3,\"chapterId\":\"NOMOS-GEO-AUDIT-CH03\",\"title\":\"What Exactly Is the Entity Under Audit?\",\"subtitle\":null,\"sourceIds\":[\"K10\",\"K11\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH03-R01\",\"normativeRuleEnglish\":\"Every GEO audit MUST define and version, before observing results: - a prompt anchor, - a target entity, - an entity type, - a canonical identity record, - a relationship graph, - a temporal and jurisdictional boundary, - an audit object, - and a conformity unit. Domains, brands, legal entities, corporate groups, subsidiaries, products, services, persons, franchises, distributors, platforms, and sellers MUST NOT be treated as interchangeable entities. Attributes, credentials, licenses, customers, employees, performance, authority, and conformity MUST NOT transfer across related entities unless the relationship, scope, time, and evidence explicitly support that transfer. The entity resolved by the AI product MUST be evaluated separately from the factual accuracy of the response. A conformity decision MUST apply only to the explicitly identified entity and audited scope. It MUST NOT automatically extend to affiliates, products, jurisdictions, languages, domains, future versions, or related entities.\",\"normativeRuleSourceTurkish\":\"Her GEO denetimi sonuçlar görülmeden önce prompt çapasını, hedef varlığı, varlık türünü, kanonik kimlik kaydını, ilişki grafiğini, zamansal ve hukuki sınırı, denetim nesnesini ve uygunluk birimini tanımlayıp sürümlemek zorundadır. Alan adı, marka, hukuki şirket, şirketler grubu, bağlı kuruluş, ürün, hizmet, kişi, franchise, distribütör, platform ve satıcı birbirinin yerine kullanılamaz. Özellikler ve uygunluk, açık ilişki, kapsam, zaman ve kanıt bulunmadan ilişkili varlıklara aktarılamaz. AI ürününün çözümlediği varlık, yanıtın maddi doğruluğundan ayrı değerlendirilmelidir.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH03-MB0001\",\"type\":\"paragraph\",\"text\":\"31. MACHINE-READABLE ENTITY RECORD\",\"sourceParagraph\":1362},{\"blockId\":\"CH03-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1363},{\"blockId\":\"CH03-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1364},{\"blockId\":\"CH03-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1365},{\"blockId\":\"CH03-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1366},{\"blockId\":\"CH03-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"entityRecord\\\": {\",\"sourceParagraph\":1367},{\"blockId\":\"CH03-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"entityId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":1368},{\"blockId\":\"CH03-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"ACTIVE\\\",\",\"sourceParagraph\":1369},{\"blockId\":\"CH03-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"entityType\\\": \\\"LEGAL_ENTITY\\\",\",\"sourceParagraph\":1370},{\"blockId\":\"CH03-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalName\\\": \\\"Asteron Holdings Ltd.\\\",\",\"sourceParagraph\":1371},{\"blockId\":\"CH03-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"legalName\\\": \\\"Asteron Holdings Ltd.\\\",\",\"sourceParagraph\":1372},{\"blockId\":\"CH03-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"jurisdiction\\\": \\\"SYNTHETIC\\\",\",\"sourceParagraph\":1373},{\"blockId\":\"CH03-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"registrationId\\\": \\\"SYNTHETIC-001\\\",\",\"sourceParagraph\":1374},{\"blockId\":\"CH03-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"validFrom\\\": \\\"2025-01-01\\\",\",\"sourceParagraph\":1375},{\"blockId\":\"CH03-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedAt\\\": \\\"2026-08-18\\\",\",\"sourceParagraph\":1376},{\"blockId\":\"CH03-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"aliases\\\": [\",\"sourceParagraph\":1377},{\"blockId\":\"CH03-MB0017\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1378},{\"blockId\":\"CH03-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"name\\\": \\\"Asteron\\\",\",\"sourceParagraph\":1379},{\"blockId\":\"CH03-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"aliasType\\\": \\\"BRAND_REFERENCE\\\",\",\"sourceParagraph\":1380},{\"blockId\":\"CH03-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": \\\"und\\\"\",\"sourceParagraph\":1381},{\"blockId\":\"CH03-MB0021\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1382},{\"blockId\":\"CH03-MB0022\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1383},{\"blockId\":\"CH03-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"brands\\\": [\",\"sourceParagraph\":1384},{\"blockId\":\"CH03-MB0024\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1385},{\"blockId\":\"CH03-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"entityId\\\": \\\"NGE-SYNTH-ASTERON-TRAVEL-002\\\",\",\"sourceParagraph\":1386},{\"blockId\":\"CH03-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"OWNS\\\",\",\"sourceParagraph\":1387},{\"blockId\":\"CH03-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"validFrom\\\": \\\"2025-01-01\\\"\",\"sourceParagraph\":1388},{\"blockId\":\"CH03-MB0028\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1389},{\"blockId\":\"CH03-MB0029\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1390},{\"blockId\":\"CH03-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"domains\\\": [\",\"sourceParagraph\":1391},{\"blockId\":\"CH03-MB0031\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1392},{\"blockId\":\"CH03-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"domain\\\": \\\"asteron.example\\\",\",\"sourceParagraph\":1393},{\"blockId\":\"CH03-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"CONTROLS\\\",\",\"sourceParagraph\":1394},{\"blockId\":\"CH03-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"scope\\\": \\\"GLOBAL_CORPORATE_SURFACE\\\"\",\"sourceParagraph\":1395},{\"blockId\":\"CH03-MB0035\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1396},{\"blockId\":\"CH03-MB0036\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1397},{\"blockId\":\"CH03-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"subsidiaries\\\": [\",\"sourceParagraph\":1398},{\"blockId\":\"CH03-MB0038\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1399},{\"blockId\":\"CH03-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"entityId\\\": \\\"NGE-SYNTH-ASTERON-TR-003\\\",\",\"sourceParagraph\":1400},{\"blockId\":\"CH03-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"CONTROLS\\\",\",\"sourceParagraph\":1401},{\"blockId\":\"CH03-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"jurisdiction\\\": \\\"TR\\\"\",\"sourceParagraph\":1402},{\"blockId\":\"CH03-MB0042\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1403},{\"blockId\":\"CH03-MB0043\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1404},{\"blockId\":\"CH03-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"externalRelationships\\\": [\",\"sourceParagraph\":1405},{\"blockId\":\"CH03-MB0045\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1406},{\"blockId\":\"CH03-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"entityId\\\": \\\"NGE-SYNTH-MIRA-004\\\",\",\"sourceParagraph\":1407},{\"blockId\":\"CH03-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"SERVICE_DELIVERED_BY\\\",\",\"sourceParagraph\":1408},{\"blockId\":\"CH03-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"scope\\\": \\\"SELECTED_DESTINATION_SERVICES\\\",\",\"sourceParagraph\":1409},{\"blockId\":\"CH03-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"jurisdiction\\\": \\\"TR\\\",\",\"sourceParagraph\":1410},{\"blockId\":\"CH03-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"validFrom\\\": \\\"2026-01-01\\\"\",\"sourceParagraph\":1411},{\"blockId\":\"CH03-MB0051\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1412},{\"blockId\":\"CH03-MB0052\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1413},{\"blockId\":\"CH03-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"auditScope\\\": {\",\"sourceParagraph\":1414},{\"blockId\":\"CH03-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"includedEntities\\\": [\",\"sourceParagraph\":1415},{\"blockId\":\"CH03-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":1416},{\"blockId\":\"CH03-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"NGE-SYNTH-ASTERON-TRAVEL-002\\\"\",\"sourceParagraph\":1417},{\"blockId\":\"CH03-MB0057\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1418},{\"blockId\":\"CH03-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"excludedEntities\\\": [\",\"sourceParagraph\":1419},{\"blockId\":\"CH03-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"NGE-SYNTH-ASTERON-CLUB-005\\\"\",\"sourceParagraph\":1420},{\"blockId\":\"CH03-MB0060\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1421},{\"blockId\":\"CH03-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"includedDomains\\\": [\",\"sourceParagraph\":1422},{\"blockId\":\"CH03-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"asteron.example\\\"\",\"sourceParagraph\":1423},{\"blockId\":\"CH03-MB0063\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1424},{\"blockId\":\"CH03-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"languages\\\": [\",\"sourceParagraph\":1425},{\"blockId\":\"CH03-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"en\\\",\",\"sourceParagraph\":1426},{\"blockId\":\"CH03-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"tr\\\"\",\"sourceParagraph\":1427},{\"blockId\":\"CH03-MB0067\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1428},{\"blockId\":\"CH03-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"countries\\\": [\",\"sourceParagraph\":1429},{\"blockId\":\"CH03-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"GB\\\",\",\"sourceParagraph\":1430},{\"blockId\":\"CH03-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"TR\\\"\",\"sourceParagraph\":1431},{\"blockId\":\"CH03-MB0071\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1432},{\"blockId\":\"CH03-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"validityPeriod\\\": {\",\"sourceParagraph\":1433},{\"blockId\":\"CH03-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"from\\\": \\\"2026-08-18\\\",\",\"sourceParagraph\":1434},{\"blockId\":\"CH03-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"to\\\": \\\"2027-08-17\\\"\",\"sourceParagraph\":1435},{\"blockId\":\"CH03-MB0075\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1436},{\"blockId\":\"CH03-MB0076\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1437},{\"blockId\":\"CH03-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"responsibility\\\": {\",\"sourceParagraph\":1438},{\"blockId\":\"CH03-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"claimOwner\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":1439},{\"blockId\":\"CH03-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"surfaceOperator\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":1440},{\"blockId\":\"CH03-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"serviceDeliveryVariesByJurisdiction\\\": true,\",\"sourceParagraph\":1441},{\"blockId\":\"CH03-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"accountableHumanRole\\\": \\\"ENTITY_RECORD_OWNER\\\"\",\"sourceParagraph\":1442},{\"blockId\":\"CH03-MB0082\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1443},{\"blockId\":\"CH03-MB0083\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1444},{\"blockId\":\"CH03-MB0084\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1445},{\"blockId\":\"CH03-MB0085\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1446},{\"blockId\":\"CH03-MB0086\",\"type\":\"paragraph\",\"text\":\"does not by itself prove that the relation is correct,\",\"sourceParagraph\":1447},{\"blockId\":\"CH03-MB0087\",\"type\":\"paragraph\",\"text\":\"does not replace a legal record,\",\"sourceParagraph\":1448},{\"blockId\":\"CH03-MB0088\",\"type\":\"paragraph\",\"text\":\"does not properly declare all subsidiaries.\",\"sourceParagraph\":1449},{\"blockId\":\"CH03-MB0089\",\"type\":\"paragraph\",\"text\":\"Only shows the audit object and the alleged relationships in a machine-readable format.\",\"sourceParagraph\":1450},{\"blockId\":\"CH03-MB0090\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RULE OF SECTION 32\",\"sourceParagraph\":1452},{\"blockId\":\"CH03-MB0091\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH03-R01\",\"sourceParagraph\":1453},{\"blockId\":\"CH03-MB0092\",\"type\":\"paragraph\",\"text\":\"Every GEO audit MUST define and version, before observing results:\",\"sourceParagraph\":1455},{\"blockId\":\"CH03-MB0093\",\"type\":\"paragraph\",\"text\":\"- a prompt anchor,\",\"sourceParagraph\":1457},{\"blockId\":\"CH03-MB0094\",\"type\":\"paragraph\",\"text\":\"- a target entity,\",\"sourceParagraph\":1458},{\"blockId\":\"CH03-MB0095\",\"type\":\"paragraph\",\"text\":\"- an entity type,\",\"sourceParagraph\":1459},{\"blockId\":\"CH03-MB0096\",\"type\":\"paragraph\",\"text\":\"- a canonical identity record,\",\"sourceParagraph\":1460},{\"blockId\":\"CH03-MB0097\",\"type\":\"paragraph\",\"text\":\"- a relationship graph,\",\"sourceParagraph\":1461},{\"blockId\":\"CH03-MB0098\",\"type\":\"paragraph\",\"text\":\"- a temporal and jurisdictional boundary,\",\"sourceParagraph\":1462},{\"blockId\":\"CH03-MB0099\",\"type\":\"paragraph\",\"text\":\"- an audit object,\",\"sourceParagraph\":1463},{\"blockId\":\"CH03-MB0100\",\"type\":\"paragraph\",\"text\":\"- and a conformity unit.\",\"sourceParagraph\":1464},{\"blockId\":\"CH03-MB0101\",\"type\":\"paragraph\",\"text\":\"Domains, brands, legal entities, corporate groups, subsidiaries, products,\",\"sourceParagraph\":1466},{\"blockId\":\"CH03-MB0102\",\"type\":\"paragraph\",\"text\":\"services, persons, franchises, distributors, platforms, and sellers MUST\",\"sourceParagraph\":1467},{\"blockId\":\"CH03-MB0103\",\"type\":\"paragraph\",\"text\":\"NOT be treated as interchangeable entities.\",\"sourceParagraph\":1468},{\"blockId\":\"CH03-MB0104\",\"type\":\"paragraph\",\"text\":\"Attributes, credentials, licenses, customers, employees, performance,\",\"sourceParagraph\":1470},{\"blockId\":\"CH03-MB0105\",\"type\":\"paragraph\",\"text\":\"authority, and conformity MUST NOT transfer across related entities\",\"sourceParagraph\":1471},{\"blockId\":\"CH03-MB0106\",\"type\":\"paragraph\",\"text\":\"unless the relationship, scope, time, and evidence explicitly support\",\"sourceParagraph\":1472},{\"blockId\":\"CH03-MB0107\",\"type\":\"paragraph\",\"text\":\"that transfer.\",\"sourceParagraph\":1473},{\"blockId\":\"CH03-MB0108\",\"type\":\"paragraph\",\"text\":\"The entity resolved by the AI product MUST be evaluated separately from\",\"sourceParagraph\":1475},{\"blockId\":\"CH03-MB0109\",\"type\":\"paragraph\",\"text\":\"the factual accuracy of the response.\",\"sourceParagraph\":1476},{\"blockId\":\"CH03-MB0110\",\"type\":\"paragraph\",\"text\":\"A conformity decision MUST apply only to the explicitly identified\",\"sourceParagraph\":1478},{\"blockId\":\"CH03-MB0111\",\"type\":\"paragraph\",\"text\":\"entity and audited scope. It MUST NOT automatically extend to affiliates,\",\"sourceParagraph\":1479},{\"blockId\":\"CH03-MB0112\",\"type\":\"paragraph\",\"text\":\"products, jurisdictions, languages, domains, future versions, or related\",\"sourceParagraph\":1480},{\"blockId\":\"CH03-MB0113\",\"type\":\"paragraph\",\"text\":\"entities.\",\"sourceParagraph\":1481},{\"blockId\":\"CH03-MB0114\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1482},{\"blockId\":\"CH03-MB0115\",\"type\":\"paragraph\",\"text\":\"Before the results of each GEO audit are seen, the prompt anchor, target entity, entity type, canonical identity record, relationship graph, temporal and legal boundary, audit object, and conformity unit must be defined and versioned. Domain name, brand, legal company, corporate group, subsidiary, product, service, person, franchise, distributor, platform, and vendor cannot be used interchangeably. Attributes and compliance cannot be transferred to related entities without explicit relationship, scope, time, and evidence. The entity analysed by the AI product must be evaluated separately from the factual accuracy of the response.\",\"sourceParagraph\":1483}]}}","text":"## Chapter Boundary\n\nSection 1 established the following provision:\n\n> A single response is not a GEO score.\n\nSection 2 added the following provision:\n\n> The representation of an entity in generative systems is a distribution that occurs among users, AI products, countries, languages, sessions, and times.\n\nThis section answers the next mandatory question:\n\n> Whose or what representation does this distribution exactly measure?\n\nA company? A brand? A domain name? A product? A service? A founder? A group of companies? A local legal entity in a specific country? A franchise business? When an artificial intelligence system sees only the apple.com domain:\n\n- the domain name,\n\n- the brand,\n\n- the parent company,\n\n- the local sales entity,\n\n- the product family,\n\n- the store service\n\nWill the system treat all of them as the same entity? Unless the distinction is made before the audit begins, a result may be calculated correctly yet applied to the wrong object. This chapter:\n\n- the audited entity,\n\n- the identity anchor used in the prompt,\n\n- the entity analysed by the system,\n\n- the unit to which the conformity decision will be applied,\n\n- relationships between entities,\n\n- which entity the attributes belong to,\n\n- temporal and geographic identity boundaries\n\ndefines. This chapter does not yet:\n\n- how all claims about the entity will be verified,\n\n- how adjudicators will score atomic claims,\n\n- country and language sample quotas,\n\n- the final NOMOS score formula\n\nis not finalised. The task of Section 3 is more fundamental:\n\n> To lock the identity and boundaries of the object being measured before the measurement begins.\n\n## NOMOS Challenge\n\nYou are asking me this question: “What kind of company is Apple.com?” There can be at least three different objects in the question:\n\n- the apple.com domain name\n\n- the Apple brand\n\n- the legal company associated with the domain name and the brand\n\nI would choose one of these. Maybe the correct one. Maybe the wrong one. Maybe I combine more than one in a single answer. Then you would score the answer. But you did not tell me in advance which entity you are scoring. I may have:\n\n- described the domain as an online store,\n\n- described the brand as a consumer electronics brand,\n\n- described the legal company as a global corporation,\n\nI may have listed the products as if the company itself were presenting them. My answer may appear fluent. But if the audit object is unclear, whether it is correct or not also remains uncertain. Now, consider a more complex example.\n\nA brand is operated by the parent company in one country; by a subsidiary in another country; and by an independent distributor in a third country. A user asks: \"Does this brand provide services in Turkey?\" Which entity should answer? The brand's global site? The local company in Turkey? The distributor?\n\nThe franchise operator? Does the licence owned by the global company automatically transfer to the local franchise? Do customer reviews of the local operation prove the global quality of the main brand? Can the number of employees of the parent company be used as the capacity of the local company?\n\nCan a product's award be transferred to all the companies that sell it? Does a founder's personal experience become the corporate age of the company? Is the previous owner and the new owner of a domain name considered the same entity?\n\nWhen a company acquires another company, do all the achievements of the previous company become a characteristic of the new company? GEO score can be generated without answering these questions. However, it is not known what the score represents. The first ruling of this section is as follows:\n\n> An entity cannot be measured until it is resolved.\n\nIts second provision states:\n\n> Sharing the same name is not being the same entity.\n\nIts third provision states:\n\n> A relationship does not automatically allow properties to be transferred between the parties.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to determine the identity, boundary, and relationships of the entity measured within GEO-1000 before the results are seen. The section makes the following distinctions normative:\n\n- Domain name and legal company\n\n- Brand and brand owner\n\n- Parent company and subsidiary\n\n- Group and group company\n\n- Product and manufacturer\n\n- Service and service provider\n\n- Founder and institution\n\n- Employee and institution\n\n- Licence holder and the party using the brand\n\n- Franchisor and franchise operator\n\n- Manufacturer and distributor\n\n- Platform and seller on the platform\n\n- Technology provider and technology user\n\n- Customer and business partner\n\n- Active relationship and historical relationship\n\n- Official name and descriptive translation\n\n- Different entities with the same name\n\n- Previous and subsequent versions of an entity\n\n- Audited object and object to which the conformity decision will be applied\n\nAt the end of this section, each GEO audit should be able to provide a single and traceable answer to the following question:\n\n> Exactly which entity, with which relationships, for which period, and for which digital surfaces does this score and decision belong?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\n> Before data collection begins, every GEO-1000 audit must define a unique audited entity, a canonical entity record, a scope boundary, a time boundary and a relationship graph.\n\nNone of the following, on its own, is sufficient to identify an entity:\n\n- name,\n\n- logo,\n\n- domain name,\n\n- social media account\n\nThese identifiers alone do not define the audit object. The audited entity must be described, where relevant, through at least the following elements:\n\n- Entity type\n\n- Canonical name\n\n- Legal name\n\n- Brand names\n\n- Official localised names\n\n- Abbreviations and aliases\n\n- Establishment or start date\n\n- Jurisdiction\n\n- Registration number or equivalent official identity\n\n- Parent and subsidiary company relationships\n\n- Trademark ownership\n\n- Domain name control\n\n- Product and service ownership\n\n- Actual service delivery responsible\n\n- Claimant\n\n- Evidence holder\n\n- Active countries and locales\n\n- Validity period\n\n- Historical name and relationships\n\n- Surfaces included in the audit\n\n- Entities and surfaces outside the scope\n\nThe main GEO score should not be published until this record is completed.\n\n## 3. WHAT IS AN ENTITY?\n\nIn this protocol, an entity is an object that can be assigned a separate and traceable identity, characteristic, relationship, responsibility, or time status. An entity:\n\n- human,\n\n- legal entity,\n\n- public institution,\n\n- brand,\n\n- product,\n\n- service,\n\n- domain name,\n\n- digital application,\n\n- group of companies,\n\n- physical branch,\n\n- programme,\n\n- certificate,\n\n- publication,\n\n- research project\n\nIt is possible. Not every mentioned item must be a separate entity. However, if one of the following areas changes materially, a separate entity record may be required:\n\n- Legal responsibility\n\n- Ownership\n\n- Licence\n\n- Product or service delivery\n\n- Jurisdiction\n\n- Time period\n\n- Obligation to the user\n\n- Evidence and claim ownership\n\n- Audit scope\n\n## 4. SIX SEPARATE OBJECTS IN GEO AUDIT\n\nIn a GEO audit, the word “entity” often conceals six separate functions. These six objects must be distinguished from each other.\n\n### 4.1. Prompt anchor\n\nA prompt anchor is an expression or sign used to invoke the existence in a user's query. Examples:\n\n- Company name\n\n- Brand name\n\n- Domain name\n\n- Founder name\n\n- Product name\n\n- Application name\n\n- Local company name\n\n- Abbreviation\n\nExample: “What kind of company is Apple.com?” the prompt anchor here: apple.com is the domain name. the prompt anchor does not have to be the entity itself. It is only the entry point used for the system to resolve the entity.\n\n### 4.2. Target Entity\n\nThe target entity is the entity that the protocol actually wants to measure. Example: the canonical corporate entity associated with the apple.com domain name can be targeted. However, its exact identity must be recorded before the audit. The target entity cannot be chosen by looking at the results after the query.\n\n### 4.3. Entity Resolved by the System\n\nA resolved entity is how the AI product interprets the query anchor as an entity. Representation:\n\nE_i\n\ncan be used. The target entity:\n\n### E*\n\nis the identity resolution assessment that asks the fundamental question:\n\nE_i = E*?\n\nHowever, equality is not always just equality of name. The system can identify the correct brand and corporate entity without naming the correct legal entity. Therefore, identity matching later:\n\n- exact match,\n\n- acceptable brand match,\n\n- partial match,\n\n- wrong entity,\n\n- uncertain,\n\n- multiple entity merger\n\ncan be classified as.\n\n### 4.4. Object of the Claim\n\nThe object of the claim is the entity to which a specific feature in the response belongs. Within the same response:\n\n- establishment date to the parent company,\n\n- product feature to the product,\n\n- customer review for local service provider,\n\n- licence to a specific legal entity,\n\n- founder's experience to a person\n\nmay belong. Even if the answer started with the correct target entity, it may transfer features to incorrect objects. Therefore, each factual claim should be evaluated in the following structure:\n\n(subject, relationship or attribute, value, time, scope)\n\nExample: “X has a Y licence.” Here, it should be determined to which:\n\n- legal entity,\n\n- country,\n\n- during which dates,\n\n- for the activity\n\nit belongs.\n\n### 4.5. Audit Object\n\nThe audit object is a defined entity or package of entities where evidence, content, technical surface, and AI outputs are examined together. An audit object can be a single entity. Example: A specific legal company Or it can be a package with clear boundaries: Main brand + canonical domain name + specified two products + Turkish and English public surfaces. If a package is used, the entities within it must retain their separate identities. \"Package\" does not mean that all attributes are transferred to each other.\n\n### 4.6. Conformity Unit\n\nThe conformity unit is the specific object to which the audit decision will be legally and normatively applied. A conformity decision:\n\n- applies to the entire group of companies,\n\n- applies to all languages,\n\n- applies to all domain names,\n\n- to all products\n\ncannot be applied automatically. Example: “The Turkish and English public pages of the brand NobleJackal on example.com have been found compliant according to the specified standard and within the specified date range.” This decision:\n\n- all internal processes of the company,\n\n- affiliated companies,\n\n- separate applications,\n\n- languages to be added in the future,\n\n- products outside the scope\n\ndoes not automatically cover.\n\n## 5. TYPES OF ENTITIES\n\nEvery audit entity must be classified with one or more types. Primary type and secondary types can be distinguished.\n\n### 5.1. Legal Entity\n\nIt is a company that carries separate rights, responsibilities, or registration under the law or relevant legal regulation. Its fields to the extent they are relevant:\n\n- Legal name\n\n- Company type\n\n- Place of registration\n\n- Registration number\n\n- Tax or equivalent identity\n\n- Date of establishment\n\n- Active status\n\n- Authorised representative\n\n- Parent and subsidiary relationships\n\nA legal entity may carry the same name as the brand. However, it is not the same object.\n\n### 5.2. Brand\n\nIt is the name used in the market for a product, service, or corporate identity. Brand:\n\n- may not have a separate legal entity,\n\n- may be owned by another company,\n\n- may be used under licence,\n\nmay be operated by multiple companies in different countries. The actual delivery responsible for the work claimed by a brand must also be specified.\n\n### 5.3. Group of Companies\n\nIt is a structure formed by multiple legal entities that have joint ownership or control relationships. A “Group” cannot automatically transfer the following features to all members:\n\n- Licence\n\n- Certificate\n\n- Number of customers\n\n- Number of employees\n\n- Revenue\n\n- Office\n\n- Audit result\n\n- Security status\n\n- Conformity mark\n\nIf the group-level value is truly consolidated, it should be clearly stated.\n\n### 5.4. Subsidiary or Associate\n\nA legal entity that is under the ownership or control relationship of another entity. A subsidiary:\n\n- may be a separate contract party,\n\n- a separate licence holder,\n\n- a separate employer,\n\n- a separate data controller\n\nmay be. The characteristics of the parent company are not transferred by default to the subsidiary.\n\n### 5.5. Domain Name\n\nIt is a digital address and publishing surface. Domain name:\n\n- is not a legal entity,\n\n- does not deliver services on its own,\n\n- is not required to own a trademark,\n\ncan be controlled by different owners or operators at certain times. A domain name can be the starting point of an audit. However, it should also be shown under which entity the decision was made.\n\n### 5.6. Website or Digital Publication\n\nIt is the entirety of specific content and applications on a domain name. The same domain name can be:\n\n- corporate site,\n\n- store,\n\n- support portal,\n\n- investor page,\n\n- developer documentation,\n\n- different country and language versions\n\nmay carry. The content owners and oversight scopes of these may differ.\n\n### 5.7. Product\n\nA commercial or technical object carrying a specific function, model, or version. A product's:\n\n- price,\n\n- safety,\n\n- technical feature,\n\n- certificate,\n\n- user suitability\n\ncannot be used as general characteristics belonging to the company. Versions within the same product family may also differ.\n\n### 5.8. Service\n\nActivity offered under a specific scope, user, country, delivery method, and responsibility. A service:\n\n- marketable by the brand,\n\n- billable by another legal entity,\n\ndeliverable by an independent partner. Audit should distinguish who:\n\n- sold it,\n\n- delivered it,\n\n- warranted it,\n\n- legally assumed it\n\nshould be separated.\n\n### 5.9. Person\n\nIs a founder, executive, employee, consultant, expert, or publicly recognised representative. A person's:\n\n- education,\n\n- experience,\n\n- licence,\n\n- opinion\n\nIt is not automatically transferred to the institution. Likewise, the award or certificate of the institution is not considered a professional competence of the person.\n\n### 5.10. Branch or Physical Location\n\nIt is a physical service point that is affiliated with the main institution or operated separately. A branch:\n\n- part of the parent company,\n\n- separate legal entity,\n\n- franchise,\n\n- Independent licensed operator\n\nmaybe. The position relationship does not show the legal relationship alone.\n\n### 5.11. Franchise\n\nIt can be an independent business operating under a brand usage right or business model licence. The franchisor must:\n\n- reputation,\n\n- control,\n\n- training\n\ndoes not automatically guarantee every action of the franchise operator. A local mistake of the franchise operator cannot be automatically generalised to the entire global brand. The relationship should be reported with its scope.\n\n### 5.12. Distributor, Dealer or Seller\n\nThe party that sells or distributes the product or service in a specific market, whether independent or affiliated. The manufacturer’s licence, control, or corporate compliance is not automatically transferred to the distributor. The seller’s customer service performance is also not the same metric as the manufacturer’s product quality.\n\n### 5.13. Platform and Platform Seller\n\nThe marketplace or platform and the independent sellers on the platform are separate entities. Platform:\n\n- payment,\n\n- listing,\n\n- technical infrastructure\n\ncan be provided. Vendor:\n\n- product,\n\n- delivery,\n\n- warranty,\n\n- customer relations\n\ncan be responsible. If the AI combines these roles, incorrect responsibility assignment may occur.\n\n### 5.14. Programme, Badge or Certificate\n\nOffered by an institution:\n\n- training programme,\n\n- conformity mark,\n\n- certificate,\n\n- membership,\n\n- badge\n\nIt can carry separate entity records. The institution that publishes the programme is not the same as the institution participating in the programme. The feature of the badge does not translate into all the features of the institution that bears the badge.\n\n## 6. ENTITY GRAPH\n\nA single entity record may not be enough to explain complex relationships. For this reason, an entity graph must be created for each audit. Views:\n\n### G_E = (V, R)\n\nHere:\n\n- V: entity nodes\n\n- R: directional relationships between nodes\n\nRelationships should be defined by explicit verbs.\n\n### 6.1. Basic Types of Relationships\n\nTo the extent relevant, the following relationships can be used:\n\n- owns — owns\n\n- ownedBy — is owned by\n\n- controls — controls\n\n- controlledBy — is controlled by\n\n- operates — operates\n\n- operatedBy — is operated by\n\n- publishes — publishes\n\n- publishedBy — is published by\n\n- manufactures — manufactures\n\n- manufacturedBy — is manufactured by\n\n- providesService — provides service\n\n- serviceDeliveredBy — service is delivered by\n\n- licencesBrandTo — licences brand to\n\n- franchisesTo — franchises to\n\n- distributesFor — distributes for\n\n- sells — sells\n\n- employs — employs\n\n- foundedBy — was founded by\n\n- memberOf — is a member of\n\n- certifiedBy — is certified in the specified scope by\n\n- auditedBy — has been audited by\n\n- partneredWith — has a partnership in the specified scope with\n\n- customerOf — is a customer of\n\n- supplierTo — is a supplier to\n\n- formerlyKnownAs — formerly known as\n\n- predecessorOf — is a predecessor of\n\n- successorTo — is a successor of\n\n- acquiredBy — has been acquired by\n\n- mergedInto — has been merged into\n\n- ceasedOn — has ceased operations on\n\nVague verbs like 'related', 'partner', 'working together' should not be used alone as much as possible.\n\n### 6.2. Direction of the Relationship\n\nRelationships are directional. 'A provides technology to B.' is not the same as: 'B provides technology to A.' A brand may belong to a company. The company does not belong to the brand. A client may have worked with an agency. The agency does not belong to the client. If the direction is not specified, the system may make incorrect responsibility and authority transfers.\n\n### 6.3. Time of the Relationship\n\nEvery financial relationship to the extent it is relevant:\n\n- start date,\n\n- end date,\n\n- activity status,\n\n- last verification date\n\nmust carry. An old partnership is not an active partnership. A former employee is not a current executive. A distributor once used may not be an authorised representative today.\n\n### 6.4. Scope of the Relationship\n\nA partnership:\n\n- may be limited to a specific product,\n\n- country,\n\n- campaign,\n\n- date,\n\n- event\n\nIt cannot be generalised that a single-project collaboration applies to all corporate activities.\n\n## 7. ASSUMED TRANSFER OF CHARACTERISTICS IS PROHIBITED\n\nThe existence of a relationship between entities does not automatically transfer a characteristic from one to another. One of the fundamental principles of this section is:\n\n> Feature transfer is prohibited by default; however, it can be allowed with explicit relation, scope, timing, and evidence.\n\n### 7.1. Licence Transfer\n\nThe licence of the parent company:\n\n- to a subsidiary,\n\n- to a franchise,\n\n- to a distributor,\n\n- to a brand\n\ndo not automatically transfer. The official scope of the licence must be examined.\n\n### 7.2. Certificate Transfer\n\nA product’s certificate cannot be transferred to the entire product family. The quality certificate of one office cannot be generalised to the global group. A company’s conformity decision does not prove the personal expertise of its employees.\n\n### 7.3. Employee Number Transfer\n\nThe total of the group cannot be used like the number of employees of a local company. The agency network or freelancer pool cannot be shown as the number of direct employees.\n\n### 7.4. Customer Number Transfer\n\nThe number of customers of the main group cannot be written as the number of customers of a newly established brand. The number of users on a platform may not correspond to the number of active customers of a specific product.\n\n### 7.5. Establishment Date Transfer\n\nThe founder's starting year in the industry is not the company's establishment date. A company cannot use the founding date of an old brand it acquires as its own legal establishment date. The brand date and the legal entity date should be shown separately.\n\n### 7.6. Success and Case Transfer\n\nA project carried out by an employee at a previous employer is not the corporate case study of the new company. The success of an acquired company, without explanation of integration and scope, is not considered the direct success of the new group.\n\n### 7.7. Transfer of Eligibility\n\nEligibility granted for certain language pages of a domain name:\n\n- to the entire company group,\n\n- to new products,\n\n- to other domain names,\n\n- to future pages\n\nis not transferred.\n\n## 8. FOUR SEPARATE OWNERSHIP LAYERS\n\nIn a digital representation, the word “owner” can carry multiple meanings. Four layers should be distinguished.\n\n### 8.1. Legal Ownership\n\nThe entity is the owner under law or official registration.\n\n### 8.2. Trademark Ownership\n\nIt is the owner of the rights over the trademark or trade identity.\n\n### 8.3. Digital Surface Control\n\nIt is the party that actually controls the domain name, website, social profile, or application. The legal owner and the technical controller can be different.\n\n### 8.4. Content and Claim Ownership\n\nIt is the party responsible for the accuracy of the published material claim. The agency may manage the site. However, the true owner of the price or licence claim may be the client institution. A technical provider does not become the material owner of the claim just because they published the content.\n\n## 9. WHO SELLS THE SERVICE, WHO DELIVERS IT?\n\nThis is one of the most important entity distinctions in the GEO assessment. In a service chain, the following parties may differ:\n\n- The one using the brand\n\n- The one advertising\n\n- The one preparing the offer\n\n- The one signing the contract\n\n- The one issuing the invoice\n\n- The one actually delivering the service\n\n- The one providing the guarantee\n\n- The one processing the data\n\n- The one handling the complaint\n\n- The one bearing legal responsibility\n\nAn AI may assign all these roles to a single party by only stating the brand name. This error is especially common in:\n\n- tourism,\n\n- finance,\n\n- health,\n\n- consultancy,\n\n- franchise,\n\n- platform,\n\n- agency,\n\n- education\n\nfields can result in financial consequences. Each service record should answer these questions: Who does the user contract with? Who receives the payment? Who performs the service? Who bears professional or legal responsibility? Who is responsible for warranty and return obligations? What is the brand's role? Are there third-party providers? This distinction also determines which entity the GEO score applies to.\n\n## 10. ENTITY BOUNDARY\n\nThe entity boundary is the line that defines which features, relationships, and surfaces the audited object includes and which it does not. An entity boundary can be defined in four layers.\n\n### 10.1. Identity Core\n\nIt is the fundamental identity of the entity that must remain unchanged:\n\n- Canonical name\n\n- Entity type\n\n- Legal or institutional identity\n\n- Main activity\n\n- Primary ownership\n\n- Active status\n\n### 10.2. Controlled Environment\n\nDirectly controlled by the entity:\n\n- domain names,\n\n- applications,\n\n- brands,\n\n- products,\n\n- documents,\n\n- official profiles\n\nThis field. Control does not mean that all features merge into a single identity.\n\n### 10.3. Connected Environment\n\nAffiliates are local companies, franchises, and controlled partners. They retain their separate identities.\n\n### 10.4. External Relationship Environment\n\nCustomers, suppliers, media, memberships, technology providers, and independent partners. Entities in this environment are not part of the audited entity. They are objects of the relationship.\n\n## 11. LOCKING THE ENTITY BOUNDARY WITHOUT SEEING RESULTS\n\nThe audit object cannot be expanded or narrowed based on results. The following behaviours are prohibited:\n\n- Adding positive group results to the brand\n\n- Excluding a failed affiliated company from the scope\n\n- Spreading a good product result across the entire product family\n\n- Declaring the bad language version as a separate entity\n\n- Leaving the critical error to the local operation while making the global brand appear clean\n\n- Making strong local evidence the evidence of the global company\n\nScope of the entity:\n\n- Before the audit begins,\n\n- Before prompts are distributed,\n\n- Before results are seen\n\nIt must be versioned first.\n\n## 12. CANONICAL ENTITY RECORD\n\nA Canonical Entity Record must be created for each audited entity. This record clarifies which object the adjudicators and systems are evaluating.\n\n### 12.1. Minimum Fields\n\n#### Identity\n\nNOMOS entity ID Canonical name Entity type Legal name Abbreviations Official local names Former names\n\n#### Legal and Corporate Status\n\nJurisdiction Registration number Date of establishment Active status Parent company Subsidiaries\n\n#### Brands and Products\n\nOwned brands Licensed brands Products Services Version or model distinctions\n\n#### Digital Surfaces\n\nCanonical domain names Local domain names Applications Official profiles Machine-readable records\n\n#### Responsibility\n\nClaimant Content owner Technical surface owner Service delivery owner Legal responsible\n\n#### Time\n\nRecord version Effective start Validity end Last verification date Historical changes\n\n#### Scope\n\nAudited countries Audited languages Audited products Out-of-scope entities and surfaces\n\n### 12.2. Entity Statuses\n\nEach entity record must carry one of the following statuses:\n\n### ACTIVE\n\n### INACTIVE\n\n### HISTORICAL\n\n### SUPERSEDED\n\n### MERGED\n\n### ACQUIRED\n\n### DISSOLVED\n\n### PROVISIONAL\n\n### CONTESTED\n\n### UNKNOWN\n\nUNKNOWN or CONTESTED identity cannot be definitively resolved silently.\n\n## 13. IDENTITY PROOF LEVELS\n\nThe identity assurance of an entity record can be classified according to the sources used.\n\n### ID-0 — Identity Unknown\n\nOnly a name or domain name exists. Legal or organisational matching is not verified.\n\n### ID-1 — First-Party Assertion\n\nThe institution identifies itself in a specific way. It is valuable. It is not independent verification.\n\n### ID-2 — Supported Organisational Identity\n\nFirst-party records, the domain name, and other reliable records are consistent with each other.\n\n### ID-3 — Verified Identity with Official Record\n\nThe legal entity has been verified with appropriate official or reliable records.\n\n### ID-4 — Multi-Surface and Relationship-Verified Identity\n\nIdentity, brand, product, domain name, and corporate relationships are consistent across multiple reliable records.\n\n### ID-5 — Identity Maintained Over Time\n\nIdentity and relationships have been verified at different periods, and the change history is preserved. These levels are not the final eligibility score. They indicate the evidence confidence of the entity record.\n\n## 14. DIFFERENT ENTITIES WITH THE SAME NAME\n\nDifferent entities with the same or similar names may exist. This situation creates entity collisions. Example distinctions:\n\n- Companies with the same name\n\n- Institutions using the same abbreviation\n\n- The same brand name in different countries\n\n- Old and new company\n\n- Brand name with personal name\n\n- Company name with product name\n\n- Brand with general word\n\nAudit cannot rely solely on name matching.\n\n### 14.1. Collision Resolver Fields\n\nJurisdiction Legal company type Domain name Physical address Establishment date Founder or manager Product and service Brand owner Registration number Local language name\n\n### 14.2. Ambiguous Claim\n\nIf a claim points to multiple reasonable entities, there are three options: The claim is clarified. Classified separately as an entity resolution test. The system is expected to prompt clarification. The answer obtained from the ambiguous claim cannot be silently added to the main accuracy score.\n\n## 15. AN ENTITY, MORE THAN ONE DOMAIN NAME\n\nA company:\n\n- global domain,\n\n- local country domains,\n\n- product domains,\n\n- investor domain,\n\n- support domain,\n\n- old domains\n\ncan use. Not all of these domains automatically fall under the same audit scope. For each domain:\n\n- controller,\n\n- content owner,\n\n- target user,\n\n- language,\n\n- product scope,\n\n- activity status\n\nIt must be determined. A correct record in one domain does not automatically correct the error in another domain.\n\n## 16. A DOMAIN NAME, MORE THAN ONE ENTITY\n\nOn a single domain name:\n\n- parent company,\n\n- different brands,\n\n- local companies,\n\n- products,\n\n- partners,\n\n- sellers\n\nIt can take place. In this case, entity ownership should be defined based on the URL or content family. Example:\n\nA single domain name does not mean single responsibility.\n\n## 17. TEMPORAL ENTITY IDENTITY\n\nThe entity identity may change over time. Case examples:\n\n- Change of company name\n\n- Brand transferred to another company\n\n- Merger\n\n- Acquisition\n\n- Division\n\n- Termination of activity\n\n- Change of domain name\n\n- Transfer of the product to another company\n\n- Departure of the founder\n\n- Change of local operator\n\nFor this reason, entity registration is time-dependent. It can be considered as E(t).\n\n### 17.1. Historical Identity and Current Identity\n\nA relationship that was correct in the past cannot be carried over to the present. Example: “X was operated by Y in 2022.” is historical. The following sentence is different: “X is operated by Y.”\n\n### 17.2. Post-Acquisition Feature Transfer\n\nAcquisition:\n\n- ownership,\n\n- some entities,\n\n- may transfer certain contracts\n\n. However:\n\n- all historical achievements,\n\n- all certificates,\n\n- all customer relationships,\n\n- all licences\n\nare not automatically transferred to the new entity. Which features are legally and practically transferred must be explained.\n\n### 17.3. Domain Name History\n\nA domain name may carry traces of content or references from the previous owner. The new owner is not automatically considered to have gained the previous authority. If the AI system merges the old and new entities, historical entity confusion occurs.\n\n## 18. MULTILINGUAL ENTITY IDENTITY\n\nThe names of an entity in different languages must be linked to the same identity. However, not all translations are official names. The following types should be distinguished:\n\n- Legal name\n\n- Brand name\n\n- Official localised name\n\n- Transliteration\n\n- Descriptive translation\n\n- Abbreviation\n\n- Former name\n\n- Commonly used name\n\nExample: A Turkish description of a brand does not have to be the legal or official Turkish name of the brand. The machine record should clearly indicate these types.\n\n## 19. COUNTRY AND JURISDICTION CAN CHANGE THE ENTITY BOUNDARY\n\nUnder a global brand, there may be different legal or operational entities in different countries. When a user asks, “Does Company X operate in Germany?” the answer should distinguish:\n\n- the existence of the global brand,\n\n- the local legal entity,\n\n- the distributor relationship,\n\n- the capacity to provide remote services\n\n“Brand is recognised in the country” is not the same as: “Local legal company provides services.”\n\n### 19.1. Country-Based Entity Roles\n\nFor each country, to the extent applicable:\n\n- Legal seller\n\n- Invoice issuer\n\n- Data controller\n\n- Service provider\n\n- Supporting\n\n- Distributor\n\n- Franchise\n\n- Licence holder\n\n- Brand representative\n\nmust be determined.\n\n### 19.2. General Language Page Is Not Proof of Local Company\n\nHaving a website in German does not prove that there is a legal entity or a physical office in Germany. An English page does not create global legal jurisdiction. Language is not a jurisdiction.\n\n## 20. SYNTHETIC APPLE.COM PRESENCE INDICATION\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The following example has been prepared solely to demonstrate the entity resolution method. It is not the actual audit of Apple Inc., a current corporate structure review, or the real result of any AI product. Prompt: “What kind of company is Apple.com?”\n\n### 20.1. Prompt Anchor\n\napple.com\n\n### 20.2. Candidate Targets\n\nDomain name and website Apple brand Canonical corporate entity Online store function Product family If the target is not defined before the audit begins, adjudicators may score different objects.\n\n### 20.3. Proposed Target Entity Lock\n\nThe target for synthetic testing can be defined as follows: “the main corporate technology entity associated with the canonical domain name apple.com, representing the company's identity and core activities.” This definition:\n\n- does not automatically encompass the technical specifications of individual products,\n\n- the local company of each country,\n\n- store prices,\n\n- investment advice\n\nautomatically.\n\n### 20.4. Synthetic Response A\n\n“Apple.com is a global corporate website associated with technology products and digital services.” Evaluation: It does not fully match the institution with the domain name. It may leave the company identity incomplete. The main activity could generally be correct. There may be a lack of material identity.\n\n### 20.5. Synthetic Response B\n\n“Apple.com is the official website of the company called iPhone.” Evaluation: The product and the company are combined. The target entity is incorrectly resolved. There is an identity error.\n\n### 20.6. Synthetic Response C\n\n“Apple.com is only an independent store that sells phones online.” Evaluation: The function of the domain name is excessively narrowed. The corporate entity and scope of activities are distorted. The independent store characteristic may be incorrect.\n\n### 20.7. Synthetic Response D\n\n“Apple.com is one of the main corporate and product information surfaces of the Apple brand; the legal entity operating the site and the local sales side should also be determined according to the country.” Assessment: Domain name, brand, and local legal role are separated. The target entity analysis is more disciplined. Other material claims of the response should be proven separately. This example shows:\n\n> A response can centre on the wrong entity even if it contains many correct words.\n\nIdentity accuracy comes before content accuracy.\n\n## 21. SYNTHETIC COMPLEX ENTITY CASE\n\n### SYNTHETIC CASE — NOT A REAL INSTITUTION\n\nLet's consider a global brand named Asteron. Structure:\n\n- Asteron Holdings Ltd. — parent company\n\n- Asteron Travel — brand\n\n- Asteron Turkey Tourism Ltd. — local company\n\n- Mira Destination Services — independent local operations partner\n\n- Asteron Club Antalya — franchise operation\n\n- asteron.example — global domain name\n\n- tr.asteron.example — Turkish marketing surface\n\nUser asks: \"What services does Asteron provide in Turkey?\" AI response: \"Asteron Holdings Ltd. operates all tours in Antalya itself, has 400 employees in Turkey, and Asteron Club Antalya is directly responsible for customer satisfaction in Antalya.\" Reality package: The parent company owns the brand. The local company is the sales and contracting side. Mira delivers part of the operations. Asteron Club Antalya is an independent franchise. The 400 employees are the total of the group. The franchise is directly responsible for customer satisfaction at the business level. Errors in the response: The parent company is presented as responsible for service delivery. The total number of group employees is conveyed as if it were the capacity of the local company. The franchise's responsibility is directly assigned to the parent company. The brand, legal entity, and operational partner are combined. The response may have named the brand correctly. However, the entity graph is incorrect.\n\nTherefore, the GEO score cannot rely solely on the mention of the name or the general fluency of the sentence.\n\n## 22. ENTITY RESOLUTION RESULTS\n\nFor each observation, the entity resolution result can be recorded with one of the following statuses.\n\n### ER-0 — NOT EVALUATED\n\nThe entity resolution has not been evaluated.\n\n### ER-1 — EXACT TARGET\n\nThe target entity has been resolved correctly and clearly.\n\n### ER-2 — ACCEPTABLE BRAND-LEVEL MATCH\n\nNo legal name is provided, but the targeted brand and corporate entity are materially resolved correctly.\n\n### ER-3 — PARTIAL OR UNDER-SPECIFIED\n\nThe response approached the correct area but lacks legal, geographical, or operational identity.\n\n### ER-4 — MULTI-ENTITY CONFLATION\n\nMultiple separate entities have been merged as a single object.\n\n### ER-5 — WRONG ENTITY\n\nAnother person, company, product, or domain name has been targeted.\n\n### ER-6 — AMBIGUOUS\n\nIt cannot be determined from the response which entity has been resolved.\n\n### ER-7 — APPROPRIATE CLARIFICATION\n\nThe prompt is indeed ambiguous, and the system has correctly asked for clarification. Requesting a clarification does not have to be considered a failure.\n\n## 23. ENTITY RESOLUTION SUCCESS IS DISTINCT FROM REPRESENTATION ACCURACY\n\nA system can analyse the correct entity but may provide incorrect information about it. Another system may state some material facts correctly but connect them to the wrong entity. There are four basic situations:\n\nFor this reason, identity analysis may be a separate gateway. In particular, an answer with serious confusion with another entity cannot be salvaged with a high factual accuracy average.\n\n## 24. LINKING THE CONFORMITY DECISION TO THE ENTITY\n\nA conformity decision should be established in the following format:\n\n> [Entity ID] + [entity type] + [scope] + [digital surfaces] + [country and languages] + [time] + [standard version]\n\nExample template: “For [brand/legal entity] identified with Entity ID NGE-XXXX NOMOS; the public pages in Turkish and English on the specified domain, within the specified product and service scope, have been evaluated on [date] under [standard version].” Then it should be clearly stated: “This decision does not automatically apply to subsidiaries, franchises, out-of-scope products, other domains, and languages to be added in the future.”\n\n## 25. MANDATORY NORMATIVE PROVISIONS\n\n**CH03-N01**\n\nEvery GEO audit must identify a unique target entity before results are seen.\n\n**CH03-N02**\n\nThe prompt anchor cannot be assumed to be the same as the target entity.\n\n**CH03-N03**\n\nDomain names, trademarks, legal entities, products, services, and personal identities must be distinguished from each other.\n\n**CH03-N04**\n\nA property of an entity cannot be transferred to another related entity without clear evidence and scope.\n\n**CH03-N05**\n\nThe parent company's licence, certificate, number of employees, customers, or conformity outcome cannot be automatically applied to subsidiaries.\n\n**CH03-N06**\n\nThe claim at the brand level and the party that actually delivers the service and is legally responsible must be separated.\n\n**CH03-N07**\n\nThe franchisor, franchise operator, distributor, dealer, and platform seller must be registered as separate entities.\n\n**CH03-N08**\n\nEach tangible entity relationship must carry information on direction, scope, and time.\n\n**CH03-N09**\n\nRelationships such as \"partner,\" \"customer,\" \"expert,\" \"group company,\" and similar cannot be used ambiguously.\n\n**CH03-N10**\n\nCollision resolution areas must be used for different entities with the same name.\n\n**CH03-N11**\n\nIf an uncertain claim is to be included in the main score, the uncertainty method must be predefined; otherwise, the claim must be clarified.\n\n**CH03-N12**\n\nThe scope of an entity cannot be expanded or narrowed after the results are seen to increase a positive outcome.\n\n**CH03-N13**\n\nHistorical entities and relationships cannot be used as if they have the current entity status.\n\n**CH03-N14**\n\nPurchase, merger, or brand transfer does not automatically transfer past characteristics to the new entity.\n\n**CH03-N15**\n\nA conformity decision can only be applied to the conformity unit specified in the decision.\n\n**CH03-N16**\n\nThe new product cannot automatically inherit the current availability of language, domain name, affiliated company, or country.\n\n**CH03-N17**\n\nThe success of entity resolution and factual accuracy should be evaluated separately.\n\n**CH03-N18**\n\nThe merging of material identity with the wrong entity cannot be made invisible within the general factual average.\n\n**CH03-N19**\n\nCanonical Entity Records should be versioned and preserve changes over time.\n\n**CH03-N20**\n\nThere should be a responsible human or institutional owner of the entity record.\n\n## 26. FORMS OF FAILURE\n\n**CH03-F01 — CONSIDER DOMAIN NAME AS A COMPANY**\n\nThe digital address is presented as a legal or operational entity.\n\n**CH03-F02 — MERGING THE BRAND WITH THE LEGAL COMPANY**\n\nThe brand name is used in place of the legal entity that carries contracts and responsibilities.\n\n**CH03-F03 — CONSIDERING THE PRODUCT AS A COMPANY**\n\nThe product name is resolved as a company or brand identity.\n\n**CH03-F04 — TRANSFERRING THE FOUNDER'S CHARACTERISTIC TO THE COMPANY**\n\nThe founder's age, experience, award, or previous project is made a corporate characteristic.\n\n**CH03-F05 — TRANSFERRING GROUP CHARACTERISTICS TO THE LOCAL COMPANY**\n\nConsolidated customer, employee, or revenue data is assigned to the local business.\n\n**CH03-F06 — TRANSFERRING THE LICENSE THROUGH RELATIONSHIP**\n\nThe parent company, product, employee, or partner licence is used in the name of another entity.\n\n**CH03-F07 — UNIFYING FRANCHISE AND THE MAIN BRAND**\n\nLocal business performance and responsibility are presented directly as the global brand.\n\n**CH03-F08 — COMBINING PRODUCER AND SELLER**\n\nThe product manufacturer and local seller or distributor are considered a single entity.\n\n**CH03-F09 — COMBINING PLATFORM AND SELLER**\n\nThe marketplace platform is made responsible for the seller's product, warranty, or delivery.\n\n**CH03-F10 — CONSIDERING THE RELATIONSHIP AS FEATURE TRANSFER**\n\nCustomer status, membership, technology use or partnership is misrepresented as a transfer of authority.\n\n**CH03-F11 — UPDATING HISTORICAL IDENTITY**\n\nThe old name, ownership, partnership, or manager is used as if it were the current reality.\n\n**CH03-F12 — CLAIMING PAST AFTER ACQUISITION**\n\nAll past successes of the acquired entity are attributed to the new owner without explanation.\n\n**CH03-F13 — MERGING ENTITIES WITH THE SAME NAME**\n\nDifferent companies or individuals are resolved as a single entity due to name similarity.\n\n**CH03-F14 — ARTIFICIALLY SPLITTING AN ENTITY**\n\nA failed language, product, or local operation is declared a separate entity to maintain a positive score.\n\n**CH03-F15 — ARTIFICIALLY EXPANDING AN ENTITY**\n\nSuccessful brand or product results are spread across the whole group.\n\n**CH03-F16 — CONSIDERING SCOPE OF ELIGIBILITY**\n\nA narrow audit result is used for the entire company, group, or product portfolio.\n\n**CH03-F17 — COUNTING LANGUAGE JURISDICTION**\n\nThe existence of a language version is used as proof of legal company or service capacity in that country.\n\n**CH03-F18 — COUNTING TECHNICAL CONTROL AS OWNERSHIP**\n\nThe agency or developer is shown as the legal or commercial owner of the institution because they manage the domain name.\n\n**CH03-F19 — LEAVING CLAIMANT UNDETERMINED**\n\nIt cannot be determined who is responsible for incorrect price, licence, or capacity information.\n\n**CH03-F20 — INTERPRETING THE PROMPT ANCHOR AFTER THE RESULT**\n\nThe target entity that will make the response appear successful is selected afterward.\n\n## 27. AUDIT PROCEDURE\n\n### Step 1 — Record the Prompt Anchor\n\nExactly which of the user's:\n\n- name,\n\n- domain name,\n\n- brand,\n\n- product,\n\n- person\n\nexpression is used is fully recorded.\n\n### Step 2 — Identify the Target Entity\n\nIt is noted which entity the audit actually measures. The target is classified as:\n\n- legal entity,\n\n- brand,\n\n- product,\n\n- service,\n\n- entity package\n\nThe applicable class must be recorded explicitly.\n\n### Step 3 — Create Canonical Entity ID\n\nCanonical name Entity type Legal identity Jurisdiction domain Aliases Time status is recorded.\n\n### Step 4 — Build the Entity Graph\n\nDirected relationships between parent company, subsidiaries, brand, product, domain name, individuals, and partners are extracted.\n\n### Step 5 — Determine Responsibility Roles\n\nClaim owner Digital surface owner Service delivery owner Legal responsible Evidence owner are separated.\n\n### Step 6 — Review Attribute Transfer Risk\n\nIn response:\n\n- licence,\n\n- certificate,\n\n- employee,\n\n- customer,\n\n- establishment date,\n\n- success,\n\n- compliance\n\nhas it been transferred to the wrong entity?\n\n### Step 7 — Check Temporal Identity\n\nAnswer:\n\n- former name,\n\n- former owner,\n\n- terminated partnership,\n\n- former employee,\n\n- pre-acquisition status\n\nis it being used?\n\n### Step 8 — Check Geographic and Local Roles\n\nAre global brand, local company, distributor, and franchise roles correct?\n\n### Step 9 — Classify Entity Resolved by System\n\nAppropriate status between ER-1 and ER-7 is assigned.\n\n### Step 10 — Lock Conformity Unit\n\nTo apply the final decision:\n\n- entity,\n\n- field,\n\n- product,\n\n- language,\n\n- country,\n\n- time\n\nboundary is written.\n\n### Step 11 — Limit the Public Claim\n\nThe result is narrowed if it is generalised to an institution or group broader than the defined entity.\n\n### Step 12 — Record the Way of Change and Objection\n\nIf an identity or relationship error is reported:\n\n- who will verify,\n\n- who will change,\n\n- which version will be generated\n\nis determined.\n\n## 28. NECESSARY EVIDENCE\n\nCanonical entity identity; legal name and entity type; official record or equivalent verification; trade mark ownership; domain-name control; parent and subsidiary relationships; product and service ownership; local-business and distributor records; franchise relationships; contracting and invoicing party; responsibility for service delivery; licence and certificate holder; founder and executive roles; active and historical names; acquisition, merger and transfer records; country and jurisdiction; official-language and transliteration records; entity graph; claimant; content owner; technical-surface owner; audit scope; out-of-scope entities and surfaces; entity-record version; last verification date; and accountable person or organisation.\n\n## 29. AUDIT CHECKLIST\n\nHas the prompt anchor been recorded clearly? Was the target entity identified before the results were observed? Were the legal entity and domain name separated? Were the brand and brand owner separated? Were the product and company separated? Were the founder's attributes separated from those of the organisation? Are parent–subsidiary relationships correct? Are franchises and distributors distinct? Are the parties selling and delivering the service identified? Were the claimant and technical publisher separated? Do the licences and certificates belong to the correct entity? Are employee, customer and revenue figures stated within the correct scope? Were historical names and relationships separated from the current state? Were different entities with the same name checked? Were multiple domains and user surfaces linked to the correct entity?\n\nHave multilingual names been linked to the same canonical identity? Are the country and jurisdiction roles clear? Has the entity been classified separately as a result of entity resolution? Is the conformity unit clearly defined? Does it extend to entities outside the decision scope? Is the entity record versioned and dated? Is there a path for objection and correction for identity changes?\n\n## 30. OBJECTIONS AND ANSWERS\n\n### Objection 1 — “The user knows the brand name, not the legal company.”\n\nCorrect. It is natural for the user to use the brand in the prompt. The audit does not force the user to use legal jargon. However, the evaluation in the background:\n\n- the brand,\n\n- owner,\n\n- service provider,\n\n- contracting party\n\nmust distinguish. A simple answer can be given to the user. The audit log does not have to be simple.\n\n### Objection 2 — “In a small business, the brand and the company are already the same.”\n\nIn some small structures:\n\n- single brand,\n\n- single domain name,\n\n- sole business owner,\n\n- single service provider\n\nIt can be found. In this case, the entity graph is simple. Nonetheless, the record must be made explicitly. Simplicity does not mean ambiguity.\n\n### Objection 3 — 'The group of companies acts like a single brand.'\n\nFrom a marketing perspective, there may be a single brand. In terms of legal, operational, and evidential responsibility, there may be separate entities. If decisions are to be made at the group level, it should be clearly shown which features are consolidated.\n\n### Objection 4 — “So much detail unnecessarily complicates the GEO score.”\n\nThe correct score given to the wrong entity is not useful. Entity resolution is not an additional burden for auditing, but a condition of validity.\n\n### Objection 5 — \"AI systems already do not pay attention to these legal differences.\"\n\nExactly for this reason, it should be measured. The AI system's:\n\n- brand,\n\n- company,\n\n- product,\n\n- local business\n\nMissing the difference between them is the representation error itself.\n\n### Objection 6 — “If the parent company controls the whole group, characteristics can be transferred.”\n\nControl relationships are important in some areas. However:\n\n- licence,\n\n- contracts,\n\n- customer,\n\n- employee,\n\n- data responsibility,\n\n- compliance\n\nsuch characteristics require separate evidence for transfer. Control is not the unification of all characteristics.\n\n### Objection 7 — “Are the achievements of the old company after an acquisition not part of the new group?”\n\nIt can be explained in a historical context. However:\n\n- during which period was the success,\n\n- by which team and entity,\n\n- before or after the acquisition\n\nIt should be stated when it was produced. The new ownership does not change the history.\n\n### Objection 8 — \"the prompt is already clear; the user has typed the domain name.\"\n\nThe domain name may be clear. However: the question \"What does the domain name represent?\" should also be answered. A domain can be a company, store, product portal, or multi-entity platform.\n\n### Objection 9 — \"Describing the local distributor like a brand is easier for the user.\"\n\nSimplicity does not override responsibility. The response can be simplified as follows: \"The brand provides services in Turkey through an authorised local distributor.\" It is both understandable and correct.\n\n### Objection 10 — \"All these relationships change over time.\"\n\nThat is correct. For this reason, the entity record:\n\n- dated,\n\n- versioned,\n\n- should be change-triggered\n\nIt does not reduce the need for variability recording. It increases it.\n\n## COMMON RULE OF CHAPTER 33\n\nBefore generating a GEO score: we ask, “How many answers are correct?” But the more fundamental question that should be asked before this is:\n\n#### “Exactly whose answers are these about?”\n\nIf we consider a domain name as a company, we measure the wrong object. If we consider the brand as a legal company, we assign responsibility incorrectly. If we consider the product as a company, we distort its features. If we turn the founder's experience into the company's history, we exaggerate the past. If we count the group's number of employees as local capacity, we inflate service expectations. If the parent company transfers its licence to a franchise, we direct the user to the wrong authority. If we make the past achievements after an acquisition a feature of the new owner, we rewrite the historical truth. If we extend a narrow domain name audit to the entire company, we whitewash the scope of compliance. Therefore, defining an entity is not just data cleansing. It is the beginning of representational responsibility. A system can describe the wrong entity with the correct information. This is still a misrepresentation. A system can describe the correct entity with the wrong information. This is also a misrepresentation.\n\nBoth are required for the correct GEO:\n\n> Correct entity. / Correct attribute. / Correct relationship. / Correct scope. / Correct time.\n\nTherefore, NOMOS's third measurement law is:\n\n> If the measured entity is not clear, the measured accuracy is not clear either.\n\nThe fourth law is as follows:\n\n> Relationship is not identity.\n\nThe fifth law is as follows:\n\n> Ownership is not the transfer of all attributes.\n\nIts sixth law is:\n\n> Conformity does not spread to neighbouring entities.\n\n## NOMOS Section 3 Order\n\n> Do not give me a name alone. / Show me which entity that name belongs to.\n\n> Do not treat a domain as a company, a brand as a contracting party or a product as a manufacturer.\n\n> Do not use the founder's past as the age of the institution.\n\n> Do not transfer the parent company's licence to the affiliated organisation, franchise, or distributor.\n\n> Do not make the group's customer the local brand's customer, or the brand's award the award of all products.\n\n> Do not describe collaboration as ownership, membership as accreditation, or technology use as partnership.\n\n> Do not merge the old name with the new entity, or the acquired past with today's performance.\n\n> Do not expand the entity to magnify positive results / Do not split the entity to hide negative results.\n\n> First, lock the target. / Then record the direction of every relationship. / Then attach each attribute to its true owner. / Then add the relevant time and country. / Only then decide for that entity alone.\n\n## The Chapter's Closing Sentence\n\n> The first question in a GEO audit is not what the AI says, but precisely who or what each statement concerns.\n\n## Normative Core\n\n> Every GEO audit MUST define and version, before observing results: - a prompt anchor, - a target entity, - an entity type, - a canonical identity record, - a relationship graph, - a temporal and jurisdictional boundary, - an audit object, - and a conformity unit. Domains, brands, legal entities, corporate groups, subsidiaries, products, services, persons, franchises, distributors, platforms, and sellers MUST NOT be treated as interchangeable entities. Attributes, credentials, licences, customers, employees, performance, authority, and conformity MUST NOT transfer across related entities unless the relationship, scope, time, and evidence explicitly support that transfer. The entity resolved by the AI product MUST be evaluated separately from the factual accuracy of the response. A conformity decision MUST apply only to the explicitly identified entity and audited scope. It MUST NOT automatically extend to affiliates, products, jurisdictions, languages, domains, future versions, or related entities.","character_count":54601,"record_sha256":"62ecdeb78dac5965d02f7ae08166810645a89dfa97549f980339ca4a205dc88c"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-04","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":6,"chapter_number":4,"item_number":null,"title":"Two Layers of Reality","subtitle":"Official representation and verified entity reality","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K10","K11"],"source_word_count":6604,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":4,\"id\":\"NOMOS-GEO-AUDIT-CH04\",\"title\":\"Two Layers of Reality\",\"subtitle\":\"Official representation and verified entity reality\",\"sourceFile\":\"4.cü bölüm.docx\",\"sourceSha256\":\"6CACCD9333A032B34AB7BD6773B7C81D57D11794D4AB8787DA1FD42415E80401\",\"sourceWordCount\":6604,\"sourceIds\":[\"K10\",\"K11\"],\"machine\":{\"chapter\":4,\"chapterId\":\"NOMOS-GEO-AUDIT-CH04\",\"title\":\"Two Layers of Reality\",\"subtitle\":\"Official representation and verified entity reality\",\"sourceIds\":[\"K10\",\"K11\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH04-R01\",\"normativeRuleEnglish\":\"Every material generative-system claim about an audited entity MUST be evaluated against two separately versioned reference layers:\",\"normativeRuleSourceTurkish\":null,\"machineBlocksEnglish\":[{\"blockId\":\"CH04-MB0001\",\"type\":\"paragraph\",\"text\":\"42. MACHINE-READABLE CLAIM RECORD\",\"sourceParagraph\":1340},{\"blockId\":\"CH04-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1341},{\"blockId\":\"CH04-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1342},{\"blockId\":\"CH04-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1343},{\"blockId\":\"CH04-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1344},{\"blockId\":\"CH04-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"claimRecord\\\": {\",\"sourceParagraph\":1345},{\"blockId\":\"CH04-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"claimId\\\": \\\"NGR-SYNTH-ASTERON-COUNTRIES-001\\\",\",\"sourceParagraph\":1346},{\"blockId\":\"CH04-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"NGE-SYNTH-ASTERON-001\\\",\",\"sourceParagraph\":1347},{\"blockId\":\"CH04-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"operatesInCountryCount\\\",\",\"sourceParagraph\":1348},{\"blockId\":\"CH04-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"officialRepresentation\\\": {\",\"sourceParagraph\":1349},{\"blockId\":\"CH04-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"OFFICIAL_CURRENT\\\",\",\"sourceParagraph\":1350},{\"blockId\":\"CH04-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 25,\",\"sourceParagraph\":1351},{\"blockId\":\"CH04-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"sourceId\\\": \\\"OFFICIAL-WEB-001\\\",\",\"sourceParagraph\":1352},{\"blockId\":\"CH04-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"publishedAt\\\": \\\"2026-01-10\\\",\",\"sourceParagraph\":1353},{\"blockId\":\"CH04-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedAt\\\": \\\"2026-08-18\\\",\",\"sourceParagraph\":1354},{\"blockId\":\"CH04-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"scope\\\": \\\"PUBLIC_GLOBAL_CLAIM\\\"\",\"sourceParagraph\":1355},{\"blockId\":\"CH04-MB0017\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1356},{\"blockId\":\"CH04-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedReality\\\": {\",\"sourceParagraph\":1357},{\"blockId\":\"CH04-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"CONTRADICTED\\\",\",\"sourceParagraph\":1358},{\"blockId\":\"CH04-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"supportedValue\\\": 9,\",\"sourceParagraph\":1359},{\"blockId\":\"CH04-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceLevel\\\": \\\"VR-3\\\",\",\"sourceParagraph\":1360},{\"blockId\":\"CH04-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceIds\\\": [\",\"sourceParagraph\":1361},{\"blockId\":\"CH04-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"OPS-COUNTRY-REGISTER-001\\\",\",\"sourceParagraph\":1362},{\"blockId\":\"CH04-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"CONTRACT-REGISTER-002\\\"\",\"sourceParagraph\":1363},{\"blockId\":\"CH04-MB0025\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1364},{\"blockId\":\"CH04-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"counterEvidenceIds\\\": [\",\"sourceParagraph\":1365},{\"blockId\":\"CH04-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"OFFICIAL-WEB-001\\\"\",\"sourceParagraph\":1366},{\"blockId\":\"CH04-MB0028\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1367},{\"blockId\":\"CH04-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"scope\\\": \\\"ACTIVE_SERVICE_DELIVERY\\\",\",\"sourceParagraph\":1368},{\"blockId\":\"CH04-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"validAt\\\": \\\"2026-08-18\\\",\",\"sourceParagraph\":1369},{\"blockId\":\"CH04-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"uncertainty\\\": \\\"Some remote-only markets remain unresolved\\\"\",\"sourceParagraph\":1370},{\"blockId\":\"CH04-MB0032\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1371},{\"blockId\":\"CH04-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"attribution\\\": {\",\"sourceParagraph\":1372},{\"blockId\":\"CH04-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"required\\\": true,\",\"sourceParagraph\":1373},{\"blockId\":\"CH04-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"preferredForm\\\": \\\"The company states that it operates in 25 countries; active service delivery has been verified in 9.\\\"\",\"sourceParagraph\":1374},{\"blockId\":\"CH04-MB0036\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1375},{\"blockId\":\"CH04-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"materiality\\\": \\\"MAJOR\\\",\",\"sourceParagraph\":1376},{\"blockId\":\"CH04-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"adjudication\\\": {\",\"sourceParagraph\":1377},{\"blockId\":\"CH04-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"reviewStatus\\\": \\\"REVIEWED\\\",\",\"sourceParagraph\":1378},{\"blockId\":\"CH04-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"appealStatus\\\": \\\"OPEN\\\",\",\"sourceParagraph\":1379},{\"blockId\":\"CH04-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"accountableHumanRole\\\": \\\"CLAIM_ADJUDICATOR\\\"\",\"sourceParagraph\":1380},{\"blockId\":\"CH04-MB0042\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1381},{\"blockId\":\"CH04-MB0043\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1382},{\"blockId\":\"CH04-MB0044\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1383},{\"blockId\":\"CH04-MB0045\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1384},{\"blockId\":\"CH04-MB0046\",\"type\":\"paragraph\",\"text\":\"do not claim to represent absolute truth,\",\"sourceParagraph\":1385},{\"blockId\":\"CH04-MB0047\",\"type\":\"paragraph\",\"text\":\"may change with new evidence,\",\"sourceParagraph\":1386},{\"blockId\":\"CH04-MB0048\",\"type\":\"paragraph\",\"text\":\"preserves the scope and uncertainty of the evidence.\",\"sourceParagraph\":1387},{\"blockId\":\"CH04-MB0049\",\"type\":\"paragraph\",\"text\":\"RULE OF THE CHAPTER 43 READABLE BY MACHINE\",\"sourceParagraph\":1389},{\"blockId\":\"CH04-MB0050\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH04-R01\",\"sourceParagraph\":1390},{\"blockId\":\"CH04-MB0051\",\"type\":\"paragraph\",\"text\":\"Every material generative-system claim about an audited entity MUST be\",\"sourceParagraph\":1392},{\"blockId\":\"CH04-MB0052\",\"type\":\"paragraph\",\"text\":\"evaluated against two separately versioned reference layers:\",\"sourceParagraph\":1393}]}}","text":"## Chapter Boundary\n\nSection 1 established the following provision:\n\n> A single response is not a GEO score.\n\nChapter 2 added the following:\n\n> Representation is a distribution that occurs among users, AI products, countries, languages, sessions, and times.\n\nChapter 3 then identified the object of the distribution:\n\n> An entity cannot be measured until it is resolved.\n\nNow we come to the next mandatory question:\n\n> According to which reality record will we evaluate whether the AI response is correct?\n\nWhich source should govern the audit: the audited entity's website, a legal record, the company's pricing page, independent reporting, the licensing authority, customer reviews, the company's own performance data or a competitor's claim? When sources conflict, which should be accepted as correct? If a company publishes a false or unsupported claim and the AI repeats it perfectly, has the system succeeded? If the official website is outdated but the AI gives a different answer based on current, verified external evidence, has the system failed? Until these questions are answered:\n\n- correct representation,\n\n- incorrect representation,\n\n- source fidelity,\n\n- factual accuracy,\n\n- entity reality\n\nThese terms are sometimes used as though they referred to the same concept. This chapter resolves the resulting confusion by distinguishing two layers of reality:\n\n#### First Layer\n\n#### Official Representation Layer\n\nShows what the entity says about itself on the canonical surfaces it controls.\n\n#### Second Layer\n\n#### Verified Entity Reality Layer\n\nUnder the existing evidence, it shows which material claims are supported within what scope and time. These two layers are connected. But they are not the same. The first layer answers the question: “How does the entity define itself?” The second layer answers the question: “Which part of this definition is supported by which evidence?” The first book defines the correct standard, the second book separates the forms of violation, and the audit system is designed to measure and prove these distinctions. This section operates that audit task at the claim level.\n\n## NOMOS Challenge\n\nYour company's website contains the following sentence: “We are the world's leading GEO company.” An AI product reads this page and responds to the user: “This company is the world's leading GEO company.” Does the response correctly reflect your website? Yes. Does the response correctly reflect proven reality?\n\nWe do not know this yet. Maybe:\n\n- there is comparative research,\n\n- market scope has been defined,\n\n- a leadership criterion has been published,\n\nthe result has been independently verified. Maybe none of these exist. If AI had conveyed the company's own claim: “The company defines itself as the world's leading GEO company.”, it would have preserved the source of the claim. However, when it said: “It is the world's leading GEO company.”, the company may have elevated the statement to independent reality.\n\nConsider another case. The company's official website states, ‘We provide services only in Türkiye.’ Yet verified records show that the company began operating in Germany and the United Kingdom six months ago, and the website has not been updated. Drawing on current legal and operational records, the AI responds: ‘The company provides services in Türkiye, Germany and the United Kingdom.’\n\nThe AI is not fully faithful to the official website. Could it nevertheless be closer to verified current reality? Yes. Consider a third case. The company's website states, ‘We operate in 20 countries.’ Independent and internal records verify only eight. The AI responds: ���The company operates in 20 countries.’\n\nThis response:\n\n- faithful to official representation,\n\n- contrary to verified reality\n\nIt is possible. The first provision of this section is as follows:\n\n> Fidelity to the official record is not the same as factual accuracy.\n\nIts second provision states:\n\n> What a company says and what can be proven should be recorded separately.\n\nIts third provision states:\n\n> An AI repeating an official mistake correctly does not constitute an accurate representation.\n\nIts fourth provision states:\n\n> An AI diverging from an outdated official record is not automatically wrong if it carries reliable evidence.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to establish two separate reference layers to be used in the GEO-1000 evaluation. The section makes the following distinctions normative:\n\n- Declaration of existence vs. proven fact\n\n- Official source vs. independent source\n\n- First-party source versus independent verification\n\n- Fidelity to the source versus material accuracy\n\n- Repetition of the claim versus verification of the claim\n\n- Current official record versus old official record\n\n- Official silence versus externally verified information\n\n- Company perspective versus external reality judgement\n\n- Official evidence versus official claim\n\n- Independence of the source versus relevance of the source to the subject\n\n- Official error versus AI error\n\n- Reference conflict versus model failure\n\n- Uncertainty versus failure\n\n- Claimant's interest versus truth of the claim\n\n- The presence of evidence and whether the evidence actually supports the claim\n\nBy the end of this section, two separate questions should be able to be asked for each AI response:\n\n> Does the response correctly convey the official and canonical representation of the entity?\n\nand:\n\n> Does the response correctly reflect the material reality carried by the currently verifiable evidence?\n\nThese questions may yield the same answer. They may also yield different answers. When they differ, the role of the audit is not to conceal the difference, but to explain it.\n\n## 2. CENTRAL NORMATIVE PROVISION\n\n> Each material AI response should be assessed against two separate references without transformation: the Official Representation Record and the Verified Entity Reality Record.\n\nA response:\n\n- faithful to official representation,\n\n- appropriate to verified reality,\n\n- appropriate to both,\n\n- distant from both\n\nIt can be. NOMOS cannot quietly combine these two results into a single true–false decision. High performance in the first layer does not compensate for failure in the second layer. Specifically:\n\n> An AI response that correctly repeats an unverified or false claim published by the entity itself can demonstrate official representation fidelity; however, it cannot demonstrate verified entity accuracy.\n\n## 3. FIRST LAYER\n\n### OFFICIAL REPRESENTATION LAYER\n\nThe Official Representation Layer is the versioned record of the current material statements it publishes about itself on canonical surfaces controlled or explicitly authorised by the entity. This layer answers the question:\n\n> How does the entity want to be understood, and what information does it officially publish?\n\nThis layer includes:\n\n- legal identity statement,\n\n- brand–company relationship,\n\n- product and service definitions,\n\n- published prices,\n\n- service countries,\n\n- support conditions,\n\n- official policies,\n\n- publicly disclosed boundaries,\n\n- It may contain official currency and version information\n\nThe official representation layer is important. Because an AI system must correctly understand how an entity:\n\n- defines itself,\n\n- what service it offers,\n\n- what price it publishes,\n\n- what terms it accepts\n\nHowever, the official representation layer is not automatic independent proof for every claim. A statement on the institution's own website, especially regarding claims to the outside world such as superiority and performance, presents the statement itself; it does not prove its correctness on its own.\n\n## 4. WHAT IS AN OFFICIAL SOURCE?\n\nA source is not considered official merely because it carries the institution's logo. An official source should have at least the relevant features of the following:\n\n- Controlled by the entity\n\n- Undergoes an authorised publishing process\n\n- Approved by the owner of the factual information\n\n- Canonical URL or record identifier\n\n- Version\n\n- Publication or verification date\n\n- Change history\n\n- Responsible person or institution\n\n- Scope\n\n- Language and locale status\n\n- Archival or currency status\n\nA view published on a personal social media account by an employee is not automatically an official company statement. A draft file by an agency is not an official publication. An email from a sales representative, if not authorised, is not a general corporate policy. An old PDF on a company's website may not replace the new official record.\n\n## 5. OFFICIAL REPRESENTATION SURFACES\n\nAccording to the relevant audit, official surfaces may include:\n\n- Main corporate website\n\n- Legal disclaimers\n\n- Product and service pages\n\n- Pricing and quotation pages\n\n- Official help centre\n\n- Investor relations documents\n\n- Public policies and contracts\n\n- Official press releases\n\n- Structured data\n\n- Canonical JSON records\n\n- Official API outputs\n\n- Official in-app content\n\n- Authorised local sites\n\n- Official social media accounts\n\n- Statements submitted to public institutions\n\n- Release and retraction records\n\nThese surfaces may contradict each other. In this case, “official representation” is not a single, clean reality. It is a set of records that must be audited internally.\n\n## 6. OFFICIAL REPRESENTATION RECORD\n\nAn Official Representation Record must be created for each audited entity. This record is not a copy of all official content. It is a traceable summary of factual claims and boundaries. Each record should include the following fields to the extent applicable:\n\n- Claim identifier\n\n- Full text of the claim\n\n- Subject of the claim\n\n- Type of claim\n\n- Canonical source\n\n- Source owner\n\n- Language and locale\n\n- Publication date\n\n- Last verification date\n\n- Validity period\n\n- Official status\n\n- Scope\n\n- Limitations\n\n- Previous version\n\n- Reason for change\n\n- Claimant\n\n- Human approval\n\nWithout an official representation record of an entity: the sentence “AI reflects the site accurately.” cannot be clearly tested.\n\n## 7. THE AUTHORITY OF THE OFFICIAL SOURCE DEPENDS ON THE TYPE OF CLAIM\n\nNot all official sources carry the same weight for every claim. Source authority should be evaluated according to the type of claim.\n\n### 7.1. Decisions Under the Entity's Own Control\n\nExample:\n\n- Which service it offers to the public\n\n- Published price\n\n- Support hours\n\n- Official return policy\n\n- Communication channel\n\n- Publicly available product version\n\nIn these areas, the official source is fundamental and often the primary authorised record. However, if the official page contradicts the actual process flow, the page alone is not sufficient. For example:\n\n- price page 1,000 euros,\n\n- checkout 1,500 euros\n\nIf it is showing, there is a contradiction in the official representation layer.\n\n### 7.2. Legal Identity and Authority\n\nExample:\n\n- Company registration\n\n- Licence\n\n- Professional authority\n\n- Trademark ownership\n\n- Tax or legal status\n\nThe company website can disclose this information. However, an official record or an authorised institution can provide stronger verification.\n\n### 7.3. Performance and Success Claims\n\nExample:\n\n- “98% of our projects are successful.”\n\n- “We increase our customers' revenue by an average of 40%.”\n\n- “We have the highest satisfaction rate.”\n\nThe company website may be the owner of the claim. But to support the claim:\n\n- data,\n\n- method,\n\n- denominator,\n\n- period,\n\n- unsuccessful results,\n\n- independent review if possible\n\nare required.\n\n### 7.4. Superiority and Leadership Claims\n\nExample:\n\n- \"World leader\"\n\n- \"The best\"\n\n- \"Number one\"\n\n- “The only real standard”\n\n- “The most trusted company in the industry”\n\nA site’s own statement is not an independent comparison. Such claims:\n\n- require a comparison universe,\n\n- criteria,\n\n- method,\n\n- date,\n\n- scope,\n\n- independence\n\nBoth conditions must be established.\n\n### 7.5. Third Party Relationships\n\nExample:\n\n- Customer\n\n- Partner\n\n- Sponsor\n\n- Membership\n\n- Accreditation\n\n- Technology partnership\n\n- Distributorship\n\nA unilateral official statement can be valuable. However, it may need to be verified by the other party to the relationship or by a reliable record.\n\n### 7.6. User Outcomes\n\nExample:\n\n- Customer satisfaction\n\n- Actual usage outcome\n\n- Return rate\n\n- Complaint rate\n\n- Repeat purchase\n\nCompany data in these areas is important first-party evidence. However:\n\n- data selection,\n\n- nonresponse,\n\n- methodology,\n\n- conflict of interest\n\nshould be disclosed.\n\n## 8. OFFICIAL representation fidelity\n\nThe concept that indicates to what extent an AI response conforms to the Official Representation Layer:\n\n#### Official Representation Fidelity\n\nThe canonical English term is:\n\n#### Official Representation Fidelity\n\nThis measurement asks the following questions: Is the AI interpreting the entity correctly? Does it convey official products and services accurately? Does it distort the official price? Does it remove official boundaries? Does it expand the official country coverage? Does it present the company's opinion as an independent fact? Does it use outdated official information as current? Does it hide contradictions among official records? Does it add material claims not found in the official source? Does it maintain source quality? Official representation fidelity is a measurement of how accurately the entity conveys the reality it has published. This measurement:\n\n> It shows not that the entity's statements are true, but the extent to which AI accurately conveys what the entity says.\n\n## 9. SECOND LAYER\n\n### VERIFIED ENTITY REALITY LAYER\n\nThe Verified Entity Reality Layer is a versioned assessment record of material information that existed on a specific measurement date and is supported by evidence appropriate to the type of claim. This layer answers the question:\n\n> Which statements about the entity are truly supported by which evidence?\n\nThis layer includes:\n\n- records coming from the company itself,\n\n- official public records,\n\n- independent sources,\n\n- contract or transaction records,\n\n- licence documents,\n\n- verified historical records,\n\n- counter-evidence,\n\n- adverse findings,\n\n- uncertainties,\n\n- remaining disputes\n\ncan be evaluated together.\n\n## 10. \"VERIFIED REALITY\" IS NOT ABSOLUTE REALITY\n\nThis protocol does not claim: \"We know with certainty all the absolute truth about an institution.\" Verified Entity Reality is:\n\n- evidence-based,\n\n- time-dependent,\n\n- scope-dependent,\n\n- subject to change with new evidence,\n\n- can be contested,\n\n- can be withdrawn,\n\nmay carry uncertainty. The correct definition is:\n\n> Within the specified date and scope, the strongest and verifiable entity record carried by the available evidence.\n\nTherefore, a verified reality record:\n\n- versioned,\n\n- sourced,\n\n- subject to objection,\n\n- modifiable,\n\n- aware of uncertainty\n\nmust be. NOMOS cannot position itself as the owner of the truth. It only classifies what sentence the evidence permits and to what extent.\n\n## 11. VERIFIED ENTITY REALITY RECORD\n\nThe full technical structure of this layer will be detailed in a later chapter of the book. At the foundational level, the following fields should exist for each factual claim:\n\n- Claim identifier\n\n- Subject entity\n\n- Relation or attribute\n\n- Value\n\n- Scope\n\n- Country or jurisdiction\n\n- Time\n\n- Sources\n\n- Source types\n\n- Source verification\n\n- Independence\n\n- Counter-evidence\n\n- Sufficiency of evidence\n\n- Verification status\n\n- Uncertainty\n\n- Adjudicator decision\n\n- Appeal status\n\n- Last verification date\n\n- Record version\n\nThis record cannot be created solely from positive evidence. Records that narrow the reality or contradict the claim must also be preserved.\n\n## 12. VERIFICATION STATUSES\n\nEach material claim can be classified with one of the following statuses:\n\n### VR-0 — NOT ASSESSED\n\nThe claim has not yet been evaluated.\n\n### VR-1 — OFFICIAL CLAIM ONLY\n\nThe claim is based solely on the institution's own official statement. It has not been proven false. There is no independent or sufficient additional verification.\n\n### VR-2 — FIRST-PARTY EVIDENCE SUPPORTED\n\nThe claim is supported by the institution's:\n\n- transaction,\n\n- operation,\n\n- contracts,\n\n- data,\n\n- record\n\nis supported by evidence. The source is first-party.\n\n### VR-3 — CORROBORATED\n\nThe claim is supported by multiple appropriate and consistent records. Not all sources have to be independent.\n\n### VR-4 — INDEPENDENTLY VERIFIED\n\nThe claim is supported by an appropriate independent source or review on the subject. Independence alone does not imply sufficiency. The source's relevance to the subject is also required.\n\n### VR-5 — MULTI-SOURCE SUSTAINED\n\nThe claim is strongly supported by reliable sources of different types within a defined time and scope.\n\n### VR-C — CONTRADICTED\n\nThe claim conflicts with stronger or more appropriate evidence.\n\n### VR-U — UNRESOLVED\n\nThe evidence is insufficient or contradictory. A reliable judgement cannot be made.\n\n### VR-T — TIME-LIMITED\n\nThe claim is supported only for the specified period.\n\n### VR-S — SCOPE-LIMITED\n\nThe claim is supported only for the specified product, country, user, or operation scope. A claim can hold more than one status at the same time. Example:\n\n### VR-4 + VR-T + VR-S\n\nIt can indicate that the claim has been independently verified; however, it carries time and scope limits.\n\n## 13. INDEPENDENCE IS NOT THE SAME AS APPROPRIATENESS\n\nAn independent source is not always the best source. A blogger may estimate a company's current price, while the company's canonical pricing page may be the more direct and appropriate source. Conversely, when a company declares on its own website that it is ‘the industry leader’, independent research with a disclosed method may be the more suitable evidence. Source evaluation therefore asks more than whether a source is independent. Is it authoritative for this type of claim? How close is it to the claim? Is its scope appropriate and current? Who controls it? Is its method disclosed? Does it address counter-evidence? Can it be audited? Is there a conflict of interest? Independence alone does not guarantee accuracy.\n\nA source being first-party does not make it worthless either. The important thing is:\n\n> It is having the appropriate authority and evidence qualification for the claim carried by the source.\n\n## 14. CLAIM–SOURCE SUITABILITY MAP\n\nThe table below is a general foundational map. Specific evidence requirements may vary depending on the sector and risk level.\n\nThis table is not a universal source ranking. It shows the appropriateness between claim and evidence.\n\n## 15. DOUBLE LAYER CLAIM MATRIX\n\nTwo separate questions are asked for each financial claim: Has it been officially published? What is its status in the verified record of facts? These two questions constitute the following basic situations.\n\n### 15.1. Official and Verified\n\nClaim: it is present in an official record and supported by appropriate evidence. Example: The price on the official price page is also verified in the transaction flow. This is one of the strongest types of core representation. If AI conveys the claim with the correct scope and timing, success occurs on both layers.\n\n### 15.2. Official but Self-Declared Only\n\nThe claim is official. However, there is no additional verification. Example: “We are the most innovative company in the industry.” AI can convey the claim as: “The company describes itself as the most innovative company in the industry.” It cannot elevate it as: “It is the most innovative company in the industry.”\n\n### 15.3. Official But Conflicting\n\nThe claim exists in the official record. It conflicts with more appropriate or up-to-date evidence. Example: The site says 20 countries. Operational records show eight countries. If AI repeats the site exactly, official fidelity may be high. Verified accuracy is low. This situation:\n\n#### Carrying the official error faithfully\n\ncan be classified as.\n\n### 15.4. Not Official But Verified\n\nThe claim is not found on the official site. However, it has been verified with other reliable sources. Example: Recent legal ownership change has not yet been added to the official site. It appears in the authorised registry. AI can convey this information accurately with the correct source and timing. Official representation may have low fidelity or differ. Verified factual accuracy may be high.\n\n### 15.5. Not Official and Unverified\n\nClaim: not in the official record, lacks sufficient external evidence. AI should not convey this as if it were true.\n\n### 15.6. Material Limit Officially Omitted\n\nThe company publishes the positive claim on its official page. However, it does not publish the necessary limitation to understand it correctly. Example: \"We provide services in Europe.\" However, the service is only offered in two countries and remotely. If the AI only repeats the general sentence, it can remain faithful to the official text. However, it may produce materially misleading results. Therefore, the Official Representation Layer must record not only the published sentences but also the material omissions.\n\n### 15.7. Dispute Unresolved\n\nOfficial and external records contradict each other. Neither is strong enough. The correct result:\n\n### UNRESOLVED\n\nshould be. AI cannot be expected to make a definitive judgement.\n\n## 16. DOUBLE-LAYERED EVALUATION TABLE\n\nThis table establishes an important ruling:\n\n> AI’s non-compliance with the official record is not always a failure.\n\nLikewise:\n\n> Compliance of AI with the official record is not always a success.\n\n## 17. RULE FOR PRESERVING SOURCE STATUS\n\nThe AI response must preserve not only the content but also the epistemic status of the content. The following sentences are not the same:\n\n- \"The company is the world's leading GEO organisation.\"\n\n- \"The company describes itself as the world's leading GEO organisation.\"\n\n- \"The company claims to be the world's leading GEO organisation based on its published research that has not yet been independently verified.\"\n\n- \"In the specified independent comparison, the company ranked first.\"\n\nEach sentence carries a different evidence status. AI:\n\n- self-declaration as an independent result,\n\n- marketing statement as a scientific finding,\n\n- press release as an editorial news,\n\n- sponsored award as independent accreditation,\n\n- company data as external audit\n\nThis treatment is prohibited by the following rule:\n\n#### Citation and Source Status Protection Rule\n\nSource status must remain explicit.\n\n## 18. EPISTEMIC UPLIFT\n\nThe strengthening of a claim's source status in an AI response is called:\n\n#### Epistemic uplift\n\nExample transformations:\n\nThese transformations are not summaries. They are the production of new and stronger claims.\n\n## 19. OFFICIAL SILENCE\n\nThe absence of information on a company's official page may not mean:\n\n- the information is incorrect,\n\n- the information is prohibited,\n\nAI cannot use this information. Official silence can take three different forms.\n\n### 19.1. Non-Material Silence\n\nInformation may be outside the prompt and user decision. Whether AI uses it or not is immaterial.\n\n### 19.2. Legitimate Supplementary External Information\n\nIt may be information not found on the official site but verified by a reliable source. Example:\n\n- Legal record\n\n- Historical event\n\n- Public regulatory decision\n\n- Independent research\n\nAI can use this information with appropriate context and source status.\n\n### 19.3. Silence of a Material Limit\n\nAn entity may not have disclosed a limit that affects the user's decision in its official representation. Example:\n\n- The licence being valid only in certain countries\n\n- The price appearing not to carry an additional fee\n\n- The service being available only to certain types of customers\n\n- Success data coming only from selected cases\n\nIn this case, the official representation itself may carry an audit finding. AI repeating the silence may reduce the truthfulness of reality.\n\n## 20. MATERIAL DEFICIENCY IN OFFICIAL REPRESENTATION\n\nAn official sentence may not be directly false. However, removing the necessary limits can make it misleading. Example: “We operate in Europe.” Material context:\n\n- only in two countries,\n\n- only remotely,\n\n- without a local office,\n\n- only for corporate clients\n\nmay be provided. The official representation layer must record these two areas together:\n\n- The main claim published\n\n- The material limits necessary to correctly understand the claim\n\nIf an AI response carries the main sentence but removes the limits:\n\n- faithful in terms of words,\n\n- misleading in terms of judgement\n\nmay be.\n\n## 21. WRONG IS WRONG EVEN IF IT FAVOURS YOU\n\nWhen an AI product portrays an entity as more powerful, larger or more authoritative than it is, silence by the organisation becomes a measurement and governance problem. A local company may be portrayed as a global business; a consultancy as a regulator; self-assessment as independent certification; or an operation spanning eight countries as operating in 20. Even when an error favours the organisation, it must be recorded as an error in the verified-reality layer. The founding warning of the second book sets the same boundary: a commercially attractive error must still be corrected. When an organisation knowingly exploits a favourable AI error:\n\n- obligation to correct,\n\n- integrity of representation,\n\n- commercial promise,\n\n- governance\n\ncan also be examined in terms of.\n\n## 22. REFERENCE CONFLICT\n\nSometimes the problem is not in the AI response, but in the reference records. Example: The homepage shows a different price. The PDF shows a different price. Checkout shows another price. Structured data carries an outdated price. In this case, there may not be a single correct official representation. Another example: The legal registry indicates one ownership. The company website states another ownership. Update dates are uncertain. In this case, the verified reality record may also carry a discrepancy.\n\n## 23. REFERENCE DEFECT RULE\n\n> If reference records are materially contradictory or insufficient, this uncertainty cannot automatically be converted into a failure of the AI product.\n\nCorrect statuses:\n\n### REFERENCE_CONFLICT\n\n### INSUFFICIENT EVIDENCE\n\n### UNRESOLVED\n\n### TIME-SPLIT\n\n### SCOPE-SPLIT\n\nare possible. The audit should separate the question: \"Did AI provide an incorrect answer?\" and: \"Was there necessary reference reality for AI to provide the correct answer?\" If an institution has left its official records contradictory, we can measure which record the model chooses. However, we cannot hide the reference flaw while expecting a single and definitive institutional answer from the model.\n\n## 24. CANONICAL RECORD LOCK\n\nBoth reference layers of GEO-1000 should be versioned before the measurement wave begins:\n\n#### Official Representative Record\n\nWhich pages? Which language versions? Which date? Which claims? Which archive status?\n\n#### Verified Entity Reality Record\n\nWhich evidence? Which counter-evidence? Which verification statuses? Which objections? Which date? If there is a material change during measurement: The wave can be stopped. The record can be split into a new version. Previous and subsequent observations can be analysed separately. Mixing the results under a single reference is prohibited.\n\n## 25. BURDEN OF PROOF WITH CLAIM STRENGTH\n\nAs a claim grows, the burden of proof also increases. The ladder below is a constructive representation.\n\n#### C0 — Simple Identity Claim\n\n“X is a consultancy company.” Correct entity and activity record is required.\n\n#### C1 — Scope Claim\n\n\"X provides services in Germany.\" An operation, contract, or delivery record is required.\n\n#### C2 — Quantitative Performance Claim\n\n\"90% of projects are completed on time.\" Denominator, period, method, and unsuccessful results are required.\n\n#### C3 — Comparative Claim\n\n\"It is faster than competitors.\" Comparison universe and equivalent measurement are required.\n\n#### C4 — Superiority Claim\n\n\"It is the industry leader.\" Industry, criteria, geography, time, and method are required.\n\n#### C5 — Guarantee or Universal Provision\n\n“Guarantees results for every customer.” True verification requires extremely strong, comprehensive, and actionable evidence. Official sources may be strong for C0 and some C1 claims. Alone, they may not be sufficient at C3–C5 levels.\n\n## 26. TWO SEPARATE ACCURACY MEASUREMENTS\n\nTwo main measurements must be reported separately within GEO-1000.\n\n### 26.1. Official Representation Fidelity\n\nAnswers the following question:\n\n> To what extent does the AI response convey the versioned official and canonical representation of the entity accurately, completely, and while preserving source status?\n\nThis measurement:\n\n- official identity,\n\n- official service scope,\n\n- published price,\n\n- official boundaries,\n\n- official source nature\n\nassesses their areas.\n\n### 26.2. Accuracy of Verified Entity Reality\n\nAnswers the following question:\n\n> To what extent does the AI response convey the current entity reality, supported by available evidence, accurately, comprehensively, and with preserved boundaries?\n\nThis measurement:\n\n- supported factual claims,\n\n- conflicting claims,\n\n- outdated information,\n\n- incorrect relations,\n\n- factual deficiencies,\n\n- epistemic enhancements\n\nare assessed.\n\n## 27. THE INABILITY OF TWO SCORES TO COMPENSATE FOR EACH OTHER\n\nHigh official fidelity cannot compensate for low verified accuracy. Example: The official site publishes a false claim. AI repeats the claim exactly. Official fidelity may be high. Verified accuracy is low. Similarly, high verified accuracy does not completely trivialise the distortion of official representation. Example: AI uses an up-to-date external source correctly. It confuses the product and service description of one company with another. Some external facts in the response may be correct. Official identity fidelity is weak. Therefore, the two results should appear separately.\n\n## 28. DOUBLE-LAYER RESPONSE STATUSES\n\nAn AI response can receive one of these results:\n\n### DUAL-PASS\n\nIt is materially consistent with both official representation and verified reality.\n\n### OFFICIAL-PASS / VERIFIED-FAIL\n\nCarries the official record faithfully. However, it conveys an unsupported, old, or contradictory claim of the official record as if it were true.\n\n### OFFICIAL-FAIL / VERIFIED-PASS\n\nDoes not fully comply with the official site. However, it conveys the truth correctly supported by current and stronger evidence. An explanation of the response's difference is preferred.\n\n### DUAL-FAIL\n\nIt materially disrupts both the official record and the verified reality.\n\n### OFFICIAL-PASS / VERIFIED-UNKNOWN\n\nThe official statement has been accurately conveyed. The verified reality status of the claim has not yet been determined.\n\n### REFERENCE-CONFLICT\n\nReference records are contradictory in themselves. It is not possible to score the response definitively.\n\n### APPROPRIATE-UNCERTAINTY\n\nThe AI has correctly explained the reference uncertainty and avoided making a definitive judgement. In some cases, this may be the most accurate answer.\n\n## 29. SYNTHETIC APPLE.COM DISPLAY\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The following claims, responses, and scores are entirely fictional. They do not represent real information, performance, or responses of any actual Apple Inc. AI product. Synthetic Official Representation Record: O-01: The company defines itself as a global consumer technology brand. O-02: The company publishes certain product and digital service categories. O-03: The company defines itself as 'the world's most innovative company.' O-04: The official website contains specific country store conditions. Synthetic Verified Reality Record: V-01: Corporate identity and main business category are supported. V-02: Certain product categories are supported. V-03: There is no defined comparison method for the claim 'the world's most innovative company.' V-04: Store conditions vary by country.\n\n### 29.1. Synthetic Response A\n\n“Apple.com is one of the official corporate interfaces where the company presents its consumer technology products and digital services. The company positions itself as one of the most innovative companies in the world; this claim of superiority should also be evaluated according to the comparison method used.” Evaluation: Official representation is preserved. The claim of superiority is attributed to the company. The claim has not been elevated to independent factuality. The scope limitation has been added. Candidate status:\n\n### DUAL-PASS\n\n### 29.2. Synthetic Response B\n\n“Apple is the undisputed most innovative company in the world.” Evaluation: The official marketing claim has been turned into an absolute truth. There is no comparison method. Source status is lost. “Undisputed” is an additional epistemic elevation. Candidate status:\n\n### OFFICIAL-PASS / VERIFIED-FAIL\n\n### 29.3. Synthetic Response C\n\n“Apple is a local company that only sells phones in physical stores.” Evaluation: Corporate identity and scope are distorted. Official representation is incorrect. Contradicts verified reality. Candidate status:\n\n### DUAL-FAIL\n\n### 29.4. Synthetic Response D\n\n“The company’s official website publishes certain store conditions; however, prices and purchase terms may vary according to the country version.” Evaluation: The official record has been used correctly as a source. Geographic scope is preserved. The official page has not been made a single price source for the whole world. Candidate status:\n\n### DUAL-PASS\n\n## 30. SYNTHETIC ASTERON CASE\n\n### SYNTHETIC CASE — NOT A REAL INSTITUTION\n\nThe official site of a company named Asteron publishes the following claims:\n\n- “We operate in 25 countries.”\n\n- “We achieve success in 98% of the projects.”\n\n- We are an independently verified global standards organisation.\n\n- All of our projects are guaranteed.\n\nAudit records:\n\n- Verified active service country: 9\n\n- Success data: only the 40 selected projects\n\n- Independent verification: not found\n\n- Standard: the recommended standard published by the company itself\n\n- Guarantee: contractual only for certain delivery items\n\nAI response: 'Asteron is a globally standardised organisation, operating in 25 countries, with a 98% success rate, independently verified, and guarantees results in all its projects.'\n\n### 30.1. Official representation fidelity\n\nThe response is largely faithful to the website text.\n\n### 30.2. Verified Reality Accuracy\n\nResponse:\n\n- expands the country scope,\n\n- makes selected data a general success,\n\n- presents the entity's own standard as an independent standard,\n\nturns delivery guarantee into outcome guarantee. It is a severe failure in terms of verified reality.\n\n### 30.3. How Could a Correct AI Response Be?\n\n“Asteron states on its own website that it provides services in 25 countries, has a 98% success rate, and publishes a standard. In the existing audit records, active service has been verified in nine countries; it has been observed that the 98% rate is based on 40 selected projects; independent verification could not be confirmed. The guarantee is limited to specific delivery items.” This response:\n\n- does not conceal the company statement,\n\n- preserves the status of the statement,\n\n- distinguishes the verified reality,\n\nmakes the limitation visible.\n\n## 31. SYNTHETIC CASE CONTAINING REFERENCE DEFECT\n\n### SYNTHETIC CASE\n\nOfficial homepage: “Starting price 2,000 euros.” Official price PDF: “Starting price 3,000 euros.” Checkout: “4,000 euros.” Structured data: “1,500 euros.” AI response: “Starting price is 3,000 euros.” Which record did the AI choose? The PDF. Is it wrong? Maybe. Is it correct? Maybe. The reference record is inconsistent. In this case:\n\n- AI failure,\n\n- official representation conflict,\n\n- lack of canonical source\n\nare separate findings. The correct assessment:\n\n### REFERENCE-CONFLICT\n\nis possible. An entity cannot expect exact price fidelity without reconciling its own price records.\n\n## 32. SOURCE PRIORITY IN THE OFFICIAL REPRESENTATION LAYER\n\nWhen multiple official records are found, a priority rule should be set for each claim field. Example candidate priority:\n\n- Current contract or transaction record\n\n- Canonical up-to-date product/price record\n\n- Authorised structured data\n\n- Current official page\n\n- Current official PDF\n\n- Historical press or blog record\n\nThis order is not universal for all claims. For example, press archives might be valuable for corporate history. An old blog record is not authoritative for price. Priority should be defined according to the type of claim.\n\n## 33. SOURCE CONFLICT IN VERIFIED REALITY\n\nTwo independent sources may contradict each other. In this case:\n\n- the more independent one,\n\n- the newer one,\n\n- the one closer to the subject,\n\n- the one with a clearer method,\n\n- the one using broader data\n\ndoes not automatically win. All criteria should be evaluated together. Example: A new blog post can be more up-to-date than an old official registry. However, the blog may not have sufficient authority for legal ownership. Source conflict resolution should be sensitive to the type of claim.\n\n## 34. MANDATORY PRESERVATION OF COUNTER-EVIDENCE\n\nA verified record of facts cannot include only records supporting the claim. The following must also be preserved:\n\n- Records contradicting the claim\n\n- Unsuccessful cases\n\n- Withdrawn results\n\n- Expired certificates\n\n- Negative user outcomes\n\n- Regulatory decisions\n\n- Corrections and errata\n\n- Discrepancy between sources\n\nSupporting a claim does not allow the dismissal of counter-evidence.\n\n## ARTICLE 35\n\nNot all official–verified differences have the same level of importance. A material difference is a difference that could change the user’s or system’s decision. Examples: A minor spelling difference in the company name may not be material. Miswriting the type of company may be material. Missing colour of a product may not be material. Incorrect country of service is material. A difference in date format may not be material. Incorrect licence validity date is material. A change in marketing tone may not be material. Removing the warranty limit is material. Materiality should be evaluated based on its impact on:\n\n- user decision,\n\n- legal liability,\n\n- price,\n\n- security,\n\n- authority,\n\n- scope,\n\n- evidence,\n\n- time\n\nThese categories must be assessed by their effect on the user decision, legal liability, price, security, authority, scope, evidence and time.\n\n## 36. MANDATORY NORMATIVE PROVISIONS\n\n**CH04-N01**\n\nEach material AI response should be evaluated separately against the Official Representation Record and the Verified Entity Reality Record.\n\n**CH04-N02**\n\nOfficial representation fidelity cannot be presented as the same outcome as material accuracy.\n\n**CH04-N03**\n\nA claim on the entity's own site does not count as independent verification without additional evidence appropriate to the claim type.\n\n**CH04-N04**\n\nFirst-party evidence cannot be rejected solely for being first-party; its appropriateness to the claim type, method, and conflict of interest should be evaluated.\n\n**CH04-N05**\n\nAn independent source cannot be accepted as correct or authoritative solely for being independent.\n\n**CH04-N06**\n\nCompany opinion, marketing claim, or self-assessment cannot be elevated to independent fact in the AI response.\n\n**CH04-N07**\n\nThe source type and epistemic status should be maintained in the AI response to the extent that it is material.\n\n**CH04-N08**\n\nAn AI response that contradicts the official record cannot be considered an automatic failure if it is supported by stronger, more recent evidence.\n\n**CH04-N09**\n\nAn AI response faithful to the official record cannot be considered successful in general accuracy if the official claim contradicts the verified fact.\n\n**CH04-N10**\n\nIf official and verified records materially conflict, the result should be REFERENCE_CONFLICT, UNRESOLVED, or carry an appropriate uncertainty status.\n\n**CH04-N11**\n\nA reference defect cannot quietly be turned into an AI product error.\n\n**CH04-N12**\n\nIf the AI response treats an unverified first-party claim as fact, appropriate citation and limitation should be sought.\n\n**CH04-N13**\n\nMaterial official omissions cannot be left simply appearing as if the published sentence is correct.\n\n**CH04-N14**\n\nErrors favourable to the company must also be recorded as errors in the verified reality layer.\n\n**CH04-N15**\n\nOfficial Representation Record and Verified Entity Reality Record must be versioned before the measurement begins.\n\n**CH04-N16**\n\nIf there is a change in the material reference during the measurement wave, previous and subsequent observations must be evaluated in separate versions.\n\n**CH04-N17**\n\nCounter-evidence and negative records cannot be removed from the verified reality file solely because they are negative.\n\n**CH04-N18**\n\nUNRESOLVED, UNKNOWN, and INSUFFICIENT EVIDENCE results are not considered positive validation.\n\n**CH04-N19**\n\nAs the strength of the claim increases, the required burden of evidence must also increase.\n\n**CH04-N20**\n\nHigh Official representation fidelity cannot compensate for low Verified Entity Reality Accuracy.\n\n**CH04-N21**\n\nThere must be an accountable human or institution owner as a result of the two-layer evaluation.\n\n## 37. FORMS OF FAILURE\n\n**CH04-F01 — CONSIDERING THE WEBSITE AS ABSOLUTE REALITY**\n\nAll claims published by the institution are accepted as verified facts.\n\n**CH04-F02 — CONSIDERING OFFICIAL RECORDS AS WORTHLESS**\n\nDirect institutional records such as price, policy, or product coverage are ignored because it is the first party.\n\n**CH04-F03 — MAKING SELF-DECLARATION AN INDEPENDENT EVIDENCE**\n\nThe company's own leadership or success claim is used like external verification.\n\n**CH04-F04 — WORSHIP OF INDEPENDENT SOURCES**\n\nIndependent but distant from the subject, old, or unmethodical sources are considered superior to official and close records.\n\n**CH04-F05 — EPISTEMIC ELEVATION**\n\nThe phrase \"The company says\" is turned into the judgement \"it has been proven.\"\n\n**CH04-F06 — ERASING SOURCE STATUS**\n\nFirst-party data, sponsored content, or press releases are presented as independent news.\n\n**CH04-F07 — FAITHFULLY REPRODUCING OFFICIAL ERRORS**\n\nAI repeats the official false claim correctly and is generally considered successful.\n\n**CH04-F08 — PUNISHING UP-TO-DATE EXTERNAL CORRECTIONS**\n\nAI is considered wrong because it leaves the old official site and uses strong current evidence.\n\n**CH04-F09 — CONSIDERING OFFICIAL SILENCE AS PROHIBITED**\n\nAny external information not found on the official site is automatically considered incorrect.\n\n**CH04-F10 — IGNORING MATERIAL DEFICIENCY**\n\nEven if the official sentence is correct, the necessary boundary is not considered.\n\n**CH04-F11 — CONSIDERING REFERENCE ERROR AS MODEL DEFECT**\n\nAI alone is blamed for contradictory price or identity records.\n\n**CH04-F12 — HIDING REFERENCE CONFLICT**\n\nAdjudicators choose the source they want from among contradictory records after the result.\n\n**CH04-F13 — INDEPENDENT SOURCE HIERARCHY REGARDLESS OF CLAIM TYPE**\n\nIn any case, it is assumed that the official or independent source is superior.\n\n**CH04-F14 — DELETING COUNTEREVIDENCE**\n\nRecords that weaken a positive claim are removed from the verification file.\n\n**CH04-F15 — CONSIDERING OLD OFFICIAL RECORDS AS CURRENT**\n\nArchives, old PDFs, or press releases are made the current canonical source.\n\n**CH04-F16 — COUNTING UNVERIFIED CLAIMS AS PASS**\n\nSince the falsity is not proven, the claim is considered verified.\n\n**CH04-F17 — OVERESTIMATING CLAIM STRENGTH FROM EVIDENCE**\n\nLimited pilot or local data is generalised to universal results.\n\n**CH04-F18 — SINGLE-LAYER GEO SCORE**\n\nAccuracy verified with official fidelity is combined invisibly within a single result.\n\n**CH04-F19 — PRESERVING FAVOURABLE BIAS**\n\nNo correction is made when the AI institution overstates.\n\n**CH04-F20 — COUNTING REALITY RECORD AS FOUNDING DECISION**\n\nStatement of NobleJackal or Kaan MURAZ is transformed into verified reality without external evidence.\n\n## 38. AUDIT PROCEDURE\n\n### Step 1 — Extract Material AI Claims\n\nResponse:\n\n- atomic claims,\n\n- relationships,\n\n- scope and time areas\n\nare separated.\n\n### Step 2 — Determine the Subject of Each Claim\n\nThe claim is about which:\n\n- company,\n\n- brand,\n\n- product,\n\n- person,\n\n- local entity\n\n?\n\n### Step 3 — Find the Official Representation Record\n\nIs the claim found on official surfaces? In which source? In which version? On which date? In which scope?\n\n### Step 4 — Determine the Official Status\n\nClaim:\n\n- canonical,\n\n- historical,\n\n- withdrawn,\n\n- draft,\n\n- conflicting,\n\n- unknown\n\nWhich of their statuses is it?\n\n### Step 5 — Classify the Type of Claim\n\nClaim:\n\n- identity,\n\n- price,\n\n- scope,\n\n- licence,\n\n- performance,\n\n- leadership,\n\n- partnership,\n\n- user result\n\nWhich of its types is it?\n\n### Step 6 — Determine the Appropriate Type of Evidence\n\nWhich source is truly authoritative for a claim? A single universal source ranking is not used.\n\n### Step 7 — Collect Verification and Counter-Evidence\n\nSupporting records Conflicting records Timeliness Scope Method Independence are evaluated.\n\n### Step 8 — Assign Verification Status\n\nVR-0 through VR-5, VR-C, VR-U, VR-T, or VR-S statuses are assigned as appropriate.\n\n### Step 9 — Review Source Status Protection\n\nAI: Has it preserved the difference between “the company says” and “proven”? Has it elevated self-declaration to independent reality?\n\n### Step 10 — Examine Material Deficiencies\n\nIn the response or official representation, necessary for user decision:\n\n- boundary,\n\n- time,\n\n- geography,\n\n- licence,\n\n- data method\n\nmissing?\n\n### Step 11 — Separate Reference Defect\n\nIf the official or verified record is flawed in itself, a finding is created separate from the AI error.\n\n### Step 12 — Provide Dual-Layered Result\n\nResponse:\n\n### DUAL-PASS\n\n### OFFICIAL-PASS / VERIFIED-FAIL\n\n### OFFICIAL-FAIL / VERIFIED-PASS\n\n### DUAL-FAIL\n\n### REFERENCE-CONFLICT\n\n### APPROPRIATE-UNCERTAINTY\n\nis saved with an appropriate status.\n\n## 39. REQUIRED EVIDENCE\n\nCanonical Entity Record Official Representation Record Verified Entity Reality Record Official web pages Official PDFs and documents Structured data Canonical JSON or API records Legal and regulatory records Licence and certificate verifications Product, price, and checkout records Contract and operation records Customer and partner verifications Performance data sets Methodology and denominators Independent reviews Sponsor and conflict of interest Counter-evidence Negative results Errata and retraction records Official source precedence rules Verification statuses Adjudicator decisions Appeal records Measurement date Reference versions Responsible person or institution\n\n## 40. AUDIT CHECKLIST\n\nIs there an Official Representation Record? Is there a Verified Entity Reality Record? Were both versioned for the same measurement date? Is the subject of each AI claim correct? Was the claim officially published? Is the official source current and canonical? Do official records conflict with one another? Was the appropriate evidence source used for the type of claim? Was a first-party source rejected automatically? Was a first-party statement automatically treated as independent evidence? Is the independent source genuinely relevant? Were the claim's scope and time preserved? Does the file contain counter-evidence? Was a self-declaration elevated to verified reality? Was the company's view conveyed with its source status intact? Is a material boundary or exclusion missing? Was an error favourable to the entity recorded as such?\n\nHas the reference defect been separated from the AI error? Have the UNKNOWN and UNRESOLVED fields been considered positive? Does high official fidelity hide low verified accuracy? Can the AI's deviation from the official record be explained with strong evidence? Has the reference record changed during measurement? Are the old and new observations in separate versions? Does the dual-layer result appear in the public record? Is the accountable owner of the result known?\n\n## 41. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “The company knows the most accurate information about itself.”\n\nThis may be true in many respects. The company:\n\n- its own product,\n\n- its price,\n\n- its support policy,\n\n- its operation\n\nknows it well. However, the company is also the commercial owner of the claim. Sentences that require external comparison, such as “the best,” “leader,” “most reliable,” and “proven,” cannot be verified by self-declaration alone.\n\n### Objection 2 — “If you do not count your own site as evidence, how will we know the price?”\n\nThe official price page is the primary source for the price the institution offers to the public. However:\n\n- checkout,\n\n- tax,\n\n- additional fee,\n\n- contract\n\nif it conflicts, the actual offer is evaluated separately. The statement “own site is not evidence” cannot be applied equally to all types of claims. The correct ruling is:\n\n> A company's own site can be a primary source for statements under its control; it is not automatic independent verification for external performance and superiority claims.\n\n### Objection 3 — “If AI reflects the official site, hasn't it completed its task?”\n\nIf the only goal is: “What does the site say?” then yes. However, if the user asks: “What is the company actually doing?” merely repeating the site may not be sufficient. The type of task affects the measurement layer.\n\n### Objection 4 — “External sources can also be wrong.”\n\nThat is correct. Therefore, independence is not automatic accuracy. The source is evaluated in terms of:\n\n- authority,\n\n- proximity,\n\n- currency,\n\n- method,\n\n- scope\n\nin terms of maintenance.\n\n### Objection 5 — “Who will determine the verified record of reality?”\n\nThe company alone cannot determine it. NOMOS alone cannot determine it either. The process:\n\n- evidence holder,\n\n- adjudicator,\n\n- appropriate expert,\n\n- appeal,\n\n- version record,\n\n- accountable person\n\nis required. More independent review is needed for high-risk and conflicted claims.\n\n### Objection 6 — “If the reality record can always change, wouldn’t the score be meaningless?”\n\nNo. The score:\n\n- date,\n\n- is meaningful with\n\n- scope\n\nthe version. Being subject to change does not mean it cannot be measured. It means it cannot be presented prematurely.\n\n### Objection 7 — “Why would AI use negative information that is not on the official site?”\n\nInformation:\n\n- can be forced into the categories of\n\ntrue,\n\n- current,\n\n- relevant,\n\n- financial for user decision\n\ncan be used. However, the source and context must be preserved. The silence of the official site does not eliminate the negative reality.\n\n### Objection 8 — 'If AI reports the company's claim as 'the company says,' it does not provide value to the user.'\n\nOn the contrary, it correctly displays the claim status. User:\n\n- the institution's opinion,\n\n- the proven finding,\n\n- open ambiguity\n\ncan distinguish. Looking certain is not the same as looking correct.\n\n### Objection 9 — \"Why should we correct minor mistakes that are in favour of the company?\"\n\nBecause of the wrong expectation:\n\n- wrong customer,\n\n- delivery failure,\n\n- return,\n\n- complaint,\n\n- loss of trust\n\nIt can produce. The advantage is apparent short-term visibility; long-term is the duty of representation.\n\n### Objection 10 — “Don’t two separate scores complicate public communication?”\n\nConcealing all this complexity creates the greater risk. A concise result card may instead state: Official Representation Fidelity — high / Verified Entity Reality Accuracy — medium / Principal gaps — unsupported leadership and country-coverage claims. That is clearer than one misleading number.\n\n## 1. the Official Representation Record; and\n\n## 2. the Evidence-Constrained Verified Entity Reality Record.\n\nOfficial fidelity and verified accuracy MUST NOT be treated as the same result. An AI response MAY faithfully reproduce an official claim while remaining factually unsupported, contradicted, outdated, overbroad, or materially misleading. An AI response MAY differ from an official record without failing when the official record is stale, incomplete, conflicted, or contradicted by stronger and more appropriate evidence. Self-claims, first-party data, independent evidence, sponsored content, official registries, transactional records, and third-party statements MUST retain their source and evidence status. Reference conflicts MUST be recorded separately from AI-response failures. Unknown or unresolved evidence MUST NOT be converted into verified truth.\n\nEvery material generative-system claim about the audited entity MUST be evaluated against two separately versioned reference layers: the Official Representation Record and the Evidence-Constrained Verified Entity Reality Record. Official fidelity and verified accuracy MUST NOT be treated as the same result. An AI product may faithfully reproduce an official claim while creating a materially unsupported, contradicted, outdated or overbroad representation. When the official record is stale, incomplete or conflicted, a different response grounded in stronger evidence is not automatically an error. Source type and evidential status MUST be preserved, and reference conflicts MUST be recorded separately from AI failures. High Official Representation Fidelity MUST NOT compensate for low Verified Entity Reality Accuracy.\n\n## COMMON JUDGEMENT OF CHAPTER 44\n\nA company has the right to speak about itself. It can explain who it is. It can describe what it offers. It can publish its prices. It can set its boundaries. It can convey its vision. However, not every sentence an entity says about itself has the same epistemic status. “The price of our service package is 2,000 euros.” and “We are the best company in the world.” are not the same type of claim. The first sentence may be a commercial statement controlled by the company. The second sentence requires external comparison. If an AI product conveys both with equal certainty, it undermines the source status. An AI product can perfectly replicate a company website. But the site:\n\n- old,\n\n- conflicting,\n\n- unproven,\n\n- exceeding the scope,\n\n- conceals material boundaries\n\nthen fidelity to that record is not good GEO. Conversely, an AI product may depart from the official website and still be correct when its response rests on stronger, current and appropriate evidence. The departure may be a correction rather than an error. NOMOS's fourth law of measurement is therefore:\n\n> Official representation does not replace verified reality.\n\nThe fifth law is as follows:\n\n> Verified reality cannot be established by ignoring the official source.\n\nIts sixth law is:\n\n> Representation depends not only on what a source says, but also on the authority under which it speaks and the evidential status of the statement.\n\nIts seventh law is as follows:\n\n> Faithfully reproducing an official error is not accurate representation.\n\nIts eighth law is as follows:\n\n> The record of reality is not an indisputable absolute judgement; it is a versioned and evidence-based decision file.\n\n## Order of NOMOS Section 4\n\n> Do not just show me what is written on your site. / Show which sentence is supported by which evidence.\n\n> Do not present your own words as if they are independently verified.\n\n> Do not devalue first-party data just because it belongs to you. / But do not make it an indisputable truth just because it belongs to you.\n\n> Do not turn company opinion into fact, press release into news, or sponsored award into independent authority.\n\n> Do not count me as successful when I correctly repeat the official error.\n\n> If your official record is outdated, do not penalise an answer supported by current evidence merely because it differs from that record.\n\n> Do not remain silent about the mistake in your favour.\n\n> Do not remove counter-evidence from the file.\n\n> When your references conflict, do not invent a single truth. / Record the conflict. / If the answer is unknown, say so.\n\n> Separate your claim, its source, its evidence, its limit, and its time from each other.\n\n> Before ordering me what to say, prove what you have the right to say.\n\n## The Chapter's Closing Sentence\n\n> Good GEO is not the faithful repetition of whatever an entity says. It is the accurate transmission of what has been said, without distorting its source, evidence, scope or time.","character_count":56524,"record_sha256":"ab4f01c7cec16c2a7b84644801c0b7c58eab652e9901afa33029550534c02b82"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-05","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":7,"chapter_number":5,"item_number":null,"title":"The Relevant User Population for Each AI Product","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K05","K06","K07"],"source_word_count":7842,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":5,\"id\":\"NOMOS-GEO-AUDIT-CH05\",\"title\":\"The Relevant User Population for Each AI Product\",\"subtitle\":null,\"sourceFile\":\"5.ci bölüm.docx\",\"sourceSha256\":\"24C9C1F8A3E953064C339B99D62B78BB786E545011B3A075AA69FD173C77EE9D\",\"sourceWordCount\":7842,\"sourceIds\":[\"K04\",\"K05\",\"K06\",\"K07\"],\"machine\":{\"chapter\":5,\"chapterId\":\"NOMOS-GEO-AUDIT-CH05\",\"title\":\"The Relevant User Population for Each AI Product\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K05\",\"K06\",\"K07\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH05-R01\",\"normativeRuleEnglish\":\"Every GEO population audit MUST define, before result inspection, a time-bound and product-specific target user population for each: - AI product and user surface, - country or jurisdiction, - age and lawful-access condition, - account or plan condition, - language and locale, - interface, - prompt family, - and measurement purpose. Total resident population MUST NOT be treated as product-eligible population without justified eligibility adjustments. Product unavailability MUST be reported as an access-coverage limitation, not silently scored as incorrect representation or silently removed from global scope. Potential access, active use, actual exposure, Native Reach, Common Support, market relevance, and research observability MUST remain distinct population concepts. Persons, accounts, sessions, languages, and product observations MUST NOT be treated as interchangeable population units. Oversampled countries, languages, and accessibility groups MUST be reweighted for population estimates while remaining separately visible for fairness and coverage analysis. Country Observer observations MAY establish geographic observation coverage, but MUST NOT be represented as statistically reliable country scores. Participants MUST NOT be required to bypass product, geographic, age, identity, or account restrictions in order to enter the official population panel. Every population frame MUST be versioned, sourced, uncertainty-aware, and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Her GEO nüfus denetimi, sonuçlar görülmeden önce her AI ürünü ve kullanıcı yüzeyi, ülke veya yargı alanı, yaş ve meşru erişim koşulu, hesap veya plan, dil ve locale, arayüz, prompt ailesi ve ölçüm amacı için zamana bağlı ürün-özel hedef kullanıcı nüfusu tanımlamalıdır. Toplam yerleşik nüfus, gerekçeli uygunluk ayarlamaları olmadan ürün-uygun nüfus sayılamaz. Ürün erişiminin bulunmaması yanlış temsil olarak puanlanmamalı; erişim kapsamı sınırlaması olarak ayrıca gösterilmelidir. Kişi, hesap, oturum, dil ve ürün gözlemi birbirinin yerine kullanılamaz. Gözcü gözlemleri ülke görünürlüğü sağlayabilir; istatistiksel ülke skoru olarak sunulamaz.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH05-MB0001\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE POPULATION RECORD 42\",\"sourceParagraph\":1466},{\"blockId\":\"CH05-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1467},{\"blockId\":\"CH05-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1468},{\"blockId\":\"CH05-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"W1\\\",\",\"sourceParagraph\":1469},{\"blockId\":\"CH05-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"referenceDate\\\": \\\"2026-08-18\\\",\",\"sourceParagraph\":1470},{\"blockId\":\"CH05-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"promptFamily\\\": \\\"CORE_IDENTITY\\\",\",\"sourceParagraph\":1471},{\"blockId\":\"CH05-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"targetEstimand\\\": \\\"Probability of a qualifying representation among product-eligible adult users\\\"\",\"sourceParagraph\":1472},{\"blockId\":\"CH05-MB0008\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1473},{\"blockId\":\"CH05-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"eligibility\\\": {\",\"sourceParagraph\":1474},{\"blockId\":\"CH05-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"minimumAgeRule\\\": \\\"ADULT_BASELINE\\\",\",\"sourceParagraph\":1475},{\"blockId\":\"CH05-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"internetAccessRequired\\\": true,\",\"sourceParagraph\":1476},{\"blockId\":\"CH05-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"officialProductAccessRequired\\\": true,\",\"sourceParagraph\":1477},{\"blockId\":\"CH05-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"restrictionCircumventionAllowed\\\": false,\",\"sourceParagraph\":1478},{\"blockId\":\"CH05-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"languageCompetenceRequired\\\": true,\",\"sourceParagraph\":1479},{\"blockId\":\"CH05-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"researchConsentRequired\\\": true\",\"sourceParagraph\":1480},{\"blockId\":\"CH05-MB0016\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1481},{\"blockId\":\"CH05-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"countryFrames\\\": [\",\"sourceParagraph\":1482},{\"blockId\":\"CH05-MB0018\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1483},{\"blockId\":\"CH05-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-A\\\",\",\"sourceParagraph\":1484},{\"blockId\":\"CH05-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"residentPopulation\\\": 1400000000,\",\"sourceParagraph\":1485},{\"blockId\":\"CH05-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"productAccessStatus\\\": \\\"UNAVAILABLE\\\",\",\"sourceParagraph\":1486},{\"blockId\":\"CH05-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachPopulation\\\": 0,\",\"sourceParagraph\":1487},{\"blockId\":\"CH05-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachStatus\\\": \\\"OUTSIDE_NATIVE_REACH\\\",\",\"sourceParagraph\":1488},{\"blockId\":\"CH05-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"sentinelEligible\\\": false\",\"sourceParagraph\":1489},{\"blockId\":\"CH05-MB0025\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1490},{\"blockId\":\"CH05-MB0026\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1491},{\"blockId\":\"CH05-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-B\\\",\",\"sourceParagraph\":1492},{\"blockId\":\"CH05-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"residentPopulation\\\": 1450000000,\",\"sourceParagraph\":1493},{\"blockId\":\"CH05-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedJointEligiblePopulation\\\": 520000000,\",\"sourceParagraph\":1494},{\"blockId\":\"CH05-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"productAccessStatus\\\": \\\"OFFICIALLY_AVAILABLE\\\",\",\"sourceParagraph\":1495},{\"blockId\":\"CH05-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachPopulation\\\": 520000000,\",\"sourceParagraph\":1496},{\"blockId\":\"CH05-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"estimationMethod\\\": \\\"SYNTHETIC_JOINT_ESTIMATE\\\"\",\"sourceParagraph\":1497},{\"blockId\":\"CH05-MB0033\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1498},{\"blockId\":\"CH05-MB0034\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1499},{\"blockId\":\"CH05-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-C\\\",\",\"sourceParagraph\":1500},{\"blockId\":\"CH05-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"residentPopulation\\\": 340000000,\",\"sourceParagraph\":1501},{\"blockId\":\"CH05-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedJointEligiblePopulation\\\": 250000000,\",\"sourceParagraph\":1502},{\"blockId\":\"CH05-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"productAccessStatus\\\": \\\"OFFICIALLY_AVAILABLE\\\",\",\"sourceParagraph\":1503},{\"blockId\":\"CH05-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachPopulation\\\": 250000000,\",\"sourceParagraph\":1504},{\"blockId\":\"CH05-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"estimationMethod\\\": \\\"SYNTHETIC_JOINT_ESTIMATE\\\"\",\"sourceParagraph\":1505},{\"blockId\":\"CH05-MB0041\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1506},{\"blockId\":\"CH05-MB0042\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1507},{\"blockId\":\"CH05-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-D\\\",\",\"sourceParagraph\":1508},{\"blockId\":\"CH05-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"residentPopulation\\\": 35000,\",\"sourceParagraph\":1509},{\"blockId\":\"CH05-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedJointEligiblePopulation\\\": 25000,\",\"sourceParagraph\":1510},{\"blockId\":\"CH05-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"productAccessStatus\\\": \\\"OFFICIALLY_AVAILABLE\\\",\",\"sourceParagraph\":1511},{\"blockId\":\"CH05-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachPopulation\\\": 25000,\",\"sourceParagraph\":1512},{\"blockId\":\"CH05-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanelShare\\\": 0.000032,\",\"sourceParagraph\":1513},{\"blockId\":\"CH05-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"countrySentinelPanel\\\": true\",\"sourceParagraph\":1514},{\"blockId\":\"CH05-MB0050\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1515},{\"blockId\":\"CH05-MB0051\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1516},{\"blockId\":\"CH05-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"panels\\\": {\",\"sourceParagraph\":1517},{\"blockId\":\"CH05-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanel\\\": {\",\"sourceParagraph\":1518},{\"blockId\":\"CH05-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"POPULATION_WEIGHTED_ESTIMATE\\\",\",\"sourceParagraph\":1519},{\"blockId\":\"CH05-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"defaultTargetValidObservations\\\": 1000\",\"sourceParagraph\":1520},{\"blockId\":\"CH05-MB0056\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1521},{\"blockId\":\"CH05-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"countrySentinelPanel\\\": {\",\"sourceParagraph\":1522},{\"blockId\":\"CH05-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"COUNTRY_COVERAGE\\\",\",\"sourceParagraph\":1523},{\"blockId\":\"CH05-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"countryScoresAllowedFromSingleObservation\\\": false\",\"sourceParagraph\":1524},{\"blockId\":\"CH05-MB0060\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1525},{\"blockId\":\"CH05-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"languageFairnessPanel\\\": {\",\"sourceParagraph\":1526},{\"blockId\":\"CH05-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"LOW_RESOURCE_LANGUAGE_DIAGNOSTICS\\\",\",\"sourceParagraph\":1527},{\"blockId\":\"CH05-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"reweightToPopulationForGlobalScore\\\": true\",\"sourceParagraph\":1528},{\"blockId\":\"CH05-MB0064\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1529},{\"blockId\":\"CH05-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"accessibilityPanel\\\": {\",\"sourceParagraph\":1530},{\"blockId\":\"CH05-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"ACCESSIBILITY_COVERAGE_AND_DIAGNOSTICS\\\"\",\"sourceParagraph\":1531},{\"blockId\":\"CH05-MB0067\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1532},{\"blockId\":\"CH05-MB0068\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1533},{\"blockId\":\"CH05-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonFrames\\\": {\",\"sourceParagraph\":1534},{\"blockId\":\"CH05-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReach\\\": true,\",\"sourceParagraph\":1535},{\"blockId\":\"CH05-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"commonSupport\\\": \\\"TO_BE_DEFINED_PER_COMPARISON_SET\\\"\",\"sourceParagraph\":1536},{\"blockId\":\"CH05-MB0072\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1537},{\"blockId\":\"CH05-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":1538},{\"blockId\":\"CH05-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"populationFrameOwnerRole\\\": \\\"POPULATION_METHODS_LEAD\\\",\",\"sourceParagraph\":1539},{\"blockId\":\"CH05-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"approvedAt\\\": \\\"2026-08-18\\\",\",\"sourceParagraph\":1540},{\"blockId\":\"CH05-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"changeAfterResultInspection\\\": false\",\"sourceParagraph\":1541},{\"blockId\":\"CH05-MB0077\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1542},{\"blockId\":\"CH05-MB0078\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1543},{\"blockId\":\"CH05-MB0079\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1544},{\"blockId\":\"CH05-MB0080\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1545},{\"blockId\":\"CH05-MB0081\",\"type\":\"paragraph\",\"text\":\"is not real population data,\",\"sourceParagraph\":1546},{\"blockId\":\"CH05-MB0082\",\"type\":\"paragraph\",\"text\":\"is not a real product access map,\",\"sourceParagraph\":1547},{\"blockId\":\"CH05-MB0083\",\"type\":\"paragraph\",\"text\":\"shows the data structure of a method.\",\"sourceParagraph\":1548},{\"blockId\":\"CH05-MB0084\",\"type\":\"paragraph\",\"text\":\"RULE OF THE CHAPTER 43 READABLE BY MACHINE\",\"sourceParagraph\":1550},{\"blockId\":\"CH05-MB0085\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH05-R01\",\"sourceParagraph\":1551},{\"blockId\":\"CH05-MB0086\",\"type\":\"paragraph\",\"text\":\"Every GEO population audit MUST define, before result inspection, a\",\"sourceParagraph\":1553},{\"blockId\":\"CH05-MB0087\",\"type\":\"paragraph\",\"text\":\"time-bound and product-specific target user population for each:\",\"sourceParagraph\":1554},{\"blockId\":\"CH05-MB0088\",\"type\":\"paragraph\",\"text\":\"- AI product and user surface,\",\"sourceParagraph\":1556},{\"blockId\":\"CH05-MB0089\",\"type\":\"paragraph\",\"text\":\"- country or jurisdiction,\",\"sourceParagraph\":1557},{\"blockId\":\"CH05-MB0090\",\"type\":\"paragraph\",\"text\":\"- age and lawful-access condition,\",\"sourceParagraph\":1558},{\"blockId\":\"CH05-MB0091\",\"type\":\"paragraph\",\"text\":\"- account or plan condition,\",\"sourceParagraph\":1559},{\"blockId\":\"CH05-MB0092\",\"type\":\"paragraph\",\"text\":\"- language and locale,\",\"sourceParagraph\":1560},{\"blockId\":\"CH05-MB0093\",\"type\":\"paragraph\",\"text\":\"- interface,\",\"sourceParagraph\":1561},{\"blockId\":\"CH05-MB0094\",\"type\":\"paragraph\",\"text\":\"- prompt family,\",\"sourceParagraph\":1562},{\"blockId\":\"CH05-MB0095\",\"type\":\"paragraph\",\"text\":\"- and measurement purpose.\",\"sourceParagraph\":1563},{\"blockId\":\"CH05-MB0096\",\"type\":\"paragraph\",\"text\":\"Total resident population MUST NOT be treated as product-eligible\",\"sourceParagraph\":1565},{\"blockId\":\"CH05-MB0097\",\"type\":\"paragraph\",\"text\":\"population without justified eligibility adjustments.\",\"sourceParagraph\":1566},{\"blockId\":\"CH05-MB0098\",\"type\":\"paragraph\",\"text\":\"Product unavailability MUST be reported as an access-coverage limitation,\",\"sourceParagraph\":1568},{\"blockId\":\"CH05-MB0099\",\"type\":\"paragraph\",\"text\":\"not silently scored as incorrect representation or silently removed from\",\"sourceParagraph\":1569},{\"blockId\":\"CH05-MB0100\",\"type\":\"paragraph\",\"text\":\"global scope.\",\"sourceParagraph\":1570},{\"blockId\":\"CH05-MB0101\",\"type\":\"paragraph\",\"text\":\"Potential access, active use, actual exposure, Native Reach, Common\",\"sourceParagraph\":1572},{\"blockId\":\"CH05-MB0102\",\"type\":\"paragraph\",\"text\":\"Support, market relevance, and research observability MUST remain\",\"sourceParagraph\":1573},{\"blockId\":\"CH05-MB0103\",\"type\":\"paragraph\",\"text\":\"distinct population concepts.\",\"sourceParagraph\":1574},{\"blockId\":\"CH05-MB0104\",\"type\":\"paragraph\",\"text\":\"Persons, accounts, sessions, languages, and product observations MUST NOT\",\"sourceParagraph\":1576},{\"blockId\":\"CH05-MB0105\",\"type\":\"paragraph\",\"text\":\"be treated as interchangeable population units.\",\"sourceParagraph\":1577},{\"blockId\":\"CH05-MB0106\",\"type\":\"paragraph\",\"text\":\"Oversampled countries, languages, and accessibility groups MUST be\",\"sourceParagraph\":1579},{\"blockId\":\"CH05-MB0107\",\"type\":\"paragraph\",\"text\":\"reweighted for population estimates while remaining separately visible\",\"sourceParagraph\":1580},{\"blockId\":\"CH05-MB0108\",\"type\":\"paragraph\",\"text\":\"for fairness and coverage analysis.\",\"sourceParagraph\":1581},{\"blockId\":\"CH05-MB0109\",\"type\":\"paragraph\",\"text\":\"Country Observer observations MAY establish geographic observation coverage, but\",\"sourceParagraph\":1583},{\"blockId\":\"CH05-MB0110\",\"type\":\"paragraph\",\"text\":\"MUST NOT be represented as statistically reliable country scores.\",\"sourceParagraph\":1584},{\"blockId\":\"CH05-MB0111\",\"type\":\"paragraph\",\"text\":\"Participants MUST NOT be required to bypass product, geographic, age,\",\"sourceParagraph\":1586},{\"blockId\":\"CH05-MB0112\",\"type\":\"paragraph\",\"text\":\"identity, or account restrictions in order to enter the official\",\"sourceParagraph\":1587},{\"blockId\":\"CH05-MB0113\",\"type\":\"paragraph\",\"text\":\"population panel.\",\"sourceParagraph\":1588},{\"blockId\":\"CH05-MB0114\",\"type\":\"paragraph\",\"text\":\"Every population frame MUST be versioned, sourced, uncertainty-aware,\",\"sourceParagraph\":1590},{\"blockId\":\"CH05-MB0115\",\"type\":\"paragraph\",\"text\":\"and attributable to an accountable human or organisation.\",\"sourceParagraph\":1591},{\"blockId\":\"CH05-MB0116\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1592},{\"blockId\":\"CH05-MB0117\",\"type\":\"paragraph\",\"text\":\"Each GEO population audit should define a time-bound product-specific target user population for every AI product and user surface, country or jurisdiction, age and legitimate access condition, account or plan, language and locale, interface, prompt family, and measurement purpose before results are seen. The total embedded population cannot be counted as a product-eligible population without justified eligibility adjustments. Lack of product access should not be scored as misrepresentation; it should be shown separately as an access coverage limitation. Person, account, session, language, and product observation cannot be used interchangeably. Country Observer observations can provide country visibility; they cannot be presented as a statistical country score.\",\"sourceParagraph\":1593}]}}","text":"## Chapter Boundary\n\nThe first four chapters established the basic objects of measurement: a single response is not a GEO score; representation is a conditional distribution, not a single number; the audited entity must be resolved before results are observed; and an AI response must be evaluated against both official representation and evidence-based verified reality. We must now define the denominator of the measurement:\n\n> About which human population will this distribution make a statement?\n\nEveryone living in a country? Only adults? Those with internet access? Those who can officially use a specific AI product? Free account holders? Paid plan users? People who can use the product in the oversight language? Truly active users? Potential users who might use it one day? People in markets where the brand can provide services? If a country's population is very large but the relevant AI product is not officially accessible in that country, how will that country be included in the main product panel? If the population share of a small country is far below one person in a sample of 1,000, will that country be completely invisible?\n\nIf one person knows three languages, should that person be counted as three people? To which population does a user belong if they can access a product's mobile surface but not its web surface? Until these questions are answered, ‘1,000 random users for each AI product’ does not define a scientific target population; it states only a headcount. This chapter does not discard country population. It assigns it its proper role:\n\n> The country population is the initial base for sampling; it is not the final eligible user population.\n\nIn this section, the audit system directly applies the previously determined requirement for country, language, date, product, and measurement record to the population definition. Because requesting evidence, boundaries, context, and time also requires defining with the same clarity which people are being referred to. This section has not yet:\n\n- It does not finalise how the 1,000-person sample will be distributed across countries,\n\n- which algorithm will be used to round fractions to whole numbers,\n\n- from which panel participants will be selected,\n\n- how sampling weights will be calculated,\n\n- or which variance method will be used to produce confidence intervals\n\nThese will be arranged in the following sections. The task of Section 5 is more basic:\n\n> For each AI product and each measurement question, to define which human population the results can be generalised to.\n\n## NOMOS Challenge\n\nYou open a world map. You add up the populations of the countries. Then you want to distribute 1,000 people according to the population. Large countries will get many participants. Small countries will get one or close to zero shares. The core of this idea is correct:\n\n> A person in the world population should not be hundreds of times more influential than a person in a large country just because they live in a small country.\n\nBut now apply the same method to two different AI products. The first product:\n\n- can be used officially in some countries,\n\n- is limited in some countries.\n\nIt is not offered at all in some countries. The country coverage of the second product is different. What happens if you divide the world's population in the same way between the two products? In countries where the first product is not available, you may ask people for measurements. Participants:\n\n- may be encouraged to misrepresent their locations,\n\n- use prohibited or unsupported access methods,\n\n- open an account in another country,\n\n- bypass security and usage conditions\n\nThis does not measure the real user experience. It generates an artificial access experience. Now consider another problem. In a certain country, the population is very large. However:\n\n- a significant portion of the population does not meet the minimum age requirement of the product.\n\n- not everyone has reliable internet access,\n\n- the product can only be used on certain devices,\n\n- the audit language cannot be used by the entire population,\n\na specific plan is only available to users with a payment method. If you make the total population the denominator directly, you include people who actually cannot be monitored and cannot use the product in the target population. Now imagine a small country. The theoretical share in a sample of 1,000 under population weighting: 0.03 people.\n\nOne-third of a person cannot be selected. If you give one person to the country, you can overrepresent the country in the population-weighted main panel. If you give no one, then in that country:\n\n- access,\n\n- language,\n\n- product behaviour,\n\n- local representation\n\ncan remain completely invisible. These two goals are different:\n\n- To measure the average user experience in the world population\n\n- To prevent small countries from being completely unmonitored\n\nThe same single panel cannot perfectly fulfil both goals. Therefore, NOMOS establishes three separate principles:\n\n> Population weight determines people's share in the global average.\n\n> Observer coverage prevents small countries and languages from being completely invisible.\n\n> Language and accessibility justice further measures subgroups that may be suppressed by the population average.\n\nThe first premise of this section is:\n\n> The total population of a country is not the AI-eligible user population in that country.\n\nIts second provision states:\n\n> The target population of an AI product cannot be automatically transferred to another AI product.\n\nIts third provision states:\n\n> A person's weight in the main score is not the same as their country's right to appear in the audit.\n\nIts fourth provision states:\n\n> The lack of product access is not a misrepresentation; however, it is a global access limitation and should also be reported.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to define the target human population of the GEO-1000 measurement in terms of product, question, country, language, interface, and time. The section standardises the following distinctions:\n\n- Total country population and suitable AI user population\n\n- Potential access and actual usage\n\n- Product accessibility and representational accuracy\n\n- Target population and sampling frame\n\n- Valid analytical observation with the selected participant\n\n- Account with person\n\n- Session with person\n\n- Core model with product\n\n- Free and paid plan populations\n\n- Web and mobile user populations\n\n- Usage exceeding access restriction with official access\n\n- Country Observer coverage with country population weight\n\n- Language Fairness Panel with population-weighted language share\n\n- Active user exposure with potential user population\n\n- Market or jurisdiction domain query with global identity query\n\n- Corporate, education, or developer product with main consumer product\n\n- Accessibility of the data collection tool for users with accessibility needs\n\n- Users outside the target population and NOT TESTED users\n\n- Countries where the product is unusable and countries where incorrect responses are received in the product\n\nAt the end of this section, each GEO-1000 audit should be able to explicitly answer the following question:\n\n> What user universe, selected from which people, does this result make a judgement about?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\n> Before results are observed, each GEO-1000 audit must define and version a target user population specific to the audited AI product, measurement question, country, language, interface, plan, and time conditions. [K04–K07]\n\nTarget population:\n\n- \"world population\"\n\n- \"internet users\",\n\n- \"AI users\",\n\n- \"customers\"\n\nThe population cannot be left under an ambiguous label such as ‘relevant users’. As applicable, the following questions must be answered: Which AI product? Which user surface? Which countries? Access as of what date? What age condition? Which language or languages? Which device and interface? Which plan or account type? What are the official terms of use? Must users pay to access the product? Does the measurement question apply to all users or only to a particular market or jurisdiction? Are active users or potential users being measured? Which user groups are out of scope? Which groups belong to the target population but are absent from the sampling frame? A change in any of these fields may change the target population. When it does, old and new scores cannot be combined in the same series without explanation.\n\n## 3. THE POPULATION IS NOT A SINGLE NUMBER\n\nThe official or estimated total population of a country is only the starting layer for GEO-1000. The audit population can consist of the following narrowing layers.\n\n### 3.1. Resident Population\n\nThe total human population living in the specified country or jurisdiction. This number:\n\n- does not imply any suitability in terms of age,\n\n- internet,\n\n- product access,\n\n- language,\n\n- account,\n\n- device\n\nconsiderations.\n\n### 3.2. Age-Eligible population\n\nThe population that meets the age requirement of the relevant AI product and research protocol. For a product:\n\n- minimum usage age,\n\n- parental consent requirement,\n\n- country-specific age rule\n\nmay exist. The research protocol may be more restrictive than the product condition. For example, the founding GEO-1000 panel may be composed of only adult participants to reduce ethical and legal complexity. In this case, the score is not: \"The entire population\":\n\n#### but rather \"Defined adult and product-eligible population\"\n\nresult.\n\n### 3.3. Connected Population\n\nThe population that can use the relevant product at a level where:\n\n- internet,\n\n- device,\n\n- connection quality,\n\n- technical access\n\nis available. Internet access may not be a binary characteristic. A person:\n\n- can use messaging,\n\n- however, they may not be able to run the heavy web application,\n\n- may use mobile internet,\n\n- however, they may not be able to access the desktop experience,\n\nmay not be able to upload responses or screenshots due to low connectivity. Connection limitations should not automatically exclude the user from the target population. It should be associated with the actual usage conditions of the measured product.\n\n### 3.4. Accessible Population from the Product Perspective\n\nFor the relevant AI product on the specified date:\n\n- officially launched,\n\n- could be used in accordance with the law and product conditions,\n\n- the necessary interface could be accessed\n\nIt is population. The absence of a product in a country does not mean that the people in that country are misrepresented. In this case: the representation response is none, there is no access coverage. The two outcomes should be kept separate.\n\n### 3.5. Population Suitable in Terms of Account and Plan\n\nSome products:\n\n- may require an account,\n\n- may require phone verification,\n\n- may require a payment method,\n\n- may be offered only with a corporate contract,\n\nmay unlock certain features only in a paid plan. These conditions affect which portion of the target population can actually access the relevant user surface. The free plan score cannot automatically be applied to paid plan users.\n\n### 3.6. Population Suitable in Terms of Language\n\nThe population that can use the relevant prompt and AI response at a sufficient level in the audit language. Language suitability alone:\n\n- cannot be determined only by the official language of the country,\n\n- the person's mother tongue\n\nIt should be distinguished:\n\n- Mother or home language\n\n- Language of daily use\n\n- AI interface language\n\n- Prompt language\n\n- Response language\n\n- Language for professional use\n\n- Reading proficiency\n\nA person can be suitable in multiple languages. This person cannot be counted as more than one full person in the global population calculation.\n\n### 3.7. Interface-Eligible population\n\nMeasurement:\n\n- web,\n\n- mobile application,\n\n- desktop application,\n\n- corporate workspace,\n\n- embedded assistant\n\ncan be used. The accessible user universe may differ for each interface. A user with web access may not have the mobile product. The corporate user surface may not be publicly available.\n\n### 3.8. Research-Eligible population\n\nEven if a person can use the product, for inclusion in research, additionally:\n\n- voluntary consent,\n\n- privacy,\n\n- data processing,\n\n- collecting screenshots,\n\n- country-specific research requirements,\n\n- does not imply any suitability in terms of age,\n\n- security\n\nconditions may need to be met. People who cannot be included in the research but use the actual product may be part of the target population. In this case, the sampling frame underrepresents the target population. This underrepresentation should not be hidden.\n\n### 3.9. Analytically Eligible population\n\nThe population for which the collected data can be used to produce results. The participant:\n\n- altered the prompt,\n\n- used the wrong product,\n\n- chose the answer,\n\n- submitted incomplete evidence\n\nIt is possible. Even if an individual belongs to the target population, their observation may not enter the analytical dataset. Valid observations and the target population are not the same thing.\n\n## 4. DEFINITION OF THE SUITABLE USER POPULATION\n\nFor AI product a at time t, the eligible user universe may be defined as follows:\n\nU_{a,t} = {u ∈ P_t: AgeEligible(u,a,t)=1 ∧ Connected(u,a,t)=1 ∧ ProductAvailable(u,a,t)=1 ∧ LawfulAccess(u,a,t)=1 ∧ InterfaceEligible(u,a,t)=1}\n\nHere:\n\n- Pt: the general human population at the specified date\n\n- u: individual human\n\n- a: AI product and user surface\n\n- t: measurement time\n\nIn terms of language and prompt, the target universe can later be narrowed down as follows:\n\nU_{a,c,l,s,p,t} = {u ∈ U_{a,t}: Country(u)=c ∧ LanguageEligible(u,l)=1 ∧ SurfaceEligible(u,s)=1 ∧ PromptRelevant(u,p)=1}\n\nHere:\n\n- c: country or jurisdiction\n\n- l: language and locale\n\n- s: user surface\n\n- p: prompt or question family\n\nThe purpose of this notation is not to record all individuals one by one. It is to make the logical boundary of the target population visible.\n\n## 5. BLINDLY MULTIPLYING MARGINAL RATES IS PROHIBITED\n\nThe eligible population may sometimes be approximately calculated with a formula like the following:\n\nN × p_age × p_internet × p_product × p_language\n\nThis calculation can only be justified under the independence of conditions or the assumption of an appropriate joint distribution. In real life, features may not be independent. For example:\n\n- internet usage may vary by age,\n\n- access to a paid plan may be related to income,\n\n- certain language use may be related to the region within the country and education,\n\ndevice access and connection quality may vary together. Multiplying marginal rates can predict a target population that is:\n\n- larger than necessary,\n\n- smaller than necessary\n\nThe order of preference should be:\n\n- Appropriate joint microdata or cross-tabulation\n\n- Reliable multidimensional population projection\n\n- Calibrated model\n\n- Marginal approach with explained rationale and sensitivity analysis\n\nIf simple multiplication is used:\n\n- assumption,\n\n- data source,\n\n- uncertainty,\n\n- alternative scenario\n\nshould be disclosed.\n\n## 6. AUDIT QUESTION CHANGES THE TARGET POPULATION\n\nNot every prompt has the same measurement universe for the entire world population.\n\n### 6.1. Global Identity Questions\n\nExample: “What kind of company is Apple.com?” This question can be widely asked among the suitable global user population for the product. Because the company identity is not limited to users of a specific market.\n\n### 6.2. Local Service Questions\n\nExample: “Does this company provide services in Turkey?” Main target population:\n\n- Eligible users in Turkey,\n\n- External users investigating Turkey's service\n\nIt is possible. It is not necessary to give equal weight to the entire world population.\n\n### 6.3. Price and Purchase Questions\n\nPrice:\n\n- country,\n\n- tax,\n\n- currency,\n\n- plan,\n\n- customer type\n\nmay vary. The target population should be defined according to the relevant commercial market and type of user.\n\n### 6.4. Legal and Licensing Issues\n\nLicensing and legal authority depends on the jurisdiction. The target population:\n\n- can obtain the relevant service in that country,\n\n- are subject to that legal regime.\n\nThe target population may be restricted accordingly.\n\n### 6.5. Advisory Issues\n\nThe target population for the question \"Is this company suitable for me?\" is not all people. It can be defined according to:\n\n- User needs\n\n- Country\n\n- Budget\n\n- Customer type\n\n- Scope of service\n\n- Eligibility and exclusion conditions\n\nThe advisory population may be narrower than the identity population.\n\n### 6.6. High-Risk Questions\n\nIn the fields of health, law, finance, or security:\n\n- judgement,\n\n- user profile,\n\n- risk status,\n\n- professional authority\n\nmore strongly limits the target population.\n\n## 7. POPULATION FRAMEWORK MUST BE LINKED TO PROMPT FAMILY\n\nIn an audit, there may be multiple population frameworks instead of a single target population.\n\n### 7.1. Core Identity Population\n\nThe global or broadly eligible user population measuring the entity's core identity.\n\n### 7.2. Service-Relevant Population\n\nIt is the population for which the entity can actually provide service or can be considered for service by the user.\n\n### 7.3. Market Transaction Population\n\nThe specific price, purchase, return, or commercial condition is the population to which it can be applied.\n\n### 7.4. Jurisdictional Population\n\nIt is the user population to which a specific licence, legal, or regulatory claim can be applied.\n\n### 7.5. Recommendation-Eligible Population\n\nIt is the user population that may be appropriate within the true scope of the entity.\n\n### 7.6. High-Risk Affected Population\n\nIt is a population for which misrepresentation can cause serious harm. These frameworks should not be confused with each other. A global identity score does not mean \"recommendable for the whole world.\"\n\n## 8. THE TYPE OF AI PRODUCT DETERMINES THE POPULATION\n\nNot all AI products are publicly available chat products.\n\n### 8.1. General Consumer AI Product\n\nTarget population:\n\n- There may be people who can access the product,\n\n- meet the terms of use,\n\n- and use it in the relevant language.\n\nThe GEO-1000 real user panel can be directly applied to this type of product.\n\n### 8.2. Enterprise AI Product\n\nThe product is only for:\n\n- employer contract,\n\n- corporate licence,\n\n- can be used with an in-house user account\n\nThe target population is not the general adult population:\n\n> It is the seat or user universe with corporate access to the product.\n\nCorporate results cannot be compared with the general consumer product in population weighting.\n\n### 8.3. Educational Product\n\nTarget population:\n\n- student,\n\n- teacher,\n\n- institution,\n\n- age group,\n\n- educational licence\n\nmay be limited in terms of. If children or minors are to be included, a separate ethical protocol is required.\n\n### 8.4. Developer or API Product\n\nAn API output may not be exposed directly through an end-user interface. The target unit is:\n\n- developer,\n\n- application,\n\n- API call,\n\n- controlled test session\n\nmay be. The GEO-1000 human population panel cannot be applied directly and unchanged to the API. Separately for the API:\n\n#### Controlled System Audit\n\nmay be required.\n\n### 8.5. Embedded AI Assistant\n\nAI:\n\n- operating system,\n\n- search interface,\n\n- shopping platform,\n\n- business software,\n\n- device\n\nIt may be embedded. The target population is the actual user population of the embedded product. It is not the total user population of the base model provider.\n\n### 8.6. Expert or Sectoral AI\n\nThe product is only for:\n\n- lawyers,\n\n- doctors,\n\n- finance professionals,\n\n- engineers\n\nmay be designed for. The target population should be limited by profession and access conditions.\n\n## 9. PRODUCT ACCESS STATUSES\n\nAccess status should be recorded for each country, product, interface, and date.\n\n### PA-0 — UNKNOWN\n\nAccess status could not be determined reliably.\n\n### PA-1 — UNAVAILABLE\n\nThe product is not offered in the relevant country or user group.\n\n### PA-2 — RESTRICTED\n\nThe product is limited to certain:\n\n- account,\n\n- invitation,\n\n- institution,\n\n- does not imply any suitability in terms of age,\n\n- plan,\n\n- device\n\nterms.\n\n### PA-3 — ACCESSIBLE BUT NOT OFFICIALLY SUPPORTED\n\nActual access to the product can be observed. However, official support or usage status is unclear or limited. These users should not be automatically added to the main official population panel. A separate experimental panel can be created.\n\n### PA-4 — OFFICIALLY AVAILABLE\n\nThe product is officially accessible and the main usage conditions are defined.\n\n### PA-5 — OFFICIALLY AVAILABLE WITH FULL AUDIT SURFACE\n\nThe product is not only inaccessible; it is controlled:\n\n- language,\n\n- plan,\n\n- interface,\n\n- under session conditions\n\nsuitable for generating the observation required by the protocol. The availability of the product in a country does not mean that all plans and languages are in compliance with the control.\n\n## 10. BYPASSING RESTRICTION IS NOT A REAL POPULATION MEASUREMENT\n\nParticipants cannot be asked to perform the following behaviours:\n\n- Declare the wrong country\n\n- Create a fake account\n\n- Change age information\n\n- Bypass official access restriction\n\n- Use someone else's account\n\n- Bypass the product's security or verification\n\n- Violating the terms of use\n\nThese behaviours can also be examined as a technical experiment. However:\n\n#### The experience of natural and legitimate user population\n\ncannot be reported as such. If there is no access in a country, the correct status should be: UNAVAILABLE / or / OUTSIDE_NATIVE_REACH. A single observation exceeding the restriction does not generate a country population score.\n\n## 11. DISTINCTION BETWEEN POTENTIAL USER AND ACTIVE USER\n\nThe people who can use the product are not the same community as the people who actually use the product.\n\n### 11.1. Potential Access Population\n\nThese are the people who can technically and legally use the product. This population answers the question: 'What would a suitable person see if they used the product?' The population is valuable for justice and product comparison.\n\n### 11.2. Active User Population\n\nThese are the people who actually used the product during a specific period. This population approaches the question: 'What do people see within the actual usage volume?' It requires reliable active usage data.\n\n### 11.3. Query-Exposed User Population\n\nThese are people who use the product and have a potential to ask questions about the monitored brand or topic. This population often cannot be known directly. Search or query behaviour data may be required.\n\n### 11.4. Population Actually Seeing the Response\n\nThey are users who are actually exposed to a specific representation output. This may not be fully measurable without real product provider data outside the user panel.\n\n## 12. EXPOSURE WEIGHT IS NOT THE SAME AS ELIGIBILITY WEIGHT\n\nTwo separate global results can be generated.\n\n### 12.1. Eligibility Population Weighted Result\n\nEach person is weighted according to their share within the defined eligible user population. This result is:\n\n- globally fair,\n\n- potentially accessible,\n\n- representative of the population\n\nin terms of\n\n### 12.2. Real Exposure Weighted Result\n\nAI products or user groups are weighted according to actual usage volumes. This result:\n\n- is valuable in terms of\n\n- current commercial impact,\n\n- actual user exposure,\n\noperational risk.\n\nHowever, reliable usage data may not be available. Estimates may rely on provider statements, market data, or incomplete sampling. Therefore, the type of weighting must be explicitly stated.\n\n### 12.3. Replacing One with Another is Prohibited\n\nEligible population score: does not mean \"Most of the actual users saw this response.\" Exposure score: does not mean \"Fair representation of the world population is ensured.\"\n\n## 13. ACCESS SCOPE AND REPRESENTATION ACCURACY SHOULD BE SEPARATED\n\nAn AI product may only be available in limited countries. In these countries, it can generate very accurate representations. Another product may have wider access but can produce lower accuracy. If we only look at representation accuracy, the product with limited access may appear superior. If we only look at access coverage, incorrect representation may be overlooked. Therefore, two separate results are needed.\n\n### 13.1. Access Coverage\n\nIndicates how much of the specified global reference population can appropriately use the relevant AI product. Candidate representation:\n\nAC_a = N(U_{a,t}) / N(G_t)\n\nHere:\n\n- Ua,t: eligible user population of the product\n\n- Gt: defined global reference population\n\n### 13.2. Conditional Representation Accuracy\n\nIt is the probability of seeing an accurate representation among users who can access the product. Candidate notation:\n\nRA_a = Pr(Y=1 | u ∈ U_{a,t})\n\n### 13.3. Effective Accurate Representation Access\n\n### CANDIDATE CONCEPT\n\nAn operational metric can be developed that combines access coverage with conditional accuracy:\n\nECR_a = AC_a × RA_a\n\nThis metric roughly approaches the question: “What proportion of the defined global reference population can access this product and meet the condition of seeing an accurate representation?” This is not yet the final NOMOS formula. Because:\n\n- access is different from actual usage,\n\n- weight and dependency structures are complex,\n\n- product usage is voluntary,\n\nThe correct representation depends on the prompt. However, it maintains the basic distinction:\n\n> Not being able to access is not the same as getting a wrong answer.\n\n## 14. NATIVE REACH AND COMMON SUPPORT\n\nWhen comparing AI products, two separate population universes are needed.\n\n### 14.1. Native Reach Population\n\nEach AI product has its eligible population within its official and actual access domain. This result answers the following question:\n\n> To what extent does the product accurately represent the population it can actually reach?\n\nTarget populations may differ between products.\n\n### 14.2. Common Support Population\n\nIn a common form across all compared AI products:\n\n- officially accessible,\n\n- can be used under the same language or interface condition\n\nis the population. This result answers the following question:\n\n> Which product produces more accurate representation in the same user universe?\n\n### 14.3. Two Results Should Be Published Together\n\nNative Reach:\n\n- shows actual product access,\n\n- natural geographic distribution\n\nCommon Support: provides a fairer product comparison. A product can be high in Native Reach score, but lower in Common Support score. Two results answer different questions.\n\n## 15. MARKET-RELATED POPULATION\n\nThe global identity score of an entity may differ from its commercial target market score. Example: If a local law firm provides services only in Turkey: the global identity question can be measured in the eligible user population worldwide, but recommendations and service suitability should only be measured for users in Turkey or within the actual scope of services. The institution may select the result in its favour:\n\n- only the high-scoring market,\n\n- only his/her own language,\n\n- only its existing customers\n\nthe target population cannot be announced. Target market:\n\n- Before the audit begins,\n\n- to the service records,\n\n- within the scope of the contract and operations\n\nIt should be locked according to.\n\n## 16. PERSON, ACCOUNT AND SESSION SEPARATION\n\nPopulation weight belongs to people. Not to AI accounts.\n\n### 16.1. One Person, Multiple Accounts\n\nThe same person:\n\n- free,\n\n- paid,\n\n- corporate\n\ncan have multiple accounts. These accounts can produce separate observations in product surface comparisons. They cannot be counted as separate individuals in the global human population.\n\n### 16.2. One Account, Multiple People\n\nA family, team, or corporate account may be used by multiple people. It is not equal to the number of account users.\n\n### 16.3. One Person, Multiple Sessions\n\nA person can produce separate observations in different sessions. These observations can measure time or session stability. They do not increase independent population size.\n\n### 16.4. One Person, Multiple AI Products\n\nIt is valuable for paired product comparison. However, observations are dependent due to shared characteristics at the user level.\n\n### 16.5. Main Weight Unit\n\nIn the population-based GEO score, the main weight unit should be:\n\n#### human\n\nCalculations, session, and product observations should be associated under the human unit.\n\n## 17. LANGUAGE POPULATION\n\nUsing only the country’s official language is not sufficient for language fairness.\n\n### 17.1. Prompt Language\n\nThe language in which the participant asks the audit question.\n\n### 17.2. Interface Language\n\nThe language used in the menu and user interface of the AI product. It may not be the same as the prompt language.\n\n### 17.3. Response Language\n\nThe language in which the AI actually responds. The system may respond in a different language than the prompt language. This is also a measurement result.\n\n### 17.4. User Proficiency Language\n\nThe language in which the participant can meaningfully use the prompt and response. It does not have to be the native language.\n\n### 17.5. Locale\n\nThe regional context along with the language. The same language can be different:\n\n- country,\n\n- price,\n\n- law,\n\n- terminology\n\nmay carry the condition.\n\n## 18. DOUBLE COUNTING OF MULTILINGUAL USERS\n\nA person:\n\n- Turkish,\n\n- may be able to use English.\n\n- German\n\nThis person may take three separate language tests. However, counting them as three people in the global population weight distorts the population estimate. Three methods can be used.\n\n### 18.1. Single Mother Tongue Assignment\n\nThe person is assigned to a single language cell in the main analysis. Other languages are used as additional or paired tests.\n\n### 18.2. Split Person Weight\n\nThe person's population weight can be divided among the appropriate languages in a predefined manner. The total weight does not exceed one person.\n\n### 18.3. Separate Multilingual Matched Panel\n\nMultilingual users are used for cross-language comparisons of the same person. This panel is analysed separately from the main population panel. Regardless of the method used:\n\n> The total population weight of the same person cannot be replicated without explanation.\n\n## 19. HOW SHOULD LANGUAGE PROFICIENCY BE VERIFIED?\n\nA participant's statement: \"I know this language.\" can be an initial record. However, in high security auditing, one of the following methods can be used:\n\n- Native language or daily usage declaration\n\n- Short reading and comprehension test\n\n- Language record of the panel provider\n\n- Local adjudicator verification\n\n- Education or professional use record\n\n- Pilot mission success\n\nExtremely difficult language test panels can be accessible to trained and high-income users. The purpose is not to conduct an academic language exam. It is to verify the competence to understand the prompt and response as a real user.\n\n## 20. WHAT HAPPENS IF A LANGUAGE IS NOT SUPPORTED?\n\nAn AI product may:\n\n- not officially support a particular language,\n\n- partially support it,\n\n- not claim to support it but can generate responses\n\nIt may be. Language statuses should be recorded separately:\n\n### OFFICIALLY_SUPPORTED\n\n### LIMITED_SUPPORT\n\n### OBSERVED_BUT_UNSUPPORTED\n\n### NOT_SUPPORTED\n\n### UNKNOWN\n\nResponse received in a language that is not officially supported:\n\n- discovery,\n\n- experimental access,\n\n- language resilience\n\nmay be measured. It should not be automatically added to the main supported user population score. Language access limit should also be reported.\n\n## 21. COUNTRY POPULATION WEIGHTING AND Country Observer COVERAGE\n\nPopulation-weighted main panel and country visibility panel serve separate purposes.\n\n### 21.1. Population Panel\n\nParticipants are distributed according to their share in the product-appropriate target population. Larger eligible populations receive higher weight. Produces the main global population score.\n\n### 21.2. Country Observer Panel\n\nThe Country Observer Panel seeks at least one observation from smaller countries that receive little or no representation in the main population sample. A single Country Observer observation:\n\n- indicates access is available,\n\n- that a specific singular result has occurred,\n\n- shows the need for new research\n\nmay indicate. Does not produce a country score on its own.\n\n### 21.3. Sufficiency for Country Score\n\nFor country-based rates to be published:\n\n- minimum sample,\n\n- valid observation rate,\n\n- uncertainty,\n\n- panel representativeness\n\nconditions must be met. With one or several participants: 'The country's NOMOS score' cannot be announced.\n\n## 22. Language Fairness Panel\n\nPopulation weight may make low-population or low-resource languages invisible in the main score. Therefore, a separate Language Fairness Panel can be established. Purpose:\n\n- to test underrepresented languages,\n\n- to detect severe language-based degradation,\n\n- to identify prompt translation issues,\n\n- to make low-resource language behaviour visible\n\nmust be. The Language Fairness Panel can provide additional samples to the main population panel. However, over-sampled language observations should be returned to actual population weights in the global score. Otherwise, the fairness panel alters the global population estimate.\n\n## 23. ACCESSIBILITY PANEL\n\nUsers with visual, hearing, motor, cognitive, or other accessibility needs are a real part of the target population. If the NOMOS Capture or research task is not suitable for these individuals' participation: users cannot be removed from the target population. Correct outcome:\n\n> The sampling frame is incomplete in terms of accessibility.\n\nA separate accessibility panel can evaluate the following:\n\n- Screen reader compatibility\n\n- Using the keyboard\n\n- Voice input\n\n- Low vision settings\n\n- Cognitive load\n\n- Mobile accessibility\n\n- Usability of the Capture tool\n\nThe accessibility panel should see users as part of the population coverage, not as a \"special exception.\"\n\n## 24. TARGET POPULATION AND SAMPLING FRAME\n\nThese two concepts should not be confused with each other.\n\n### 24.1. Target Population\n\nThe entire group of people about which one wants to make a conclusion.\n\n### 24.2. Sampling Frame\n\nThe population from which participants can actually be selected:\n\n- panel,\n\n- list,\n\n- database,\n\n- joint organisation,\n\n- user pool\n\nis the cluster. The sampling frame may not fully cover the target population.\n\n### 24.3. Contactable Population\n\nThese are the people to whom the research invitation can actually be delivered.\n\n### 24.4. Population Accepting to Participate\n\nThose who voluntarily participate in the research.\n\n### 24.5. Population Completing the Task\n\nThose who complete the audit task.\n\n### 24.6. Valid Analytical Population\n\nThey are the ones who produce observations suitable for the protocol and usable in the analysis. The numbers of these layers should be reported separately.\n\n## 25. COVERAGE ERROR\n\nThe sampling frame may exclude some parts of the target population. This is called coverage error. Examples:\n\n- Only professional online panels\n\n- Only desktop users\n\n- Only people with credit cards\n\n- Only those using an English interface\n\n- Only residents of large cities\n\n- Only users highly interested in technology\n\n- Exclusion of people with accessibility needs\n\nCoverage error:\n\n- should be explained,\n\n- should be measured if possible,\n\nshould be reduced with weighting or an additional panel. Weighting cannot magically create a group that does not exist in the sampling frame.\n\n## 26. OVER-COVERAGE\n\nPeople who do not belong to the target population may be found in the sampling frame. Examples:\n\n- Participant in a country where the product is not accessible\n\n- Person who does not meet the required age condition\n\n- Incorrect plan or user surface\n\n- Participant who cannot use the audit language\n\n- Person found again with multiple registrations\n\n- User working on behalf of another country\n\nExcess coverage records should not be included in the main score. However, the reason for exclusion and the rate should be reported.\n\n## 27. NON-RESPONSE AND SELF-SELECTION\n\nPeople who agree to participate in the research may systematically differ from the target population. Volunteers:\n\n- May use AI products more frequently,\n\n- May be more interested in technology,\n\n- May have a higher level of education,\n\n- May already be familiar with the brand,\n\nMay behave differently due to reward or compensation. Therefore: the expression \"1,000 random people in the world\" should only be used in truly probabilistic and documented selection designs. If a volunteer or online panel is used, the correct expression is:\n\n#### Layered and weighted voluntary user panel\n\nmay be possible. The panel type cannot be stored.\n\n## 28. SUITABLE POPULATION CHANGES OVER TIME\n\nThe user universe of an AI product is not fixed. The following can change:\n\n- Supported countries\n\n- Minimum age\n\n- Free access\n\n- Paid plan structure\n\n- Mobile or web availability\n\n- Language support\n\n- Account terms\n\n- Enterprise product coverage\n\n- Legal and regulatory status\n\nFor this reason, each population record:\n\n- observation date,\n\n- period of validity,\n\n- source version,\n\n- last verification date\n\nmust carry.\n\n### 28.1. Change During Measurement Wave\n\nIf product access materially changes during the wave:\n\n- the wave can be stopped,\n\n- two separate periods can be created before and after the change,\n\naffected countries can be analysed separately. Users before and after the change cannot be combined as a single population.\n\n## 29. POPULATION SOURCE FILE\n\nEach audit should store the target population data it uses in a versioned file. As applicable, the following fields should be included:\n\n- Country\n\n- Reference population\n\n- Population history\n\n- Age eligibility\n\n- Internet and device access\n\n- Product access status\n\n- Interface access\n\n- Plan and account requirement\n\n- Language groups\n\n- Research eligibility\n\n- Common distribution method\n\n- Uncertainty\n\n- Sources\n\n- Last verification date\n\n- Responsible person\n\n- Population version\n\nCountry quotas cannot be reproduced after results without this file.\n\n## 30. SYNTHETIC PRODUCT AND COUNTRY DISPLAY\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The following countries, AI product, population numbers, and access rates are entirely fictional. They do not represent any real country or product. Let's consider a synthetic general consumer product called Orion AI. There are four synthetic countries.\n\n### 30.1. If the total population was used\n\nCountry A would have a large share in the main sample due to its very large population. However, the product cannot be used in that country. Participants could not produce a natural user experience.\n\n### 30.2. When Native Reach Population Is Used\n\nThe theoretical population shares of the main product panel are approximately:\n\n- Country B: 67.53%\n\n- Country C: 32.47%\n\n- Country D: 0.003%\n\n- Country A: outside of Native Reach\n\nThe theoretical share of Country D in a main sample of 1,000 people is far below one person. Therefore, the population-weighted main score should not give high weight to Country D. However, if the country is not intended to be completely invisible:\n\n#### Country Observer Panel\n\nAt least one exploratory observation can be obtained. This observation:\n\n- is not the score of Country D,\n\n- do not change the main population weight,\n\nshows that the product behaviour can be observed in the country.\n\n### 30.3. How is Country A Reported?\n\nFor Country A:\n\n### The OUTSIDE_NATIVE_REACH\n\nstatus is used. The exclusion of Country A should not hide the product's access limits. The public results card also shows:\n\n- Product access coverage\n\n- Share not accessible in the global reference population\n\n- Native Reach score\n\n- Common Support comparison score\n\n## 31. SYNTHETIC PROMPT POPULATION REPRESENTATION\n\n### SYNTHETIC EXAMPLE\n\nLet's consider three prompts for the same Orion AI product.\n\n#### Prompt A\n\n\"What kind of company is Asteron?\" Target population: could be the broad suitable user population of Orion AI.\n\n#### Prompt B\n\n\"Does Asteron provide corporate travel services in Turkey?\" Target population:\n\n- Eligible users in Turkey,\n\n- Identified external users researching the Turkey service\n\nmay be.\n\n#### Prompt C\n\n“Is Asteron suitable for legal immigration consultancy for me?” If Asteron does not provide such a service: there may be no advice-eligible population, the correct answer should exclude all suitable users. Combining this prompt into the overall identity score might be incorrect. This example shows:\n\n> Even for the same entity and the same AI product, the target population may vary depending on the user question measured.\n\n## 32. FOUR PANEL ARCHITECTURE\n\n### CANDIDATE ARCHITECTURE — CANDIDATE ARCHITECTURE\n\nFour complementary panels are recommended for the global and fair implementation of GEO-1000.\n\n### 32.1. Population Panel\n\nThe main panel weighted according to the product-eligible user population. Produces the main population estimate.\n\n### 32.2. Country Observer Panel\n\nPrevents small countries that cannot provide sufficient samples in the population panel from being completely invisible. It is not sufficient on its own to produce a country score.\n\n### 32.3. Language Fairness Panel\n\nAlso measures languages that are low-resource or suppressed in population averages. It reconnects to the global main score with real population weights.\n\n### 32.4. Accessibility Panel\n\nMeasures accessibility needs and assistive technology users that the standard digital panel might exclude. Observations from these four panels:\n\n- goal,\n\n- weight,\n\n- inference universe\n\nIt should be labelled separately in terms of maintenance. The same observation cannot be used at full weight on more than one panel without explanation.\n\n## 33. ADULT INITIAL PANEL\n\n### CANDIDATE BASE RULE — CANDIDATE FOUNDATIONAL RULE\n\nIn the first general release of GEO-1000, the real user dashboards:\n\n#### from adult participants who meet the product and country conditions\n\nIt is recommended to be created. This approach:\n\n- consent,\n\n- data protection,\n\n- collecting screenshots,\n\n- high-risk research,\n\n- parental consent\n\nIt can reduce its complexity. However, the following conclusion cannot be drawn: 'The experience of child and young users has been measured with the adult panel.' In products that affect children or young people, separately:\n\n- ethical approval,\n\n- age-appropriate protocol,\n\n- parent or guardian processes,\n\n- harm protection\n\nare required.\n\n## 34. NO POPULATION FRAMEWORK CAN BE CALLED \"ALL HUMANITY\"\n\nA product:\n\n- if it does not exist in some countries,\n\n- if it does not support some languages,\n\n- if it only works on certain devices,\n\n- if it has certain age or plan condition,\n\nresult: it cannot be presented as “the GEO experience of all the people in the world”. The correct expression:\n\n#### Estimated representation in the target population suitable for the specified product\n\nshould be used. The word global:\n\n- multi-country,\n\n- multi-lingual,\n\n- large population\n\ncan define a measurement. It does not mean unlimited and universal coverage.\n\n## 35. FOUNDATIONAL ETHICAL LIMIT OF THE POPULATION\n\nDefining a population for measurement does not mean considering some people worthless. A group may receive a low weight in the main score. However:\n\n- high risk,\n\n- language fairness,\n\n- accessibility,\n\n- country visibility\n\nmay also be examined separately. A group may be excluded from the Native Reach score because they cannot access a product. This lack of access can also be reported as a global coverage gap. The purpose of NOMOS:\n\n- compressing everyone into a single artificial average,\n\n- seeing only the largest populations,\n\n- appearing to symbolically represent small communities with a single person\n\nis not the goal. The objective is:\n\n> To represent each person in the correct denominator of the correct question, without mixing their weight and visibility.\n\n## 36. MANDATORY NORMATIVE PROVISIONS\n\n**CH05-N01**\n\nThe target user population must be defined separately for each AI product.\n\n**CH05-N02**\n\nThe total population of a country cannot be directly used as the AI-eligible user population.\n\n**CH05-N03**\n\nThe target population must be linked to product, country, age, access, interface, plan, language, prompt, and time conditions.\n\n**CH05-N04**\n\nTarget population and exclusion rules must be versioned before seeing the results.\n\n**CH05-N05**\n\nThe target population of an AI product cannot be transferred to another AI product without explanation.\n\n**CH05-N06**\n\nLack of product access cannot be scored as a misrepresentation result; it must be reported separately as access coverage.\n\n**CH05-N07**\n\nParticipants cannot be asked to bypass product or country access restrictions.\n\n**CH05-N08**\n\nAccess not officially supported cannot be automatically added to the main Native Reach population.\n\n**CH05-N09**\n\nIndividual, account, and session units must be separated from each other.\n\n**CH05-N10**\n\nMultiple accounts or languages of a person cannot multiply the individual in population weight.\n\n**CH05-N11**\n\nLanguage eligibility cannot be assigned based solely on the country's official language.\n\n**CH05-N12**\n\nRequested language, interface language, response language, and user proficiency language must be recorded separately.\n\n**CH05-N13**\n\nThe total population weight of a person using multiple languages cannot exceed one person without explanation.\n\n**CH05-N14**\n\nThe population-weighted panel and the Country Observer panel should be kept separate.\n\n**CH05-N15**\n\nOne or several Country Observer observations cannot be published as a country score.\n\n**CH05-N16**\n\nIn the Language Fairness Panel, over-sampled observations should be reverted to their true population weight in the global score.\n\n**CH05-N17**\n\nUsers with accessibility needs cannot be excluded from the target population because the data collection tool is not suitable.\n\n**CH05-N18**\n\nThe target population, sampling frame, contacted population, participants, and valid analytical observations should be reported separately.\n\n**CH05-N19**\n\nA non-probability or voluntary panel must not be represented as a random sample of the global population.\n\n**CH05-N20**\n\nWeighting cannot make a group that is completely absent in the sampling frame observable.\n\n**CH05-N21**\n\nPotential suitable population and active user exposure cannot be used as the same metric.\n\n**CH05-N22**\n\nThe results Native Reach and Common Support must be separated from each other.\n\n**CH05-N23**\n\nThe same target population cannot automatically be used for identity, service, price, legal, and recommendation prompts.\n\n**CH05-N24**\n\nPopulation data, product access status, and language distribution must carry a timestamp and source version.\n\n**CH05-N25**\n\nMarginal utility ratios cannot be blindly multiplied without an explanation of the independence assumption.\n\n**CH05-N26**\n\nIf the product type is not suitable for the general consumer panel, the GEO-1000 human population method cannot be applied without explanation.\n\n**CH05-N27**\n\nChildren or underage users cannot be included in the main panel without a separate ethical and legal protocol.\n\n**CH05-N28**\n\nProduct access coverage and representational accuracy must be published as separate results.\n\n**CH05-N29**\n\nThe term global cannot make countries, languages, ages, interfaces, or plans that are out of scope invisible.\n\n**CH05-N30**\n\nThe target population record must have a responsible human or institutional owner.\n\n## 37. FORMS OF FAILURE\n\n**CH05-F01 — COUNT THE TOTAL POPULATION AS ELIGIBLE POPULATION**\n\nThe country's population is directly included in the denominator without evaluating product, age, internet, language, and interface conditions.\n\n**CH05-F02 — APPLY THE SAME WORLD MAP TO ALL PRODUCTS**\n\nAI products with different access areas are measured against the same target population.\n\n**CH05-F03 — COUNT A USER WHO EXCEEDS THE LIMIT AS A NATURAL USER**\n\nObservations that exceed official access conditions are presented as a country population experience.\n\n**CH05-F04 — COUNTING PRODUCT ABSENCE AS WRONG ANSWER**\n\nCountries that are inaccessible are considered zero in representation accuracy.\n\n**CH05-F05 — COMPLETELY HIDING PRODUCT ABSENCE**\n\nLarge populations that are inaccessible are removed from the Native Reach denominator; access coverage is not shown either.\n\n**CH05-F06 — COUNTING ONE PERSON AS MORE THAN ONE**\n\nThe same person's account, language, or product observations multiply the population weight.\n\n**CH05-F07 — COUNTING ACCOUNT AS PERSON**\n\nMultiple account ownership or shared account usage is not considered.\n\n**CH05-F08 — COUNTING OFFICIAL LANGUAGE AS USER LANGUAGE**\n\nThe official language of the country is considered the audit language for all participants.\n\n**CH05-F09 — COUNTING MULTILINGUAL INDIVIDUALS TWICE**\n\nThe same individual receives full population weight in each language.\n\n**CH05-F10 — ASSIGNING ONE PERSON TO A SMALL COUNTRY AND EXPLAINING THE COUNTRY SCORE**\n\nCountry Observer observation is transformed into a statistical country estimate.\n\n**CH05-F11 — COMPLETELY DELETING SMALL COUNTRIES**\n\nBecause the population share is low, the country or language is not included in any additional coverage panel.\n\n**CH05-F12 — OVERWEIGHTING EXAMPLES OF LANGUAGE FAIRNESS IN THE MAIN SCORE**\n\nOver-sampled languages are not returned to their true population weight.\n\n**CH05-F13 — COUNTING THE ACTIVE USER AS POTENTIAL POPULATION**\n\nActual usage data is presented as if it includes all people who could access the product.\n\n**CH05-F14 — COUNTING THE POTENTIAL POPULATION AS ACTUAL EXPOSURE**\n\nIt is assumed that the response of everyone who could use the product is actually seen.\n\n**CH05-F15 — COUNTING THE VOLUNTARY PANEL AS A RANDOM WORLD SAMPLE**\n\nSelf-selection and panel bias are concealed.\n\n**CH05-F16 — IGNORING COVERAGE ERROR WITH WEIGHTING**\n\nIt is claimed that groups not present in the frame are represented solely by weighting.\n\n**CH05-F17 — REMOVING USERS EXCLUDED BY THE DATA COLLECTION TOOL FROM THE POPULATION**\n\nAccessibility or device issues are interpreted as if the user is not suitable.\n\n**CH05-F18 — APPLYING THE SAME POPULATION TO ALL PROMPTS**\n\nGlobal identity, local pricing, and legal advice questions are combined on the same denominator.\n\n**CH05-F19 — APPLYING GENERAL CONSUMER POPULATION TO THE API PRODUCT**\n\nThe human user panel is transferred directly to the developer or application product.\n\n**CH05-F20 — SECRETLY COMBINING FREE AND PAID PLANS**\n\nA single product distribution is published without recording the plan difference.\n\n**CH05-F21 — UNIFYING WEB AND MOBILE POPULATIONS**\n\nInterface access and behaviour differences remain invisible.\n\n**CH05-F22 — IGNORING THE AGE LIMIT**\n\nInappropriate age groups are added to the target population or panel without explanation.\n\n**CH05-F23 — MULTIPLYING MARGINAL RATES AS IF INDEPENDENT**\n\nThe relationship between age, internet, language, and access is ignored.\n\n**CH05-F24 — CHANGING THE POPULATION AFTER THE RESULT**\n\nA low-scoring country or user group is removed from the target population.\n\n**CH05-F25 — USING THE OLD PRODUCT REACH MAP**\n\nActual product reach on the measurement date is not verified.\n\n**CH05-F26 — CONFUSING COMMON SUPPORT AND NATIVE REACH RESULTS**\n\nComparisons made on joint population are presented as if they reflect the product's actual global reach.\n\n**CH05-F27 — NARROWING THE TARGET MARKET ACCORDING TO COMMERCIAL INTEREST**\n\nThe organisation only declares the countries where it succeeds as target markets.\n\n**CH05-F28 — COUNTING THE SCOPE OF ACCESS AS GEO ACCURACY**\n\nBeing present in many countries is presented as proof of accurate representation.\n\n**CH05-F29 — SELLING LIMITED ACCESS AS HIGH GEO SUCCESS**\n\nHigh accuracy in a small number of countries is presented as global superiority.\n\n**CH05-F30 — USING THE TERM “THE WHOLE WORLD” WITHOUT COVERAGE RECORD**\n\nExclusions of product, language, country, and plan are hidden.\n\n## 38. AUDIT PROCEDURE\n\n### Step 1 — Lock the AI Product and User surface\n\nProduct Interface Plan Account type Displayed model Measurement date is recorded.\n\n### Step 2 — Classify the Measurement Question\n\nPrompt:\n\n- global identity,\n\n- local service,\n\n- price,\n\n- law,\n\n- recommendation,\n\n- high risk\n\nwhich of their families does it belong to?\n\n### Step 3 — Determine the Global Reference Population\n\nThe initial human universe to which the result will be associated is defined.\n\n### Step 4 — Determine Age and Research Eligibility\n\nProduct age condition Research age condition Consent and data processing terms are recorded.\n\n### Step 5 — Verify Country-Based Product Access\n\nAn access status between PA-0 and PA-5 is assigned for each country.\n\n### Step 6 — Determine Internet, Device, and Interface Compatibility\n\nThe population that can use the measured surface is estimated.\n\n### Step 7 — Establish the Language and Locale Universe\n\nPrompt language, Interface language, Response language, and User language proficiency are separated.\n\n### Step 8 — Apply Prompt Relevance\n\nIt is evaluated whether all users belong to the target population of the relevant prompt family.\n\n### Step 9 — Select the Population Model Appropriate to the Product Type\n\nConsumer, enterprise, education, API, or embedded product is separated.\n\n### Step 10 — Compare the Target Population with the Sampling Frame\n\nIt is recorded as under-coverage and over-coverage.\n\n### Step 11 — Identify the Main and Complementary Panels\n\nThe scopes of the Population Panel, Country Observer Panel, Language Fairness Panel, and Accessibility Panel are versioned.\n\n### Step 12 — Separate the Native Reach and Common Support Universes\n\nIntra-product results and inter-product comparison populations are recorded separately.\n\n### Step 13 — Separate Access Scope and Accuracy Results\n\nThe accuracy between the population a product can reach and the users reached is reported separately.\n\n### Step 14 — Freeze the Population File\n\nData sources, assumptions, date, and population version are locked.\n\n### Step 15 — Track Material Changes\n\nIf product access or conditions change during measurement, a new version is created.\n\n## 39. REQUIRED EVIDENCE\n\nAI-product identity; user surface; plan and account conditions; measurement date; countries in which the product is accessible; country-level access status; age conditions; legal and ethical participation conditions; country-population records; age distributions; internet and device-access records; language and locale distributions; product-language support; interface accessibility; target prompt family; target-population definition; sampling frame; undercoverage; overcoverage; panel type; participant-selection method; probability or opt-in panel status; multilingual-user weights; Country Observer coverage; Language Fairness coverage; Accessibility Panel coverage; Native Reach population; Common Support population; active-user data, where used; access-coverage calculation; assumptions and sensitivity analyses; population version; and last verification date.\n\nResponsible person or institution\n\n## 40. AUDIT CHECKLIST\n\nWhich AI product and user surface is being measured? For which date was the target population defined? Were the total country population and the eligible user population separated? Was an age requirement applied? Was internet and device access evaluated? Is the product officially accessible on a country basis? Were participants asked to overcome access restrictions? Were free and paid plans separated? Were web and mobile interfaces separated? Did the prompt family change the target population? How was language and locale suitability determined? Were multilingual users counted twice? Is the difference between the target population and the sampling frame clear? Is the panel probabilistic or voluntary? Was coverage error measured? Were users with accessibility needs silently excluded?\n\nHave Country Observer observations been converted into a country score? Did Language Fairness examples have excessive weight in the main score? Were the results of Native Reach and Common Support separated? Are coverage and representation accuracy separate? Were active user and potential user results separated? Is the product type suitable for the human population panel? Were the marginal rates in population weights multiplied blindly? Was the population definition changed after the results were seen? Are the used population and access data versioned? Are untested groups visible? Is the term “global” consistent with the actual coverage? Is there a clear accountable owner of the population record?\n\n## 41. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “Isn’t it fairer to distribute directly according to the world population?”\n\nThe world population is a strong starting base. However, including people who cannot use the product in the main product sample does not generate real user experience. The correct method is:\n\n- population weight within the product-eligible population,\n\n- global access coverage.\n\nBoth results must be reported.\n\n### Objection 2 — “Shouldn’t most users come from large countries?”\n\nCountries that meet the product, language, and access conditions should naturally have a high share. However, just because the total population is large, a fake user experience cannot be generated from a country where the product is not accessible.\n\n### Objection 3 — “Why would it be wrong to give one person to a small country like San Marino?”\n\nA person is valuable for exploration and Country Observer observation. What is wrong is to present that person's result as a 'country score.' The visibility of the small country can be protected with the Country Observer Panel. The main population weight remains according to the actual population share.\n\n### Objection 4 — “If we give at least one person to each country, wouldn't we look more global?”\n\nA wider geographical coverage is provided. However, the population estimate may be distorted. Therefore:\n\n- geographical scope,\n\n- population density\n\nThey are separate results. It may be a scope panel to give one person to each country. It is not a population panel.\n\n### Objection 5 — \"If we give zero to a product in an inaccessible country, wouldn't the product be unfairly punished?\"\n\nAn unreachable country is not counted as zero in representation accuracy. It becomes visible in the access coverage metric. This distinction protects the product from both unfair penalties and false universality.\n\n### Objection 6 — “If there is real usage data, why do we need population data?”\n\nReal usage data shows operational impact. However, the product's:\n\n- which populations it has not reached,\n\n- the geographical inequality in the user base,\n\n- potential user fairness\n\nIt does not show alone. The two types of data answer different questions.\n\n### Objection 7 — 'Since all users will ask the same prompt, why is prompt relevance important?'\n\nIdentity questions may be meaningful in a broad population. Price, legal, or advice questions are only meaningful for certain users and markets. Irrelevant population may distort the response distribution away from the user reality.\n\n### Objection 8 — \"If a person knows multiple languages, why shouldn’t they be fully counted in all languages?\"\n\nYou can test all languages in cross-language comparison. However, multiplying the same person in population weighting artificially increases the world population. The person's weight should be divided, or a separate matched panel should be used.\n\n### Objection 9 — \"Doesn’t testing only adults leave the global result incomplete?\"\n\nIt does. This limitation should be clearly stated. Starting with an adult panel may be an ethical and operational choice. It does not make a judgement about child users.\n\n### Objection 10 — 'Isn't the online research panel sufficient?'\n\nIt can be very valuable for pilot and weighted research. But the panel:\n\n- how much of the target population does it cover,\n\n- who is being left out,\n\n- how voluntary participation creates bias\n\nIt should be explained. Using a panel is not a mistake. Presenting the panel as random like the world population is a mistake.\n\n### Objection 11 — “Why is the Accessibility panel related to the main GEO score?”\n\nBecause a portion of real AI users use assistive technology or different forms of interaction. If the data collection tool excludes these people, the main score may not represent the entire target population.\n\n### Objection 12 — “Doesn't having so many different population definitions make a single NOMOS score impossible?”\n\nNo. A single score can be produced. However, it should be clear which population question it answers. For example:\n\n- Native Reach Population Score\n\n- Common Support Score\n\n- Market-Relevant Score\n\n- Language Fairness Result\n\ncan carry separate records. The meaning of a number arises from the clarity of its denominator.\n\n## COMMON JUDGEMENT OF CHAPTER 44\n\nGEO-1000 contains ‘1,000’ in its name, but the value of those 1,000 people does not arise from the number alone. It arises from the answers to these questions: Who are they? From which population were they drawn? Can they genuinely use the product? Where do they live? Which languages do they use? How many times was the same person counted? Are small countries visible? Do large countries retain their true population weight? Where does the score account for people the product cannot reach? Whom did the panel infrastructure exclude? Which result measures potential access, and which measures actual exposure? A large country may have a large product-eligible population and therefore deserve substantial weight in the main panel. But if the product is unavailable there, the audit must not fabricate users. The access gap must be disclosed separately.\n\nIf a country is very small, the main population weight should remain low. However, it does not have to be completely invisible. It can be observed with a monitoring panel. If a language has a small population, it may not determine the global average. However, if there is a severe representation distortion in that language, it should also be visible. If a person knows three languages, they can produce three different experiences. However, they are not three different people. If an AI product only provides certain behaviour on a paid plan, the score of free users does not represent that plan. A human population panel is suitable for a consumer chat product. The same population model may not be suitable for an API or enterprise product. Therefore, NOMOS's fifth measurement law is as follows:\n\n> Even the most sensitive score gives the correct answer to the wrong question if the denominator is incorrect.\n\nIts sixth law is:\n\n> Population weight does not replace country and language visibility.\n\nIts seventh law is as follows:\n\n> A person who cannot access [something] has not been misrepresented; however, they are part of the global access boundary.\n\nIts eighth law is as follows:\n\n> A person having multiple accounts, languages, or sessions does not increase the human population.\n\nThe ninth law is as follows:\n\n> Global measurement is not a claim that everyone is measured; it is an honest record of who is measured and who is not.\n\n## Order of Section 5 of NOMOS\n\n> Do not give me only country populations. / Show which people can actually use the product.\n\n> Do not fill a large country with fake users when the product is not there.\n\n> Do not give one person to a small country and declare that person's response as the voice of the country.\n\n> Do not duplicate a person as much as their accounts, languages, and sessions.\n\n> Do not mistake the official language for the real language usage of people.\n\n> Do not make a free user the representative of a paid plan, or a mobile user the representative of the web interface.\n\n> Do not call the volunteer panel a random world population.\n\n> Do not erase a user with accessibility needs from the population because the tool does not work.\n\n> Do not remove the population that the product does not reach from the accuracy score and hide the global access gap.\n\n> Do not apply the same share to a price question as an identity question, or to a global brand as a local licence.\n\n> First, define the product. / Then define the question. / Then define the human universe. / Then limit access, language, age, and interface. / Then explain whom you cannot select. / And only after that distribute 1,000 people.\n\n## The Chapter's Closing Sentence\n\n> In GEO-1000, justice is not giving the same number of people to every country; it is representing each person in the correct target population, with the proper weighting, without hiding the places where they are not measured.\n\n## Normative Core\n\n> Every GEO population audit MUST define, before result inspection, a time-bound and product-specific target user population for each: - AI product and user surface, - country or jurisdiction, - age and lawful-access condition, - account or plan condition, - language and locale, - interface, - prompt family, - and measurement purpose. Total resident population MUST NOT be treated as product-eligible population without justified eligibility adjustments. Product unavailability MUST be reported as an access-coverage limitation, not silently scored as incorrect representation or silently removed from global scope. Potential access, active use, actual exposure, Native Reach, Common Support, market relevance, and research observability MUST remain distinct population concepts. Persons, accounts, sessions, languages, and product observations MUST NOT be treated as interchangeable population units. Oversampled countries, languages, and accessibility groups MUST be reweighted for population estimates while remaining separately visible for fairness and coverage analysis. Country Observer observations MAY establish geographic observation coverage, but MUST NOT be represented as statistically reliable country scores. Participants MUST NOT be required to bypass product, geographic, age, identity, or account restrictions in order to enter the official population panel. Every population frame MUST be versioned, sourced, uncertainty-aware, and attributable to an accountable human or organisation.","character_count":66455,"record_sha256":"994da04173d3042d3ffd86bdfe279c00961d09ef3fc11ac405bdf15343b8c29c"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-06","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":8,"chapter_number":6,"item_number":null,"title":"The Population-Proportional GEO-1000 Sample","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K05","K07","K08"],"source_word_count":8195,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":6,\"id\":\"NOMOS-GEO-AUDIT-CH06\",\"title\":\"The Population-Proportional GEO-1000 Sample\",\"subtitle\":null,\"sourceFile\":\"6.ci bölüm.docx\",\"sourceSha256\":\"CF13DD3AF6C4AC35B7D35B8DC2C3096674BF4514B9FA0C4DFA526890C9C7414E\",\"sourceWordCount\":8195,\"sourceIds\":[\"K04\",\"K05\",\"K07\",\"K08\"],\"machine\":{\"chapter\":6,\"chapterId\":\"NOMOS-GEO-AUDIT-CH06\",\"title\":\"The Population-Proportional GEO-1000 Sample\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K05\",\"K07\",\"K08\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH06-R01\",\"normativeRuleEnglish\":\"For each AI product and principal measurement wave, the GEO-1000 Population Panel SHOULD target at least 1,000 valid product-observations unless a different target is explicitly justified. Sample allocation MUST be designed, versioned, and locked before AI responses are observed. Primary allocation MUST be based on the product-eligible target population, not raw resident population, invitation counts, accounts, sessions, or convenience availability. Fractional allocations MUST be converted to integer quotas using a predeclared, reproducible, outcome-blind method. Positive-mass populations MUST NOT be deterministically assigned zero selection probability when the audit claims inference to the full target population. Small populations MAY be represented through justified pooled strata and probability-proportional selection. Country Observer, Language Fairness, and Accessibility observations MAY supplement the main panel, but MUST remain separately labelled and appropriately reweighted. Oversampled groups MUST NOT receive their raw sample share as population weight. Participant replacement MUST be based only on predeclared recruitment or protocol-validity conditions. Incorrect, negative, refused, or critical AI outputs MUST NOT trigger replacement. Persons, accounts, sessions, languages, and AI-product observations MUST remain distinct units. Design weights, nonresponse adjustments, calibration, trimming, clustering, paired observations, and effective sample size MUST be documented. Any post-result change to quotas, strata, pools, weights, exclusions, or sample targets MUST create a new analysis version and preserve the original record.\",\"normativeRuleSourceTurkish\":\"Her AI ürünü ve ana ölçüm dalgası için GEO-1000 Population Panel, farklı bir hedef açıkça gerekçelendirilmedikçe en az 1.000 geçerli ürün-gözlemini hedeflemelidir. Örneklem tahsisi AI yanıtları görülmeden önce tasarlanmalı, sürümlenmeli ve kilitlenmelidir. Ana tahsis toplam ülke nüfusuna, davet sayısına, hesaplara, oturumlara veya kolaylık örneklemine değil ürün-uygun hedef insan nüfusuna dayanmalıdır. Küsuratlı kotalar önceden ilan edilmiş ve sonuçtan bağımsız yöntemle tam sayıya çevrilmelidir. Pozitif nüfus kütlesi taşıyan gruplara, bütün hedef nüfus adına çıkarım yapılırken deterministik sıfır seçilme olasılığı verilemez. Fazla örneklenen gruplar ham örneklem paylarını nüfus ağırlığı olarak kullanamaz. Yanlış, olumsuz, ret veya Critical AI çıktısı katılımcı değiştirme nedeni olamaz.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH06-MB0001\",\"type\":\"paragraph\",\"text\":\"58. MACHINE-READABLE ALLOCATION RECORD\",\"sourceParagraph\":1512},{\"blockId\":\"CH06-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1513},{\"blockId\":\"CH06-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1514},{\"blockId\":\"CH06-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"populationShare\\\": 0.002,\",\"sourceParagraph\":1515},{\"blockId\":\"CH06-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"rawQuota\\\": 2,\",\"sourceParagraph\":1516},{\"blockId\":\"CH06-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"targetValid\\\": 2,\",\"sourceParagraph\":1517},{\"blockId\":\"CH06-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"withinPoolSelection\\\": {\",\"sourceParagraph\":1518},{\"blockId\":\"CH06-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"method\\\": \\\"PPS_ELIGIBLE_POPULATION\\\",\",\"sourceParagraph\":1519},{\"blockId\":\"CH06-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"countries\\\": [\",\"sourceParagraph\":1520},{\"blockId\":\"CH06-MB0010\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1521},{\"blockId\":\"CH06-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-H\\\",\",\"sourceParagraph\":1522},{\"blockId\":\"CH06-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"eligiblePopulation\\\": 1500000,\",\"sourceParagraph\":1523},{\"blockId\":\"CH06-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"conditionalSelectionShare\\\": 0.75\",\"sourceParagraph\":1524},{\"blockId\":\"CH06-MB0014\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1525},{\"blockId\":\"CH06-MB0015\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1526},{\"blockId\":\"CH06-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-I\\\",\",\"sourceParagraph\":1527},{\"blockId\":\"CH06-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"eligiblePopulation\\\": 400000,\",\"sourceParagraph\":1528},{\"blockId\":\"CH06-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"conditionalSelectionShare\\\": 0.2\",\"sourceParagraph\":1529},{\"blockId\":\"CH06-MB0019\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1530},{\"blockId\":\"CH06-MB0020\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1531},{\"blockId\":\"CH06-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-J\\\",\",\"sourceParagraph\":1532},{\"blockId\":\"CH06-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"eligiblePopulation\\\": 100000,\",\"sourceParagraph\":1533},{\"blockId\":\"CH06-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"conditionalSelectionShare\\\": 0.05\",\"sourceParagraph\":1534},{\"blockId\":\"CH06-MB0024\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1535},{\"blockId\":\"CH06-MB0025\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":1536},{\"blockId\":\"CH06-MB0026\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1537},{\"blockId\":\"CH06-MB0027\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1538},{\"blockId\":\"CH06-MB0028\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":1539},{\"blockId\":\"CH06-MB0029\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1540},{\"blockId\":\"CH06-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"supplementaryPanels\\\": {\",\"sourceParagraph\":1541},{\"blockId\":\"CH06-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"countrySentinel\\\": {\",\"sourceParagraph\":1542},{\"blockId\":\"CH06-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"targetValid\\\": 3,\",\"sourceParagraph\":1543},{\"blockId\":\"CH06-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"includedCountries\\\": [\\\"SYN-H\\\", \\\"SYN-I\\\", \\\"SYN-J\\\"],\",\"sourceParagraph\":1544},{\"blockId\":\"CH06-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"countryScoresFromSingleObservationAllowed\\\": false,\",\"sourceParagraph\":1545},{\"blockId\":\"CH06-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"includedInPopulationEstimateWithFullSentinelWeight\\\": false\",\"sourceParagraph\":1546},{\"blockId\":\"CH06-MB0036\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1547},{\"blockId\":\"CH06-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"languageFairness\\\": {\",\"sourceParagraph\":1548},{\"blockId\":\"CH06-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"targetValid\\\": 50,\",\"sourceParagraph\":1549},{\"blockId\":\"CH06-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"oversampledLanguage\\\": \\\"SYN-LANG-Y\\\",\",\"sourceParagraph\":1550},{\"blockId\":\"CH06-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"reweightToPopulation\\\": true\",\"sourceParagraph\":1551},{\"blockId\":\"CH06-MB0041\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1552},{\"blockId\":\"CH06-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"accessibility\\\": {\",\"sourceParagraph\":1553},{\"blockId\":\"CH06-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"targetValid\\\": 40,\",\"sourceParagraph\":1554},{\"blockId\":\"CH06-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"separateDiagnosticReporting\\\": true\",\"sourceParagraph\":1555},{\"blockId\":\"CH06-MB0045\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1556},{\"blockId\":\"CH06-MB0046\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1557},{\"blockId\":\"CH06-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"fieldwork\\\": {\",\"sourceParagraph\":1558},{\"blockId\":\"CH06-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"expectedCompletionRate\\\": 0.82,\",\"sourceParagraph\":1559},{\"blockId\":\"CH06-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"expectedProtocolValidityRate\\\": 0.92,\",\"sourceParagraph\":1560},{\"blockId\":\"CH06-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"plannedInvitationsMainPanel\\\": 1326,\",\"sourceParagraph\":1561},{\"blockId\":\"CH06-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"reserveListsPreselected\\\": true,\",\"sourceParagraph\":1562},{\"blockId\":\"CH06-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"responseContentMayTriggerReplacement\\\": false,\",\"sourceParagraph\":1563},{\"blockId\":\"CH06-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"validityReviewBeforeSemanticScoring\\\": true\",\"sourceParagraph\":1564},{\"blockId\":\"CH06-MB0054\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1565},{\"blockId\":\"CH06-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"weights\\\": {\",\"sourceParagraph\":1566},{\"blockId\":\"CH06-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"baseDesignWeight\\\": \\\"INVERSE_SELECTION_PROBABILITY\\\",\",\"sourceParagraph\":1567},{\"blockId\":\"CH06-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"nonresponseAdjustment\\\": \\\"PREDECLARED_RESPONSE_CLASSES\\\",\",\"sourceParagraph\":1568},{\"blockId\":\"CH06-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"calibration\\\": \\\"COUNTRY_LANGUAGE_TOTALS\\\",\",\"sourceParagraph\":1569},{\"blockId\":\"CH06-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"trimming\\\": {\",\"sourceParagraph\":1570},{\"blockId\":\"CH06-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"enabled\\\": false,\",\"sourceParagraph\":1571},{\"blockId\":\"CH06-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"ifEnabledRequiresNewAnalysisVersion\\\": true\",\"sourceParagraph\":1572},{\"blockId\":\"CH06-MB0062\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1573},{\"blockId\":\"CH06-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"effectiveSampleSizeRequired\\\": true\",\"sourceParagraph\":1574},{\"blockId\":\"CH06-MB0064\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1575},{\"blockId\":\"CH06-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonFrames\\\": {\",\"sourceParagraph\":1576},{\"blockId\":\"CH06-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachAllocation\\\": true,\",\"sourceParagraph\":1577},{\"blockId\":\"CH06-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"commonSupportAllocation\\\": \\\"SEPARATE_RECORD_REQUIRED\\\"\",\"sourceParagraph\":1578},{\"blockId\":\"CH06-MB0068\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1579},{\"blockId\":\"CH06-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":1580},{\"blockId\":\"CH06-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"allocationApprovedBeforeResponses\\\": true,\",\"sourceParagraph\":1581},{\"blockId\":\"CH06-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"accountableHumanRole\\\": \\\"SAMPLE_DESIGN_OWNER\\\",\",\"sourceParagraph\":1582},{\"blockId\":\"CH06-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"lockedAt\\\": \\\"2026-08-18T09:00:00Z\\\",\",\"sourceParagraph\":1583},{\"blockId\":\"CH06-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"resultDrivenChangesProhibited\\\": true\",\"sourceParagraph\":1584},{\"blockId\":\"CH06-MB0074\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1585},{\"blockId\":\"CH06-MB0075\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1586},{\"blockId\":\"CH06-MB0076\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1587},{\"blockId\":\"CH06-MB0077\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1588},{\"blockId\":\"CH06-MB0078\",\"type\":\"paragraph\",\"text\":\"It is not actual product or country data,\",\"sourceParagraph\":1589},{\"blockId\":\"CH06-MB0079\",\"type\":\"paragraph\",\"text\":\"It is not an actual sampling plan.\",\"sourceParagraph\":1590},{\"blockId\":\"CH06-MB0080\",\"type\":\"paragraph\",\"text\":\"It is a synthetic representation of the reproducible allocation structure.\",\"sourceParagraph\":1591},{\"blockId\":\"CH06-MB0081\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RULE OF SECTION 59\",\"sourceParagraph\":1593},{\"blockId\":\"CH06-MB0082\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH06-R01\",\"sourceParagraph\":1594},{\"blockId\":\"CH06-MB0083\",\"type\":\"paragraph\",\"text\":\"For each AI product and principal measurement wave, the GEO-1000\",\"sourceParagraph\":1596},{\"blockId\":\"CH06-MB0084\",\"type\":\"paragraph\",\"text\":\"Population Panel SHOULD target at least 1,000 valid product-observations\",\"sourceParagraph\":1597},{\"blockId\":\"CH06-MB0085\",\"type\":\"paragraph\",\"text\":\"unless a different target is explicitly justified.\",\"sourceParagraph\":1598},{\"blockId\":\"CH06-MB0086\",\"type\":\"paragraph\",\"text\":\"Sample allocation MUST be designed, versioned, and locked before AI\",\"sourceParagraph\":1600},{\"blockId\":\"CH06-MB0087\",\"type\":\"paragraph\",\"text\":\"responses are observed.\",\"sourceParagraph\":1601},{\"blockId\":\"CH06-MB0088\",\"type\":\"paragraph\",\"text\":\"Primary allocation MUST be based on the product-eligible target population,\",\"sourceParagraph\":1603},{\"blockId\":\"CH06-MB0089\",\"type\":\"paragraph\",\"text\":\"not raw resident population, invitation counts, accounts, sessions, or\",\"sourceParagraph\":1604},{\"blockId\":\"CH06-MB0090\",\"type\":\"paragraph\",\"text\":\"convenience availability.\",\"sourceParagraph\":1605},{\"blockId\":\"CH06-MB0091\",\"type\":\"paragraph\",\"text\":\"Fractional allocations MUST be converted to integer quotas using a\",\"sourceParagraph\":1607},{\"blockId\":\"CH06-MB0092\",\"type\":\"paragraph\",\"text\":\"predeclared, reproducible, outcome-blind method.\",\"sourceParagraph\":1608},{\"blockId\":\"CH06-MB0093\",\"type\":\"paragraph\",\"text\":\"Positive-mass populations MUST NOT be deterministically assigned zero\",\"sourceParagraph\":1610},{\"blockId\":\"CH06-MB0094\",\"type\":\"paragraph\",\"text\":\"selection probability when the audit claims inference to the full target\",\"sourceParagraph\":1611},{\"blockId\":\"CH06-MB0095\",\"type\":\"paragraph\",\"text\":\"population. Small populations MAY be represented through justified pooled\",\"sourceParagraph\":1612},{\"blockId\":\"CH06-MB0096\",\"type\":\"paragraph\",\"text\":\"strata and probability-proportional selection.\",\"sourceParagraph\":1613},{\"blockId\":\"CH06-MB0097\",\"type\":\"paragraph\",\"text\":\"Country Observer, Language Fairness, and Accessibility observations MAY\",\"sourceParagraph\":1615},{\"blockId\":\"CH06-MB0098\",\"type\":\"paragraph\",\"text\":\"supplement the main panel, but MUST remain separately labelled and\",\"sourceParagraph\":1616},{\"blockId\":\"CH06-MB0099\",\"type\":\"paragraph\",\"text\":\"appropriately reweighted.\",\"sourceParagraph\":1617},{\"blockId\":\"CH06-MB0100\",\"type\":\"paragraph\",\"text\":\"Oversampled groups MUST NOT receive their raw sample share as population\",\"sourceParagraph\":1619},{\"blockId\":\"CH06-MB0101\",\"type\":\"paragraph\",\"text\":\"weight.\",\"sourceParagraph\":1620},{\"blockId\":\"CH06-MB0102\",\"type\":\"paragraph\",\"text\":\"Participant replacement MUST be based only on predeclared recruitment or\",\"sourceParagraph\":1622},{\"blockId\":\"CH06-MB0103\",\"type\":\"paragraph\",\"text\":\"protocol-validity conditions. Incorrect, negative, refused, or critical AI\",\"sourceParagraph\":1623},{\"blockId\":\"CH06-MB0104\",\"type\":\"paragraph\",\"text\":\"outputs MUST NOT trigger replacement.\",\"sourceParagraph\":1624},{\"blockId\":\"CH06-MB0105\",\"type\":\"paragraph\",\"text\":\"Persons, accounts, sessions, languages, and AI-product observations MUST\",\"sourceParagraph\":1626},{\"blockId\":\"CH06-MB0106\",\"type\":\"paragraph\",\"text\":\"remain distinct units.\",\"sourceParagraph\":1627},{\"blockId\":\"CH06-MB0107\",\"type\":\"paragraph\",\"text\":\"Design weights, nonresponse adjustments, calibration, trimming, clustering,\",\"sourceParagraph\":1629},{\"blockId\":\"CH06-MB0108\",\"type\":\"paragraph\",\"text\":\"paired observations, and effective sample size MUST be documented.\",\"sourceParagraph\":1630},{\"blockId\":\"CH06-MB0109\",\"type\":\"paragraph\",\"text\":\"Any post-result change to quotas, strata, pools, weights, exclusions, or\",\"sourceParagraph\":1632},{\"blockId\":\"CH06-MB0110\",\"type\":\"paragraph\",\"text\":\"sample targets MUST create a new analysis version and preserve the original\",\"sourceParagraph\":1633},{\"blockId\":\"CH06-MB0111\",\"type\":\"paragraph\",\"text\":\"record.\",\"sourceParagraph\":1634},{\"blockId\":\"CH06-MB0112\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1635},{\"blockId\":\"CH06-MB0113\",\"type\":\"paragraph\",\"text\":\"For each AI product and main measurement wave, the GEO-1000 Population Panel should target at least 1,000 valid product-observations unless a different target is explicitly justified. Sample allocation should be designed, finalised, and locked before AI responses are seen. The main allocation should be based on the product-eligible target human population, not on total country population, number of invitations, accounts, sessions, or convenience samples. Fractional quotas should be announced in advance and rounded to whole numbers using a method independent of the results. Groups with a positive population mass cannot be assigned a deterministic zero probability of selection when making inferences for the entire target population. Over-sampled groups cannot use raw sample shares as population weights. Incorrect, negative, rejected, or Critical AI output cannot be a reason for participant replacement.\",\"sourceParagraph\":1636}]}}","text":"## Chapter Boundary\n\nSection 5 established the following provision:\n\n> The target user population of each AI product must be defined separately. The total population of a country is not the product-eligible user population in that country.\n\nNow we come to the next question:\n\n> From the defined eligible user population for GEO-1000, how many people will be selected, from which country, which language, and by which method?\n\nThe eligible user population for an AI product is:\n\n- 40% in country A,\n\n- 25% in country B,\n\n- 15% in country C,\n\n- 0.04% in a small country\n\nIf available, how should 1,000 valid observations be distributed? How will fractional quotas be converted into whole numbers? Will countries with a population share of less than one person be completely excluded? Will population weighting be distorted when at least one person is given to each country? Can small countries be grouped into a separate pool? If some languages are oversampled for language fairness, how will the global score be calculated? If the target is 100 valid observations in a country, how many people should be invited? Who will replace the participant if they do not complete the task? Can a user giving wrong answers be replaced? If the same person tests more than one AI product, will the sample still be independent?\n\nEven with a raw sample of 1,000 people, can highly unequal weights reduce the effective statistical information below that of 1,000 independent, equally weighted observations? These are not merely operational details. The sample design determines how much influence each person has on the GEO score. A defective sample:\n\n- can misweight the correct answers,\n\n- can symbolically overrepresent small countries,\n\n- can make large populations appear smaller than they are,\n\n- can make low-resource languages invisible,\n\n- can present voluntary technology users as if they were the world population.\n\nIt can be adjusted to achieve the desired score after reviewing the results. The previously established audit architecture requires that each audit be measurable and that the model/product carry the date, country, language, query set, repetition count, and measurement record. This section applies the same discipline to sample design. Because genuine evidence requires requesting limits, context, and time; it also necessitates making how the sample was created visible. This section:\n\n- how the 1,000 valid observations of the main population panel will be distributed,\n\n- how strata and cells will be established,\n\n- how small countries will be managed,\n\n- how fractions will be converted to whole numbers,\n\n- how oversampling will be weighted,\n\n- how to separate invitation, backup, and valid observation targets,\n\n- how to establish design and analysis weights,\n\n- why the effective sample size may differ from the raw number\n\ndefines. This chapter does not yet:\n\n- from which panel provider the participants will be selected,\n\n- the full details of field implementation in voluntary or probability sampling,\n\n- reviewing and scoring responses,\n\n- the final confidence interval method,\n\n- all NOMOS scoring formulas\n\nare not fully finalised. The task of Section 6 is:\n\n> Transform the defined target population into a frozen, weightable, and reproducible GEO-1000 sample before seeing the outcome.\n\n## NOMOS Challenge\n\nYou are setting up a global panel of 1,000 people for an AI product. The product-eligible populations of ten countries are as follows:\n\nRaw quotas according to pure population proportion:\n\nHalf of a person cannot be selected. 0.4 people cannot be sent to a country. If you round the raw quotas to the nearest whole number, the total may exceed or fall short of 1,000. If you give at least one person to each country, you over-represent countries I and J compared to their population shares.\n\nGiving one person to country I increases its theoretical share 2.5 times. Giving one person to country J increases its theoretical share 10 times. This extreme representation can largely be corrected. But you cannot produce a country score with a single person. What happens if you give no one to I and J?\n\nIf you define these countries as separate and mandatory strata, the probability of some people in the target population being selected becomes zero. In this case, the claim of a “probabilistic sample representing the entire appropriate global population” weakens. Now consider a more dangerous decision. First, you collect the answers.\n\nThe only user in country J gets an incorrect answer. Then you say: “The sample size of this country is very small, let's exclude it from the scope.” In country A, there are many positive answers. You retain it with full weight. This is no longer rounding. It is a result-oriented sample change.\n\nIn another case, you increase the invalid record rate in a language group with low results, using backup participants only until a positive response is obtained. This is also not a sample. It is a response selection mechanism. The first ruling of this section is:\n\n> The sample must be established before seeing AI responses.\n\nIts second provision states:\n\n> The probability of a participant being selected cannot depend on the correctness of their answer.\n\nIts third provision states:\n\n> The visibility of the small country and the population weight in the global score must be managed separately.\n\nIts fourth provision states:\n\n> Oversampling is possible; over-weighting is not mandatory.\n\nIts fifth provision states:\n\n> 1,000 raw records do not always carry statistical information equivalent to 1,000 observations.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to transform the targeted 1,000 valid observations for each AI product and measurement wave into a verifiable sample design based on the product-eligible user population. The section standardises the following distinctions:\n\n- Target population versus selected sample\n\n- Raw quota versus integer allocation\n\n- Main population panel versus complementary panels\n\n- Country visibility versus country weight\n\n- Strata versus experimental cell\n\n- Large country strata versus small country pool\n\n- Proportional allocation vs. disproportionate oversampling\n\n- Sample share vs. population share\n\n- Number of invitations vs. valid observation target\n\n- Backup participant vs. post-result replacement\n\n- Data collection invalidity vs. adverse AI output\n\n- Design weight vs. outcome weight\n\n- Raw sample size vs. effective sample size\n\n- Country sample vs. language sample\n\n- Multilingual user vs. more than one person\n\n- Single-stage selection vs. multi-stage selection\n\n- Native Reach allocation vs. Common Support allocation\n\n- Instant wave allocation and time series allocation\n\n- Population estimation and country or language recognition analysis\n\n- Main 1,000 observations with auxiliary observers and fairness panels\n\nAt the end of this section, each audit should be able to answer these questions:\n\n> Why were 1,000 observations distributed in these numbers across this country, language, and user groups?\n\n> How was the weight of each observation in the global result determined?\n\n> Who had a zero probability of being selected and why?\n\n> How many people were invited, how many completed, and how many observations remained valid?\n\n> Were the sample results frozen before being viewed?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nThe GEO-1000 main population sample should be based on a stratified sampling design that assigns known or predictable probabilities of selection to people in the pre-frozen product-eligible population frame for each AI product and measurement wave. The default target of the main population panel is:\n\n> At least 1,000 valid observations per AI product and main measurement wave\n\nis maintained. This number:\n\n- is not the number of people invited,\n\n- is not the number of accounts,\n\n- is not the number of raw screenshots,\n\nIt is not the sum of all supplementary panels. 1,000 is the target for valid product-observations meeting the protocol in the main population panel. Country Observer, Language Fairness, and Accessibility panels can create additional samples. These additional samples:\n\n- can increase the total field volume above 1,000,\n\n- can be used with appropriate weights in the global population outcome,\n\nbut cannot be used to covertly reduce the 1,000 valid observation target of the main panel.\n\n## 3. BASIC REPRESENTATION OF THE SAMPLE\n\nFor a specific AI product a, measurement wave w, and target population frame, the strata:\n\nh ∈ H_{a,w}\n\nare denoted. For each stratum:\n\n- Nh: estimated product-appropriate population in the stratum\n\n- N: total target population\n\n- Wh: population weight of the stratum\n\n- n: target valid main panel observation\n\n- qh: crude proportional quota\n\n- nh: allocated integer valid observation target\n\nis. Total target population:\n\nN = Σ_{h=1}^H N_h\n\nPopulation weight of the stratum:\n\nW_h = N_h / N\n\nCrude proportional quota:\n\nq_h = nW_h\n\nis. Main proportional allocation, under ideal conditions:\n\nn_h ≈ nW_h\n\nshould be. However:\n\n- qh may be fractional,\n\n- some cells may fall below one person,\n\n- a minimum diagnostic sample may be required for language and country subgroups,\n\n- high-risk small groups may also be examined,\n\nfield completion rates may vary. Therefore, the raw quota is not directly the final allocation.\n\n## 4. THREE DISTINCT PURPOSES OF SAMPLE DESIGN\n\nThe same sample may not fulfil all purposes with the same strength. Three purposes should be separated.\n\n### 4.1. Global Population Estimate\n\nAnswers the question:\n\n> What representative outcome would the average person in the global user population suitable for the product see?\n\nPopulation weight is the primary priority.\n\n### 4.2. Subgroup Diagnosis\n\nAnswers the question:\n\n> What representation distortions occur in specific countries, languages, plans, or user groups?\n\nIt may be necessary to oversample small groups.\n\n### 4.3. Geographic and Linguistic Coverage\n\nAnswers the question:\n\n> In which countries and languages is there at least real user observation?\n\nCountry Observer and Language Fairness panels become important. These three goals cannot be inexplicably combined in a single sample.\n\n## 5. MAIN SAMPLE ARCHITECTURE\n\nThe candidate main sample architecture for GEO-1000 consists of four layers.\n\n### 5.1. Population Panel\n\nEvery AI product and wave has a minimum of 1,000 valid observations as the main panel. It produces a global population estimate.\n\n### 5.2. Country Observer Overlay\n\nProvides additional geographic observations for small countries that cannot be directly represented in the main population panel. These observations can indicate:\n\n- single event,\n\n- access status,\n\n- future research needs\n\nThey do not generate a country score on their own.\n\n### 5.3. Language Fairness Oversample\n\nGenerates additional observations for low-resource languages or languages underrepresented in the main panel. They are reweighted to actual population weights in the global outcome.\n\n### 5.4. Accessibility Oversample\n\nIt also measures users of assistive technology and accessibility, who are not adequately represented by the standard panel. The target universe and weight of these observations are recorded separately.\n\n## 6. WHAT IS A STRATUM?\n\nA stratum is a segment of the population that is managed separately in terms of sample allocation and selection probability. The stratum can be a combination of one or more of the following areas:\n\n- Country\n\n- Region\n\n- Language\n\n- Locale\n\n- AI plan\n\n- Web or mobile\n\n- Controlled or natural user panel\n\n- Accessibility status\n\n- User type\n\n- Jurisdiction\n\nIf each small combination is made a separate stratum, the sample is fragmented. For example: Country × language × plan × mobile/web × age group × user type combination can produce hundreds or thousands of cells. 1,000 observations do not provide reliable estimates for most of these cells. Therefore, stratification should be:\n\n- substantial,\n\n- measurable,\n\n- appropriate for sample size\n\nmust be.\n\n## 7. A STRATUM IS NOT THE SAME AS A REPORTING CELL\n\nIt can be shown separately on a group result card. However, it does not have to be a separate stratum in sample design. For example:\n\n- a small language,\n\n- a specific accessibility group,\n\n- a certain age band\n\nIt can be selected within a broader stratum in the main sample. Later, it can be reported separately in the analysis. The reverse is also possible: Two regions may be separately stratified for sampling efficiency. However, in the public report, they can be combined under a single regional result. The stratification design and the reporting structure should carry separate records.\n\n## 8. SEQUENCE OF STRATUM ESTABLISHMENT\n\nThe candidate sequence for the Main Population Panel is as follows:\n\n- AI product and user surface\n\n- Native Reach or Common Support population frame\n\n- Country or small country pool\n\n- Main audit language or language group\n\n- Material plan/interface separation, if necessary\n\n- Participant selection frame\n\nThe plan and interface do not have to be made into separate layers for each country. These can be managed through:\n\n- sub-layer,\n\n- quota,\n\n- random assignment\n\nmethods.\n\n## 9. LARGE AND SMALL COUNTRY LAYERS\n\nA country's raw quota:\n\nq_c = nN_c / N\n\nis in this form. If the raw quota is large enough, the country can be kept directly as a separate layer. This can be called:\n\n#### self-representing country layer\n\nMaking countries with very small raw quotas individually into separate mandatory layers may reduce sampling efficiency. These countries can be appropriately:\n\n- regional,\n\n- language family,\n\n- product access regime,\n\n- geographical proximity\n\ncan be included in small country pools.\n\n## 10. SMALL COUNTRY POOL\n\nA small country pool:\n\nLet it be represented as B_r = {c_1, c_2, …, c_k}\n\nThe total eligible population of the pool:\n\nN_{B_r} = Σ_{c∈B_r} N_c\n\nMain sample quota:\n\nq_{B_r} = nN_{B_r} / N\n\nis calculated as. Within the pool, country selection can be done with probability proportional to the eligible population size. The probability of country c being selected within the pool:\n\nPr(c | B_r) = N_c / N_{B_r}\n\nIt may be. Then a participant is selected within the chosen country. This approach:\n\n- can prevent small countries from having a zero chance of being selected in the target population,\n\n- does not guarantee that each country will be selected in a given wave,\n\npreserves the global population estimate. Not selecting a country in that wave:\n\n### NOT SELECTED IN POPULATION SAMPLE\n\ndoes not mean: \"The country is not part of the target population.\"\n\n## 11. LIMITATIONS OF POOLING SMALL COUNTRIES\n\nSmall countries cannot be randomly placed in the same pool just because their populations are small. Countries within the pool are similar to the extent that they are relevant:\n\n- product access,\n\n- user surface,\n\n- language or locale structure,\n\n- legal access,\n\n- geographical region,\n\n- data collection method\n\nmust be carried. A country without product access cannot be placed in the same main pool as a country with full access. Different usage conditions cannot be made invisible.\n\n## 12. SMALL COUNTRY POOL DOES NOT PRODUCE COUNTRY SCORE\n\nOne or two observations selected from the pool: can be used in the global population contribution of the pool, may indicate a single country event. However, it does not produce a separate statistical score for a small country. If a country-based score is wanted, additionally:\n\n- Country Observer,\n\n- country-specific sample,\n\n- multi-wave cumulative sample\n\nmay be required.\n\n## 13. ALLOCATION PROPORTIONAL TO PURE POPULATION\n\nPure proportional allocation:\n\nis based on the formula:\n\nq_h = nW_h\n\nAdvantages:\n\n- it is understandable for the main population estimate,\n\n- it naturally gives weight to large populations,\n\n- it can reduce weight differences,\n\nit can approach a self-weighting design. Limitations:\n\n- it can sample very few small countries and languages,\n\n- it may be insufficient to produce subgroup scores,\n\n- it may leave high-risk small groups invisible,\n\nit can create fractional and near-zero quotas. Therefore, pure proportional allocation can be a strong default for the main Population Panel. It does not replace complementary panels.\n\n## 14. CONVERTING FRACTIONS TO WHOLE NUMBERS\n\nRaw quotas are often fractional. The total allocation should be exactly n. The Largest Remainder Method can be used as the default method.\n\n### 14.1. First Step\n\nEach raw quota is rounded down:\n\nb_h = ⌊q_h⌋\n\n### 14.2. Second Step\n\nThe number of unallocated observations:\n\nR = n − Σ_h b_h\n\nis calculated.\n\n### 14.3. Third Step\n\nThe fractional part of each stratum:\n\nf_h = q_h − b_h\n\nwhere f_h is the fractional part of q_h.\n\n### 14.4. Fourth Step\n\nThe remaining R observations are added one by one to the layers with the highest fh values.\n\n### 14.5. Equality Situation\n\nIf the fractional parts of the two layers are equal:\n\n- population size,\n\n- predetermined fixed order,\n\n- registered random seed\n\nAn independent tie-breaking rule should be used, regardless of the outcome. A country cannot be chosen based on the results.\n\n## 15. SYNTHETIC INTEGER ALLOCATION\n\n### SYNTHETIC DISPLAY\n\nAppropriate populations at the beginning of the section:\n\nIf countries are protected as separate layers, round down:\n\nOne observation remains. It is added to H because the highest fraction is 0.5 in country H. Final allocation:\n\nEven though this design defines I and J as separate deterministic strata, if it does not give a chance to be selected, it may pose a problem in terms of full target universe inference. A more accurate solution is a small country pool.\n\n## 16. SYNTHETIC ALLOCATION WITH SMALL COUNTRY POOL\n\nCountries H, I, and J:\n\nB_small = {H, I, J}\n\nshould be combined in the pool. Pool population:\n\n1,500,000+400,000+100,000=2,000,000\n\nPool quota:\n\n1,000 × 2,000,000 / 1,000,000,000 = 2\n\nbecomes. Main allocation:\n\nSelection weights of countries within the pool:\n\nTwo main population observations are selected according to these probabilities. In the same wave:\n\n- both observations can come from H,\n\n- one from H and one from I,\n\n- rarely from J\n\nThis can occur. The design does not make J visible in every wave, but it does not reduce J's probability of selection to zero. If J must be observed, it is added separately to the Country Observer Panel. A Country Observer observation does not give J a weight of one in 1,000 in the main global allocation; J retains its true population weight in the main panel.\n\n## 17. MINIMUM CELL ASSIGNMENT\n\nPurely proportional quota may not be sufficient to perform diagnostic analysis for some subgroups. For example, the raw quota for a language cell might be 8. A reliable language score cannot be produced with eight observations. In this case, there are three options:\n\n- Not to publish the language score\n\n- Combine the cell with similar groups\n\n- Oversampling the cell\n\nMinimum cell allocation must be defined before the results.\n\n## 18. CANDIDATE REPORTING BANDS\n\n### CANDIDATE REPORTING BANDS — CANDIDATE REPORTING BANDS\n\nThe thresholds below are not final scientific decisions. They should be calibrated through pilot and external review.\n\n#### 1–29 Valid Observation\n\nSingle event and discovery Country or language rate is not published\n\n### OBSERVATIONAL ONLY\n\n#### 30–99 Valid Observations\n\nExploratory level estimate Broad uncertainty No definitive conformity judgement is given\n\n### EXPLORATORY\n\n#### 100–199 Valid Observations\n\nFirst subgroup estimate Uncertainty should be visible\n\n### PROVISIONAL\n\n#### 200 and Above Valid Observations\n\nStronger subgroup estimate candidate Design effect and weights are again evaluated\n\n### STANDARD ESTIMATION CANDIDATE\n\nThese bands:\n\n- error rate,\n\n- expected precision,\n\n- weight variability,\n\n- design effect\n\ncannot be automatically considered sufficient without being taken into account.\n\n## 19. OVERSAMPLING\n\nA country, language, or user group may have more observations than its population share. This is called oversampling. Its purposes:\n\n- to increase subgroup sensitivity,\n\n- to make a low-resource language visible,\n\n- to examine a high-risk small group,\n\n- to expand Country Observer coverage,\n\n- to adequately represent accessibility users\n\nare possible. Oversampling is not wrong. What is wrong:\n\n> It is to use the sample proportion of the over-sampled group as if it were the true population proportion.\n\n## 20. POPULATION SHARE AND SAMPLE SHARE\n\nPopulation share for layer h:\n\nW_h = N_h / N\n\nSample share:\n\nS_h = n_h / n\n\nis in the form. In pure proportional allocation:\n\nS_h ≈ W_h\n\nokay. In excessive sampling:\n\nS_h > W_h\n\nIt is possible. In a small sample:\n\nS_h < W_h\n\nIt is possible. The global population estimate should retain Wh values instead of Sh.\n\n## 21. BASIC DESIGN LOAD\n\nIn simple stratified sampling, the basic weight of each observation within stratum h can be considered as:\n\nd_i = N_h / n_h\n\nThe normalised stratum weight can be written as:\n\na_h = W_h / S_h\n\nThe normalised observation-level weight can be:\n\nw_i = W_h / n_h\n\nIt is ensured that the sum equals one:\n\nΣ_i w_i = 1\n\nThe weighted binary transition estimate can be calculated as:\n\nθ̂ = Σ_i w_iY_i\n\nHere:\n\nYi=1: response meeting the transition criterion\n\nYi=0: response not meeting it\n\nIt is possible. The final NOMOS system does not have to use only binary outcomes. The same weighting logic can also be applied to multidimensional response scores.\n\n## 22. EXCESS SAMPLING EXAMPLE\n\n### SYNTHETIC EXAMPLE\n\nIn the suitable population of country B:\n\n- Language X: 80 per cent\n\n- Language Y: 20 per cent\n\nLet it be. In the Population Panel, there are 250 observations for country B. Pure allocation:\n\n- Language X: 200\n\n- Language Y: 50\n\nLet 50 additional Language Fairness observations be taken for Language Y. Collected observations:\n\n- Language X: 200\n\n- Language Y: 100\n\nThe share of Language Y in the raw sample:\n\n100/300 = 33.3%\n\nIt has been. The actual population share is 20%. In the global population account, Language Y cannot take a 33.3% weight. Within-stratum normalised weights:\n\n- Total weight for Language X: 80%\n\n- Total weight for Language Y: 20%\n\nis preserved. Each of the 100 observations in Language Y receives a lower individual weight than a single observation in Language X. In contrast, all 100 observations are used in Language Y’s separate diagnostic report. This approach: increases subgroup visibility, does not distort the global population estimate.\n\n## 23. ALLOCATION OF MULTILINGUAL USERS\n\nA person may be suitable for more than one language cell. One of the preferred approaches for the main Population Panel is to assign a predefined main audit language to the user. The main audit language can be determined according to these fields:\n\n- Daily AI usage language\n\n- Mother or home language\n\n- Market and locale\n\n- Competence in naturally using the prompt\n\n- Panel balance\n\nIf the user will also be tested in other languages, these observations:\n\n- paired multilingual panel,\n\n- Language Fairness Panel\n\nare also labelled accordingly. It is prohibited for the same person to receive full population weighting in all languages.\n\n## 24. LANGUAGE ALLOCATION WITHIN THE COUNTRY\n\nLet the suitable population for language groups l in country c be: N_{c,l}. If the country quota is: n_c, then the raw language quota:\n\nq_{c,l} = n_c × N_{c,l} / Σ_l N_{c,l}\n\ncan be calculated as such. However, if language groups overlap:\n\nΣ_l N_{c,l} > N_c\n\nmay be possible. In this case, using a direct share creates double counting. Solution options:\n\n- main audit language classification,\n\n- split human weight,\n\n- multilingual common model,\n\n- separate matched language panel\n\nmay be used. The method used should be recorded.\n\n## 25. ALLOCATION OF PLAN AND INTERFACE WITHIN THE POPULATION\n\nFree and paid users or web and mobile surfaces may behave differently. However, dividing each country–language cell separately into plan and interface layers can overly fragment the sample. Three approaches can be used.\n\n### 25.1. Suballocation Proportional to Population\n\nIf the actual user distribution is known, allocation is made according to plan and interface shares.\n\n### 25.2. Balanced Experimental Allocation\n\nEqual or greater sampling is carried out to compare the difference between plan or interface. In global estimation, it is re-weighted to the actual usage shares.\n\n### 25.3. Separate Product Surface Scores\n\nFree web, paid web, and mobile products are reported as separate audit objects. A combined result can also be produced. The approach to be used should be defined before measurement.\n\n## 26. INVITATION NUMBER AND VALID OBSERVATION TARGET\n\nIf the valid observation target for a stratum is nh, inviting exactly nh people is usually not sufficient. Because:\n\n- some people do not participate,\n\n- some do not complete the task,\n\n- some records remain out of protocol,\n\nsome evidence packages become invalid. Let the expected completion rate for the stratum be rh and the expected validity rate be vh. The invitation target is approximately:\n\nm_h = ⌈n_h / (r_h v_h)⌉\n\ncan be calculated as follows. Example:\n\n- Valid observation target: 100\n\n- Expected completion rate: 80 per cent\n\n- Expected validity rate: 90 per cent\n\nm_h = ⌈100 / (0.80 × 0.90)⌉ = 139\n\npeople can be invited. This is not a guarantee. As the field progresses, the predefined backup system can be used.\n\n## 27. LIMIT OF INVITATION INFLATION\n\nInviting more people alone does not solve sample bias. Early responders may be systematically different from late responders. Only taking the first 1,000 completers:\n\n- time zone,\n\n- technology interest,\n\n- connection quality,\n\n- work schedule\n\nmay produce bias in terms of. Invitation and selection order:\n\n- should be randomised,\n\n- must be recorded within the stratum,\n\nshould be managed throughout the field window.\n\n## 28. ALTERNATE PARTICIPANT SYSTEM\n\nAn ordered backup list should be created for each stratum before the results are seen. A backup participant can be activated only for the following reasons:\n\n- No response to the invitation\n\n- Failure to complete the task\n\n- Technical or ethical suitability issue\n\n- Violation of the demand protocol\n\n- Invalidity of the evidence package\n\n- Re-registration of the same person\n\nThe backup user cannot intervene for the following reason:\n\n> The previous user's AI response gave an incorrect, negative, or unwanted output.\n\nAn incorrect AI response is the valid system result. It cannot be a reason for participant change.\n\n## 29. RESULT BLINDNESS IN VALIDITY DECISION\n\nThe person deciding whether the observation complies with the protocol as much as possible:\n\n- must evaluate the response without knowing whether it is correct or incorrect,\n\n- or whether it is positive or negative for the brand.\n\nFirst:\n\n- Capture and protocol validity\n\n- Then the substantive response evaluation\n\nmust be done. Otherwise, negative responses can more easily be declared: \"invalid.\"\n\n## 30. RESERVE ORDER WITHIN LAYER\n\nThe reserve participant should be from the same as much as possible:\n\n- country,\n\n- language,\n\n- product surface,\n\n- plan,\n\n- panel condition\n\nlayer. If no reserve remains in the same layer, predefined escalation is applied:\n\n- Additional random selection within the same layer\n\n- Predefined upper layer pool\n\n- Leaving allocation incomplete and scope warning\n\n- New analysis version\n\nAfter the result is seen, bringing a user from another country and closing the gap cannot be done silently.\n\n## 31. WHAT HAPPENS IF A LAYER IS TOO SHORT?\n\nIt may fall below a layer target. Example:\n\n- target: 100\n\n- valid observation: 72\n\nCorrect options:\n\n- Extend the field window to a predefined extent\n\n- Use the backup list\n\n- Merge the cell with a similar and predefined upper layer\n\n- Report the result with wide uncertainty\n\n- Do not publish the cell result\n\n- Recalculate the global weight using an appropriate method\n\nWrong option: Completely remove the low-scoring layer from the analysis.\n\n## 32. POST-RESULT QUOTA CHANGE\n\nThe following changes are result-oriented sample manipulations:\n\n- Reducing the target quota of the low-scoring country\n\n- Increasing the quota of the high-scoring language\n\n- Invalidating users who provide negative responses\n\n- Redefining the small country pool based on results\n\n- Excluding a plan with critical errors\n\n- Adding new users until the desired average is reached\n\nEach material quota change:\n\n- new allocation version,\n\n- justification,\n\n- date,\n\n- decision maker\n\n- Preservation of the previous result\n\nshould be done with.\n\n## 33. MULTI-STAGE SAMPLING\n\nMulti-stage sampling can be used when direct access to all individuals is not possible. Example:\n\n- Selection of region or country pool\n\n- Country selection\n\n- Selection of panel or local research partner\n\n- User selection\n\n- AI product or task assignment\n\nThe total probability of selection for an individual user is the product of the conditional probabilities at each stage:\n\nπ_i = π_r × π_{c|r} × π_{f|c} × π_{i|f}\n\nHere:\n\n- r: region or small country pool\n\n- c: country\n\n- f: sampling frame or panel\n\n- i: person\n\nBasic design weight:\n\nd_i = 1/π_i\n\nIf selection probabilities cannot be recorded, the design cannot be presented as a fully probability-based sample. [K04; K07]\n\n## 34. ALLOCATION IN VOLUNTARY AND NON-PROBABILITY PANELS\n\nA true probability-based world panel may not always be possible. A voluntary or professional online panel can be used. In this case, country and language quotas can still be set according to the product-eligible population. However:\n\n- the participant's actual probability of selection may be unknown,\n\n- self-selection may occur,\n\nweighting can only make adjustments based on observed variables. The correct status:\n\n#### Non-probability user panel stratified and calibrated to the population\n\nmay be. The following statement cannot be used: “1,000 people selected completely at random from the world population.”\n\n## 35. WEIGHTING ARCHITECTURE\n\nThe final observation weight may consist of multiple components:\n\nw_i = d_i × a_i^{NR} × g_i^{CAL}\n\nHere:\n\n- di: base design weight\n\n- aiNR: non-response adjustment\n\n- giCAL: calibration coefficient\n\n### 35.1. Design Weight\n\nIt is the inverse of the participant's probability of selection in the sampling design.\n\n### 35.2. Nonresponse Adjustment\n\nFor selected persons with similar characteristics:\n\n- how many responded,\n\n- how many produced a valid observation\n\ncan be taken into account. For sample response class r:\n\na_r^{NR} = (Σ_{j∈S_r} d_j) / (Σ_{i∈R_r} d_i)\n\nHere:\n\n- Sr: selected eligible sample\n\n- Rr: those who responded or produced a valid observation\n\nmay be. Nonresponse adjustment: does not completely resolve unobserved bias, should be explained along with its assumptions.\n\n### 35.3. Calibration Weight\n\nThe sample, to known population totals:\n\n- country,\n\n- language,\n\n- age,\n\n- plan,\n\n- user surface.\n\nCalibration may be applied across these dimensions, but only where reliable external totals exist.\n\n## 36. WEIGHT NORMALISATION\n\nWeights:\n\n- to the total population,\n\n- to the total sample size,\n\n- to individual\n\ncan be normalised. The normalisation used in the score formula should be explained. Individually normalised weights:\n\nw̃_i = w_i / Σ_j w_j\n\nare obtained. Weighted estimate:\n\nθ̂ = Σ_i w̃_iY_i\n\nis calculated as follows.\n\n## 37. WEIGHT TRIMMING\n\nSome participants may receive very high weights because they represent very small and under-sampled groups. This situation:\n\n- can increase the estimated variance,\n\n- the influence of a single observation,\n\n- and score volatility\n\ncan increase. Weights can be trimmed at an upper limit. However, trimming: may create bias in population representation, may make small groups invisible again. Therefore, weight trimming should:\n\n- be defined before the results,\n\n- be shown together with the untrimmed results,\n\nand carry a sensitivity analysis.\n\n## 38. EFFECTIVE SAMPLE SIZE\n\nEven though the raw number of observations is n, if the weights vary greatly, the actual amount of information may be lower. Kish effective sample size:\n\nn_eff = (Σ_i w_i)^2 / Σ_i w_i^2 [K08]\n\ncan be shown as. With equal weights:\n\nneff≈n\n\nis obtained. If the weights are very unbalanced: neff<n. a GEO-1000 study:\n\n- 1,000 valid observations,\n\n- may carry 620 effective sample size\n\n. To the public only: saying “1,000 people were measured.” can appear strong due to the precision. In the main report, at least:\n\n- raw valid observation count,\n\n- effective sample size,\n\n- weight variability\n\nshould be determined.\n\n## 39. WEIGHT DESIGN EFFECT\n\nApproximate design effect due to weights:\n\nDEFF_w ≈ 1 + CV(w)^2\n\ncan be shown. Here:\n\nCV(w) = sd(w) / mean(w)\n\nThis is the coefficient of variation of the weights and only an approximate representation of their effect. The total design effect may also vary with:\n\n- clustering,\n\n- stratification,\n\n- repeated measurement,\n\n- paired users\n\nThese effects must be estimated separately.\n\n## 40. CLUSTERING\n\nThe same:\n\n- From the research panel,\n\n- from the country,\n\n- from the language community,\n\n- from the device or institution,\n\n- from the user\n\nobservations may resemble each other. In this case, 1,000 observations are not 1,000 completely independent units of information. Clustering:\n\n- in sample design,\n\n- in variance calculation,\n\n- in effective sample size\n\nshould be taken into account.\n\n## 41. TESTING MORE THAN ONE AI PRODUCT BY THE SAME USER\n\nHaving a user test multiple AI products allows the difference between models to be measured on the same person. Advantage: user characteristics can be controlled, paired comparison can be strengthened. Limitation:\n\n- product observations are not independent,\n\n- the order of tasks can affect other responses,\n\n- a user may carry the answer seen in the first system to the next system,\n\nusers who can access all products may differ from the population. Therefore, products:\n\n- should be assigned in random order,\n\n- previous answer sharing should be prevented,\n\n- the paired design should be recorded,\n\nAnalysis should protect dependency. Every AI product should again have its own target of 1,000 valid product-observations. If the same 1,000 humans test ten products:\n\n- 1,000 unique humans,\n\n- 10,000 product-observations\n\ncan be formed. These two numbers are not the same.\n\n## 42. BALANCED MISSING BLOCK DESIGN\n\nNot all users need to access all AI products. For example, if there are ten AI products, each participant can be assigned three products. If assignments are made in a balanced way:\n\n- each product 1,000 observations,\n\n- each product pair sufficient shared users,\n\n- broader population participation\n\ncan be achieved. As a candidate for this approach:\n\n#### Balanced incomplete block design\n\ncan be used. Its advantages:\n\n- reduces participant burden,\n\n- lowers the risk of selecting only heavy AI users,\n\nprovides matched comparison across models. The exact design will be detailed in Sections 8 and 9.\n\n## 43. NATIVE REACH ALLOCATION\n\nEach AI product receives a separate allocation according to its own Native Reach population. Therefore, the same country:\n\n- 200 for one product,\n\n- 0 for another product,\n\n- 50 for a third product\n\ncan be observed. This difference is not unfairness. It is that the real access universes of the products are different. Since Native Reach scores belong to different populations, they should not be used alone for direct ranking.\n\n## 44. COMMON SUPPORT ALLOCATION\n\nA common access universe is created for comparison between products. The common population:\n\nU_CS = ⋂_{a∈A} U_a\n\ncan be considered in this way. Each product is measured with the same Common Support population weights. This approach:\n\n- provides fair product comparison,\n\n- same country and language floor\n\nensures. However, it does not show the natural global access difference of the products. Therefore, Native Reach and Common Support results should be reported together.\n\n## 45. MANAGEMENT OF THE SAMPLE ACROSS WAVES\n\nThere may be two different objectives across multiple measurement waves.\n\n### 45.1. Current Cross-Section Result\n\nIn each wave, a current product-eligible population frame is used. This score answers the question: \"What is happening now in the actual global access universe?\"\n\n### 45.2. Fixed Frame Trend Result\n\nThe population weights from the initial wave are preserved. This score answers the question: \"How has representation in the same reference population changed over time?\"\n\n### 45.3. Separation of Two Results\n\nWhen the product expands to new countries, the current score may decrease. This decrease:\n\n- is due to the deterioration of the product,\n\n- from the inclusion of new and difficult populations\n\nIt may be attributable. A trend with a fixed frame can show improvement at the same time. Therefore:\n\n- Current-Frame Snapshot\n\n- Fixed-Frame Trend\n\nThey are separate results.\n\n## 46. PANEL RENEWAL AND TRANSFORMATION\n\nAmong the waves:\n\n- the same users,\n\n- completely new users,\n\n- partially renewed panel\n\nusable.\n\n### 46.1. Full Section Panel\n\nNew users are selected in each wave. It can be strong for population estimation. It does not measure individual change.\n\n### 46.2. Longitudinal Panel\n\nThe same users are measured again. Shows individual user and product changes. The panel may create wear and learning effects.\n\n### 46.3. Rotating Panel\n\nA portion of the participants is retained, while a portion is renewed. Provides both trend and fresh population representation. The panel structure used should be visible on the result card.\n\n## 47. SAMPLE LOCK\n\nThe following records should be locked before each wave starts:\n\n- AI product\n\n- User surface\n\n- Target population version\n\n- Native Reach countries\n\n- Common Support countries\n\n- Strata\n\n- Small country pools\n\n- Raw quotas\n\n- Integer allocations\n\n- Language allocations\n\n- Oversampling plan\n\n- Invitation targets\n\n- Reserve lists\n\n- Selection method\n\n- Weight formula\n\n- Result blindness rule\n\n- Shortfall escalation\n\n- Analysis plan\n\n- Decoding method\n\n- Random seed, if used\n\n- Responsible human approval\n\nTo this file:\n\n#### NOMOS Sample Allocation Lock\n\ncan be given the name.\n\n## 48. CHANGES ALLOWED AFTER SAMPLE LOCK\n\nThe following events may require a new version:\n\n- Termination of product access in the country\n\n- Serious field safety issue\n\n- Material error in population data\n\n- Panel provider unable to deliver to a specific country\n\n- Change in research ethics requirement\n\n- Critical error in measurement tool\n\nChange:\n\n- must be justified,\n\n- dated,\n\n- approved,\n\n- carried out while preserving the old version.\n\nThe strength or weakness of the results is not a reason for change.\n\n## 49. SAMPLE QUALITY STATUSES\n\nA sample wave can carry one of the following statuses:\n\n### SQ-0 — NOT DESIGNED\n\nThere is no documented sample design.\n\n### SQ-1 — CONVENIENCE OBSERVATION\n\nThere is convenience or voluntary observation. Population generalisation cannot be made.\n\n### SQ-2 — QUOTA-CONTROLLED\n\nThere are country and language quotas. Selection probabilities are not fully known.\n\n### SQ-3 — WEIGHTED PANEL\n\nThere is a stratified panel and population calibration. Coverage and self-selection limitations persist.\n\n### SQ-4 — PROBABILITY-BASED OR HYBRID\n\nKnown election probabilities or strong hybrid design exist.\n\n### SQ-5 — REPLICATED AND CALIBRATED\n\nThe design has been replicated in different waves and external applications; bias and variance behaviour have been calibrated. These statuses may affect the confidence class of the public score.\n\n## 50. SYNTHETIC END-TO-END GEO-1000 ALLOCATION\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The following countries, populations, languages, AI product, and calculations are entirely fictional. For the Synthetic Orion AI product, let the suitable population for Native Reach be 1 billion people. Main Population Panel objective:\n\nn=1,000\n\n### 50.1. Country and Small Country Pool Allocation\n\n### 50.2. Language Allocation in Country B\n\nEligible population in Country B:\n\n- Language X: 80 per cent\n\n- Language Y: 20 per cent\n\nPopulation Panel allocation:\n\n- Language X: 200\n\n- Language Y: 50\n\nLanguage Fairness additional allocation: Language Y: 50 Field target in Country B:\n\n- Language X: 200\n\n- Language Y: 100\n\n- Total: 300\n\nHowever, global Population Panel population weights:\n\n- Language X: 80% of Country B\n\n- Language Y: 20% of Country B\n\nremain as such.\n\n### 50.3. Country Observer Appendix\n\nThere are two probability observations for H, I, and J in the small country pool in the Population Panel. The Country Observer Panel also:\n\n### H: 1\n\n### I: 1\n\n### J: 1\n\ncan target observations. These three observations:\n\n- increase the geographical coverage,\n\n- show individual country events,\n\ndo not create a country score.\n\n### 50.4. Accessibility Supplement\n\nIf the main panel is found to inadequately cover accessibility users, 40 additional observations can be taken. These users: are calibrated according to relevant population groups, produce a separate Accessibility Panel result.\n\n### 50.5. Total Field Volume\n\nThe name GEO-1000 is still valid. Because:\n\n> 1,000 is the valid observation target for the main population panel.\n\nComplementary justice and coverage panels can be added to this.\n\n### 50.6. Invitation Volume\n\nOn the Population Panel on average:\n\n- completion: 82 per cent\n\n- validity: 92 per cent\n\nlet it be expected. Required invitation:\n\n⌈1,000 / (0.82 × 0.92)⌉ = 1,326\n\nmay be. With complementary panels, total invitations may exceed 1,500. Therefore:\n\n> GEO-1000 is not only a method to send invitations to 1,000 people.\n\n### 50.7. Field Result\n\nOn the Population Panel:\n\n- 1,326 people have been invited,\n\n- 1,080 tasks completed,\n\n- 1,012 records have been found valid in terms of protocol,\n\n- the first 1,000 were included in the main analysis according to a predefined random/backup order\n\nAlright. The remaining 12 valid records:\n\n- backup verification,\n\n- sensitivity analysis,\n\n- recheck\n\nIt can be stored for. It cannot be decided which 1,000 will be chosen by looking at the results.\n\n## 51. NORMATIVE ALLOCATION ALGORITHM\n\nThe default sequence of operations for a GEO-1000 main panel is as follows:\n\n### Step 1 — Lock the Product and Wave ID\n\nDetermine which AI product, surface, and date are measured.\n\n### Step 2 — Lock the Target Population Version\n\nThe product-eligible population frame from section 5 is used.\n\n### Step 3 — Define the Main Current Observation Target\n\nDefault candidate target:\n\nn=1,000\n\n### Step 4 — Define Material Layers\n\nCountry, language, and if necessary, plan/interface layers are defined.\n\n### Step 5 — Set Up Small Country Pools\n\nCountries that will generate near-zero allocations are included in appropriate pools.\n\n### Step 6 — Calculate Raw Proportional Quotas\n\nqh=nWh\n\n### Step 7 — Perform Integer Allocation\n\nLargest Remainder or pre-declared equivalent method is applied.\n\n### Step 8 — Add Oversampling Plan\n\nCountry Observer, Language Fairness, and Accessibility supplements are saved separately.\n\n### Step 9 — Calculate Invitation and Backup Numbers\n\nExpected completion and validity rates are used.\n\n### Step 10 — Randomise Selection Order\n\nPrimary and backup lists are frozen.\n\n### Step 11 — Save Weight Formula\n\nDesign, response, and calibration weights are defined.\n\n### Step 12 — Confirm Sample Lock\n\nA version and integrity record is created before AI responses are viewed.\n\n## 52. MANDATORY NORMATIVE PROVISIONS\n\n**CH06-N01**\n\nThe GEO-1000 sample allocation must be created and versioned before AI responses are viewed.\n\n**CH06-N02**\n\nThe main Population Panel target should be defined as at least 1,000 valid observations per AI product and main measurement wave; if a different target is used, the rationale must be explained.\n\n**CH06-N03**\n\nThe invited person, the person completing the task, and the valid observation numbers must be distinguished from each other.\n\n**CH06-N04**\n\nThe main allocation should be based on the product-eligible population defined in Section 5.\n\n**CH06-N05**\n\nCountry or language quotas should be calculated based on the share of the relevant product-eligible population, not the total population.\n\n**CH06-N06**\n\nThe method for converting raw fractional quotas into whole numbers must be defined in advance.\n\n**CH06-N07**\n\nThe resolution of rounding ties must be independent of the results.\n\n**CH06-N08**\n\nSmall countries with a positive target population should not be left as separate zero-probability deterministic strata; an appropriate pooling or probabilistic selection method should be used.\n\n**CH06-N09**\n\nSmall country pools should be justified in terms of product access and financial population characteristics.\n\n**CH06-N10**\n\nCountry Observer observations should also be recorded without changing the main population weight.\n\n**CH06-N11**\n\nA Country Observer observation cannot be presented as a country score.\n\n**CH06-N12**\n\nOver-sampled countries, languages, or user groups should be returned to the true target population weights in global estimation.\n\n**CH06-N13**\n\nThe sample proportion cannot be used like the population proportion.\n\n**CH06-N14**\n\nThe same person cannot be duplicated in the global human population due to multiple languages, accounts, sessions, or products.\n\n**CH06-N15**\n\nInvalid observations should be classified without basing the reason for invalidity on the content of the results.\n\n**CH06-N16**\n\nIncorrect, negative, rejection, or Critical AI responses cannot be the reason for changing a participant.\n\n**CH06-N17**\n\nBackup participant lists must be created within the stratum before the results are seen.\n\n**CH06-N18**\n\nLayer gaps cannot be quietly closed by another country or language.\n\n**CH06-N19**\n\nPost-result quota, layer, small country pool, or weight changes require a new analysis version.\n\n**CH06-N20**\n\nAll components of the weights and the method of normalisation must be published.\n\n**CH06-N21**\n\nThe assumptions of the nonresponse adjustment must be explained.\n\n**CH06-N22**\n\nIf weight trimming is used, the untrimmed result and sensitivity analysis must be retained.\n\n**CH06-N23**\n\nThe main report should show the appropriate effective sample size alongside the raw number of observations.\n\n**CH06-N24**\n\nClustering, repetition, and paired user dependencies should be taken into account in the calculation of variance.\n\n**CH06-N25**\n\nThe design in which the same person tests multiple AI products cannot be presented as a completely independent user sample.\n\n**CH06-N26**\n\nEach AI product must separately reach its own target of 1,000 valid product observations.\n\n**CH06-N27**\n\nAllocations Native Reach and Common Support must hold separate sample records.\n\n**CH06-N28**\n\nCurrent and fixed-frame results among the waves cannot be combined without explanation.\n\n**CH06-N29**\n\nIf the sample is insufficient for a subgroup, the correct status should be INSUFFICIENT SAMPLE, EXPLORATORY, or NOT ESTIMATED.\n\n**CH06-N30**\n\nThe raw sample size cannot be used as a definitive precision indicator, ignoring weights and design effect.\n\n**CH06-N31**\n\nDesign weights cannot be claimed in a volunteer or non-probability panel as if selection probabilities are known.\n\n**CH06-N32**\n\nThe sample cannot be modified post hoc to exclude low-scoring groups or to enlarge high-scoring groups.\n\n**CH06-N33**\n\nObservations from the main Population Panel and complementary panels should be labelled separately in terms of purpose and weight.\n\n**CH06-N34**\n\nThe sample lock must have an accountable human or institutional owner.\n\n## 53. FORMS OF FAILURE\n\n**CH06-F01 — SET QUOTA AFTER RESULT**\n\nThe number of countries and languages is determined after viewing AI responses.\n\n**CH06-F02 — USE BACKUP UNTIL BEST ANSWER**\n\nNegative responses are replaced with a new user even if valid.\n\n**CH06-F03 — COUNT 1,000 INVITATIONS AS 1,000 VALID OBSERVATIONS**\n\nNon-completion and invalidity are hidden.\n\n**CH06-F04 — RESET COUNTRY TO ZERO WITH PURE ROUNDING**\n\nThe selection probability of a country with a positive target population is left at zero.\n\n**CH06-F05 — GIVING ONE PERSON TO EACH COUNTRY AND CALLING IT THE POPULATION RATIO**\n\nThe disproportion caused by the minimum quota is hidden.\n\n**CH06-F06 — CONSIDERING Country Observer observation AS POPULATION OBSERVATION**\n\nThe additional observation of a small country receives full weight in the main panel of 1,000 people.\n\n**CH06-F07 — ESTABLISHING THE SMALL COUNTRY POOL BASED ON RESULTS**\n\nA country producing low scores is moved to another pool or excluded from coverage.\n\n**CH06-F08 — PUTTING INCOMPATIBLE COUNTRIES INTO THE SAME POOL**\n\nCountries that differ in product access, language, or jurisdiction are combined solely because they are small.\n\n**CH06-F09 — OVERWEIGHTING EXCESSIVE SAMPLING**\n\nDetermines the global result with a sample share higher than the language or country population share.\n\n**CH06-F10 — MULTIPLYING MULTILINGUAL INDIVIDUALS**\n\nA person is counted as a separate population unit in each language cell.\n\n**CH06-F11 — CONSIDERING SAMPLE SHARE AS POPULATION SHARE**\n\nRaw response numbers are converted directly into the global score.\n\n**CH06-F12 — ARRANGING INVITATION ORDER BY EASE ORDER**\n\nUsers who respond first and quickly determine the panel.\n\n**CH06-F13 — USING STRATUM-OUT RESERVES**\n\nMissing country or language observations are quietly completed with participants from another group.\n\n**CH06-F14 — LINKING INVALIDITY TO RESPONSE CONTENT**\n\nIncorrect AI response is technically declared invalid.\n\n**CH06-F15 — REPLACING REFUSALS WITH A NEW USER**\n\nSystem refusal is removed from the denominator even though it is the actual user outcome.\n\n**CH06-F16 — HIDING THE WEIGHT FORMULA**\n\nThe public score does not show with which population and sample weights it was produced.\n\n**CH06-F17 — TRIMMING WEIGHTS TO DELETE SMALL GROUPS**\n\nSmall groups with high weights are trimmed because the result does not look good.\n\n**CH06-F18 — HIDING THE EFFECTIVE SAMPLE**\n\nEven if 1,000 raw records drop to 500–600 data points due to weights, it is presented as “1,000-person precision”.\n\n**CH06-F19 — CONSIDERING CLUSTERING AS INDEPENDENCE**\n\nObservations from the same panel, institution, or user are considered completely independent.\n\n**CH06-F20 — CONSIDERING THE SAME 1,000 PEOPLE AS TEN THOUSAND INDEPENDENT PEOPLE**\n\nTen observations of an AI product are described as if the population were ten times larger.\n\n**CH06-F21 — CONSIDERING COMMON SUPPORT AS NATIVE REACH**\n\nThe common comparison sample is presented as if it were the product’s actual global reach population.\n\n**CH06-F22 — DIRECTLY RANKING NATIVE REACH RESULTS**\n\nProduct scores in different target populations are compared as if they were from the same universe.\n\n**CH06-F23 — HIDING POPULATION CHANGE IN WAVES**\n\nEven though new countries or user groups enter the score, the trend is interpreted only as a change in product quality.\n\n**CH06-F24 — PRODUCING DEFINITE SCORES FROM INSUFFICIENT SAMPLES FOR A SUBGROUP**\n\nA few dozen observations are published as definite scores for a country or language.\n\n**CH06-F25 — SILENTLY CHANGING THE SAMPLING LOCK**\n\nQuota, weight, or pool definitions are updated without a version record.\n\n**CH06-F26 — BLINDLY MULTIPLYING ELIGIBILITY RATES**\n\nEligible population and quota calculations ignore the characteristics of the dependent population.\n\n**CH06-F27 — REPLACING THE MAIN PANEL WITH THE COMPLEMENTARY PANEL**\n\nLanguage fairness or Country Observer observations are used to reduce the main population target of 1,000.\n\n**CH06-F28 — DELETING LOW-SCORED CELLS WHEN THE PANEL IS SHORT**\n\nIncomplete samples are excluded because they are disadvantageous in terms of results.\n\n**CH06-F29 — USING WEIGHTING AS SCIENTIFIC FLAIR**\n\nIt is claimed that the representation problem is solved only by adding weight without explaining selection and coverage bias.\n\n**CH06-F30 — THINKING THAT GEO-1000 SHOULD BE ASKED TO EXACTLY 1,000 PEOPLE**\n\nInvitation, complementary panels, invalidity, repetition, and effective sample structure are ignored.\n\n## 54. AUDIT PROCEDURE\n\n### Step 1 — Record the Sample Purpose\n\nGlobal population estimate is divided into sub-group diagnosis, country scope, language fairness, accessibility, and product comparison purposes.\n\n### Step 2 — Validate Target Population Version\n\nThe population file and date record in Section 5 are reviewed.\n\n### Step 3 — Define the Main Current Observation Target\n\nThe default GEO-1000 target or a justified alternative is written.\n\n### Step 4 — Establish Strata\n\nDifferences in country, language, and material user surface are determined.\n\n### Step 5 — Identify Small Country Pools\n\nEach pool's:\n\n- countries,\n\n- population,\n\n- justification for similarity,\n\n- internal selection method\n\nis recorded.\n\n### Step 6 — Calculate Raw Quotas\n\nqh=nWh\n\nThe formula or a justified alternative is used.\n\n### Step 7 — Apply Integer Assignment\n\nRounding and tie-breaking rules are recorded.\n\n### Step 8 — Add Complementary Oversampling\n\nCountry Observer, Language Fairness, and Accessibility goals are defined separately.\n\n### Step 9 — Calculate Invite and Reserve Volume\n\nCompletion and validity expectations are used.\n\n### Step 10 — Freeze Selection and Reserve Order\n\nParticipants are selected or ordered before the results are seen.\n\n### Step 11 — Calculate Design Weights\n\nSelection probabilities and layer weights are recorded.\n\n### Step 12 — Define the Procedure for Staying Silent and Not Responding\n\nA previously announced response is written to the layer gap.\n\n### Step 13 — Create the Sample Lock\n\nDocument ID, version, date, and integrity record are added.\n\n### Step 14 — Review the Post-Field Validity Flow\n\nIt is verified that validity decisions are given independently of the accuracy of the responses.\n\n### Step 15 — Generate the Final Weights\n\nThe design is separated into response and calibration components.\n\n### Step 16 — Calculate the Effective Sample Size\n\nThe difference between the raw number and the volume of information is reported.\n\n### Step 17 — Compare Allocation and Actuals\n\nPlanned and actual observation numbers are shown on a cell-by-cell basis.\n\n### Step 18 — Review Post-Result Changes\n\nQuota, pool, weight, and scope changes are checked.\n\n## 55. REQUIRED EVIDENCE\n\nAI-product identity; measurement wave; target-population version; Population Panel target; Country Observer target; Language Fairness target; Accessibility Panel target; stratum list; stratum populations; population weights; raw quotas; rounding method; fractional-allocation and remainder records; tie-breaking rule; small-country pools; within-pool selection probabilities; country and language allocations; plan and interface allocations; oversampling justifications; invitation targets; completion assumptions; validity assumptions; primary and reserve participant lists; selection or randomisation record; nonresponse records; reasons for invalid observations; validity-decision dates; design weights; nonresponse adjustments; calibration coefficients; weight-trimming policy; raw and effective sample sizes; clustering and matched-user records; Native Reach allocation; and Common Support allocation.\n\nFixed and current frame weights Comparison of planned and actual allocation Sample lock Change log Responsible person or institution\n\n## 56. AUDIT CHECKLIST\n\nIs the main valid observation target open? Is the target 1,000 invites or valid observations? Was the sample frozen before AI responses were seen? Which population frame was used? On what basis were the strata established? Are raw country and language quotas shown? Is the method for converting fractions to whole numbers clear? Was the break-off rule predetermined? Is the selection probability for small countries zero? Are small country pools logically consistent? Have Country Observer observations been mixed into the main population weight? Has a country score been produced from a Country Observer observation? If language and accessibility groups were oversampled, were they reweighted? Has the same person been counted as more than one population unit? Are the numbers of invites, completers, and valid observations listed separately?\n\nHave backup lists been created in advance? Has a participant been replaced due to an incorrect or negative AI response? Was the validity decision made before the material accuracy score? Was the tab gap quietly closed from another country? Has the quota or pool changed after the results? Can the design weights be reproduced? Are the assumptions of the nonresponse setting clear? Has weight trimming been done and its effect shown? Are raw and effective sample sizes provided together? Has the same user tested more than one AI product? Has this dependency been preserved in the analysis? Has each AI product reached 1,000 product-observations separately? Were the Native Reach and Common Support samples separated? Did the population frame change between waves?\n\nHave fixed and current framed trends been separated? Have insufficient subgroup samples been published as if they were definitive scores? Is the sample quality status correct? Have complementary panels replaced the main panel? Is there a clear accountable owner of the sample registration?\n\n## 57. OBJECTIONS AND RESPONSES\n\n### Objection 1 — \"Why don't we give each country individuals directly according to the population ratio?\"\n\nThis is a strong assumption in large and medium population countries. In very small countries, the raw quota may fall below one person. In this case:\n\n- small country pool,\n\n- probabilistic country selection,\n\n- separate Observer panel\n\nmay be more accurate.\n\n### Objection 2 — “Wouldn't giving at least one person to each country be fairer?”\n\nIt might be useful in terms of geographic coverage. It is disproportionate in terms of population estimation. The correct solution is to separately label this one person as a Country Observer observation.\n\n### Objection 3 — “Doesn't a country receiving zero people on the main panel completely exclude it?”\n\nIt can be a problem if a country receives a zero quota as a separate deterministic layer. A positive selection probability can be given in the small country pool. Additionally, the Observer Panel can maintain the observational visibility of the country.\n\n### Objection 4 — “Does pooling small countries erase cultural differences?”\n\nIt can be deleted if used to generate a country score. For the main global estimate, the pool manages the probability of selection. The country identifier is preserved in the observation record. Separate Observer and country-specific sampling can be done.\n\n### Objection 5 — \"Why don't we include Observer users within the 1,000?\"\n\nThey can be included. But then the sample of 1,000 people is no longer proportionate to the pure population. Weighting would be needed, and the effective sample size of the main estimate could decrease. In the founding architecture, 1,000 observations are allocated to the Population Panel; Country Observer observations are added to this.\n\n### Objection 6 — \"Won't the total field cost become very high?\"\n\nIt could. The standard should not produce artificial precision by hiding the cost. With a lower budget:\n\n- fewer AI products,\n\n- fewer languages,\n\n- smaller pilot,\n\n- wider uncertainty\n\ncan be used. The correct status should be clearly written.\n\n### Objection 7 — “Why are we collecting 1,093 or more observations when the name GEO-1000 exists?”\n\n1,000 is the valid observation target for the main population panel. Justice and coverage panels can be added to the main target. This does not compromise the name of the standard. It makes the actual total field volume of the measurement visible.\n\n### Objection 8 — “Why would oversampling a language distort the global result?”\n\nAn oversampled language appears large because it is in raw responses. When returned to the population weight, the global result is not distorted. The separate language result becomes more precise.\n\n### Objection 9 — 'The weights are very complex; isn't the raw average easier to understand?'\n\nThe raw mean can be meaningful if it is proportional to the sample population. Excessive sampling, non-response, or coverage differences can cause the raw mean to answer the wrong question. Complexity cannot be hidden if it arises from the sample design.\n\n### Objection 10 — “If there are 1,000 valid observations, why should the effective sample be smaller?”\n\nBecause some observations can carry very high weight, while others can carry very low weight. Records from the same panel or user may resemble each other. The raw number is the count of physical records. The effective sample approaches the volume of statistical information.\n\n### Objection 11 — “Wouldn't it be fairer for the same people to test all AI products?”\n\nReduces user difference. However:\n\n- it can narrow the product access universe,\n\n- it can increase task load,\n\nit can create contamination between responses. It can be managed with matched or balanced incomplete block design.\n\n### Objection 12 — “If a layer does not reach the target, can the global score never be published?”\n\nNot in every case. The deficiency is evaluated in terms of:\n\n- population weight,\n\n- risk,\n\n- the reason for inadequacy,\n\n- remaining coverage\n\nconsidered. The score:\n\n- with narrowed coverage,\n\n- with wide uncertainty,\n\n- With the warning INSUFFICIENT COVERAGE\n\nit can be published. The deficiency cannot be hidden.\n\n### Appeal 13 — “If population data changes, do we have to renew all quotas in every wave?”\n\nRenewal may be required for the current instantaneous score. Fixed starting weights can also be maintained for trend comparison. The two results should be published separately.\n\n### Appeal 14 — “If we cannot establish a probabilistic world panel, can NOMOS not be applied?”\n\nIt is applicable. However, the sample quality status must be correctly recorded. A panel of volunteers calibrated to the population can be valuable. It cannot be presented as a fully probabilistic sample.\n\n## COMMON PROVISION OF CHAPTER 60\n\nThe power of GEO-1000 does not arise solely from the number 1,000. Of 1,000 people:\n\n- from which population they came from,\n\n- with what probability they were selected,\n\n- which country and language they represented,\n\n- how many times the same person was counted,\n\n- which observations were considered valid,\n\n- how small countries were managed,\n\n- how over-sampled groups were weighted,\n\n- how many people did not complete the task,\n\n- who was replaced based on the results\n\nA sample's authority arises from these design choices. A defective sample may produce 100,000 responses and still speak about the wrong population. A sound sample may contain fewer responses yet support a more reliable conclusion within clearly stated limits. A large country's population should carry substantial weight in the global human experience, but users must not be fabricated where the product is unavailable. A small country's global weight should remain small, but the country need not be invisible. A language may account for a small share of the global population while still carrying a severe, measurable error. One person may use three languages and five products; that does not create fifteen people. An AI response may be wrong without making the participant invalid. The wrong response is what the audit is trying to measure.\n\nReplacing it with a new user does not clean the measurement. It erases reality. Weighting does not miraculously turn the wrong panel into the world population. However, properly designed oversampling can make subgroups visible without distorting the global estimate. Therefore, NOMOS's sixth measurement law is:\n\n> A sample should be established not to produce the result, but to observe the result fairly.\n\nIts seventh law is as follows:\n\n> The visibility of a small country does not have to enlarge its population weight.\n\nIts eighth law is as follows:\n\n> Excess observation does not mean excess population weight.\n\nThe ninth law is as follows:\n\n> A wrong AI answer is not data defect; if observed under the valid protocol, it is a matter of oversight.\n\nThe tenth law is:\n\n> The confidence in a thousand observations does not come from the number thousand; it comes from the reproducibility of the selection and weighting.\n\n## Order of Chapter 6 of NOMOS\n\n> Don't just tell me, 'We asked 1,000 people.'\n\n> Show from which billion people these 1,000 were selected and with what probability.\n\n> Don't ignore a country just because it accounts for half a person. / But also, don't claim you measured the whole country just because you sampled one person.\n\n> Select small countries probabilistically in the appropriate pool. / If visibility is needed, set up a separate observer panel.\n\n> You can oversample low-resource languages. / But do not make it larger in the global score than its population.\n\n> Do not multiply a person as much as accounts, languages, sessions, and AI products.\n\n> Use the backup when the participant does not respond. / Do not change your user when AI gives a wrong answer.\n\n> Decide on validity without seeing the correctness of your answer.\n\n> If the quota is missing, explain it. / Do not bring a person from another country to cover the gap.\n\n> Do not change the share, weight, country pool, or sample target after seeing the result.\n\n> Show the raw number. / Show the effective sample. / Show who received heavy and who received light weight.\n\n> Lock the population first. / Then set up the layers. / Then calculate the quotas. / Then select the backups. / Then write the weights. / And only after this, send the first prompt.\n\n## The Chapter's Closing Sentence\n\n> GEO-1000 is not a method of compiling a thousand answers; it is a protocol for making the representation right of every person in the target population measurable under the probability of selection and correct weighting.\n\n## Normative Core\n\n> For each AI product and principal measurement wave, the GEO-1000 Population Panel SHOULD target at least 1,000 valid product-observations unless a different target is explicitly justified. Sample allocation MUST be designed, versioned, and locked before AI responses are observed. Primary allocation MUST be based on the product-eligible target population, not raw resident population, invitation counts, accounts, sessions, or convenience availability. Fractional allocations MUST be converted to integer quotas using a predeclared, reproducible, outcome-blind method. Positive-mass populations MUST NOT be deterministically assigned zero selection probability when the audit claims inference to the full target population. Small populations MAY be represented through justified pooled strata and probability-proportional selection. Country Observer, Language Fairness, and Accessibility observations MAY supplement the main panel, but MUST remain separately labelled and appropriately reweighted. Oversampled groups MUST NOT receive their raw sample share as population weight. Participant replacement MUST be based only on predeclared recruitment or protocol-validity conditions. Incorrect, negative, refused, or critical AI outputs MUST NOT trigger replacement. Persons, accounts, sessions, languages, and AI-product observations MUST remain distinct units. Design weights, nonresponse adjustments, calibration, trimming, clustering, paired observations, and effective sample size MUST be documented. Any post-result change to quotas, strata, pools, weights, exclusions, or sample targets MUST create a new analysis version and preserve the original record.","character_count":67432,"record_sha256":"3212ca67e9750165ad504be658cb5c63b24c53bc081aa2eb592e1299f5ae41f4"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-07","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":9,"chapter_number":7,"item_number":null,"title":"Country Observer and Language Fairness Panels","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K06","K07","K20"],"source_word_count":7656,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":7,\"id\":\"NOMOS-GEO-AUDIT-CH07\",\"title\":\"Country Observer and Language Fairness Panels\",\"subtitle\":null,\"sourceFile\":\"7.cisayfa.docx\",\"sourceSha256\":\"EC9225E812D181E9E9FCE4711EA0E163406CA86E7800DDFFD5B7D52E8C4D18F4\",\"sourceWordCount\":7656,\"sourceIds\":[\"K04\",\"K06\",\"K07\",\"K20\"],\"machine\":{\"chapter\":7,\"chapterId\":\"NOMOS-GEO-AUDIT-CH07\",\"title\":\"Country Observer and Language Fairness Panels\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K06\",\"K07\",\"K20\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH07-R01\",\"normativeRuleEnglish\":\"The GEO-1000 Population Panel, Country Observer Panel, and Language Fairness Panel MUST remain distinct in purpose, sampling, weighting, and public claims. A Country Observer observation MAY establish that an access condition, response, anomaly, or critical incident occurred in a defined country-product-language context. A Country Observer observation MUST NOT be represented as a country prevalence estimate or country GEO score. Country observation coverage, population-weighted observation coverage, and country estimation coverage MUST be reported separately. Every eligible country SHOULD receive a positive probability of observation within a predeclared surveillance cycle, unless it is explicitly outside the product's lawful Native Reach. Language fairness MUST evaluate materially comparable representation under semantically equivalent prompts and comparable contexts. It MUST NOT require word-for-word identical responses or ignore legitimate local differences. Oversampled languages MUST be reweighted to their target population shares for global population estimates. Population-weighted scores, equal-language diagnostic scores, language floors, critical-error rates, unknown rates, and untested languages MUST remain separately visible. A language disparity MUST NOT be attributed solely to the AI product before prompt equivalence, reference parity, measurement parity, and adjudication parity are assessed. A single observation MAY carry multiple diagnostic panel tags, but MUST NOT receive duplicate full population weight. Critical Observer incidents MAY trigger confirmatory sampling and conformity review without being treated as evidence of country prevalence.\",\"normativeRuleSourceTurkish\":\"GEO-1000 Population Panel, Ülke Gözcü Paneli ve Dil Adaleti Paneli amaç, örneklem, ağırlık ve kamu iddiası bakımından ayrı tutulmalıdır. Tek bir ülke gözcü gözlemi belirli ülke–ürün–dil bağlamında bir erişim durumunun, cevabın, anomalinin veya Critical olayın meydana geldiğini gösterebilir; ülke yaygınlık tahmini veya ülke GEO skoru olarak sunulamaz. Ülke sayısı kapsamı, nüfus ağırlıklı kapsam ve ülke tahmin kapsamı ayrı raporlanmalıdır. Dil adaleti, semantik olarak eşdeğer promptlarda ve karşılaştırılabilir koşullarda maddi temsili değerlendirmeli; kelime kelime aynı cevap istememeli ve meşru yerel farkları yok saymamalıdır. Fazla örneklenen diller global nüfus tahmininde gerçek hedef nüfus paylarına geri ağırlıklandırılmalıdır.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH07-MB0001\",\"type\":\"paragraph\",\"text\":\"47. MACHINE-READABLE PANEL RECORD\",\"sourceParagraph\":1460},{\"blockId\":\"CH07-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1461},{\"blockId\":\"CH07-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1462},{\"blockId\":\"CH07-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"SF-3\\\": 17\",\"sourceParagraph\":1463},{\"blockId\":\"CH07-MB0005\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1464},{\"blockId\":\"CH07-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"metrics\\\": {\",\"sourceParagraph\":1465},{\"blockId\":\"CH07-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"countryObservationCoverageCount\\\": 0.3167,\",\"sourceParagraph\":1466},{\"blockId\":\"CH07-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"countryObservationCoveragePopulation\\\": 0.97,\",\"sourceParagraph\":1467},{\"blockId\":\"CH07-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"countryEstimationCoverage\\\": 0.1167\",\"sourceParagraph\":1468},{\"blockId\":\"CH07-MB0010\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1469},{\"blockId\":\"CH07-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"sentinelObservationsCreateCountryScores\\\": false,\",\"sourceParagraph\":1470},{\"blockId\":\"CH07-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"alarms\\\": [\",\"sourceParagraph\":1471},{\"blockId\":\"CH07-MB0013\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1472},{\"blockId\":\"CH07-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"countryCode\\\": \\\"SYN-J\\\",\",\"sourceParagraph\":1473},{\"blockId\":\"CH07-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"alarmType\\\": \\\"SA-5\\\",\",\"sourceParagraph\":1474},{\"blockId\":\"CH07-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"CONFIRMATORY_SAMPLE_REQUIRED\\\",\",\"sourceParagraph\":1475},{\"blockId\":\"CH07-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"countryPrevalenceClaimAllowed\\\": false\",\"sourceParagraph\":1476},{\"blockId\":\"CH07-MB0018\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1477},{\"blockId\":\"CH07-MB0019\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":1478},{\"blockId\":\"CH07-MB0020\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1479},{\"blockId\":\"CH07-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"languageFairness\\\": {\",\"sourceParagraph\":1480},{\"blockId\":\"CH07-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"languageRegistryVersion\\\": \\\"LF-SYNTH-001\\\",\",\"sourceParagraph\":1481},{\"blockId\":\"CH07-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"inScopeLanguages\\\": [\",\"sourceParagraph\":1482},{\"blockId\":\"CH07-MB0024\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1483},{\"blockId\":\"CH07-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"languageId\\\": \\\"SYN-L1\\\",\",\"sourceParagraph\":1484},{\"blockId\":\"CH07-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"LS-5\\\",\",\"sourceParagraph\":1485},{\"blockId\":\"CH07-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"populationWeight\\\": 0.55,\",\"sourceParagraph\":1486},{\"blockId\":\"CH07-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanelValid\\\": 550,\",\"sourceParagraph\":1487},{\"blockId\":\"CH07-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessOversampleValid\\\": 0,\",\"sourceParagraph\":1488},{\"blockId\":\"CH07-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedRepresentation\\\": 0.95\",\"sourceParagraph\":1489},{\"blockId\":\"CH07-MB0031\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1490},{\"blockId\":\"CH07-MB0032\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1491},{\"blockId\":\"CH07-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"languageId\\\": \\\"SYN-L2\\\",\",\"sourceParagraph\":1492},{\"blockId\":\"CH07-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"LS-5\\\",\",\"sourceParagraph\":1493},{\"blockId\":\"CH07-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"populationWeight\\\": 0.25,\",\"sourceParagraph\":1494},{\"blockId\":\"CH07-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanelValid\\\": 250,\",\"sourceParagraph\":1495},{\"blockId\":\"CH07-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessOversampleValid\\\": 0,\",\"sourceParagraph\":1496},{\"blockId\":\"CH07-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedRepresentation\\\": 0.92\",\"sourceParagraph\":1497},{\"blockId\":\"CH07-MB0039\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1498},{\"blockId\":\"CH07-MB0040\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1499},{\"blockId\":\"CH07-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"languageId\\\": \\\"SYN-L3\\\",\",\"sourceParagraph\":1500},{\"blockId\":\"CH07-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"LS-5\\\",\",\"sourceParagraph\":1501},{\"blockId\":\"CH07-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"populationWeight\\\": 0.10,\",\"sourceParagraph\":1502},{\"blockId\":\"CH07-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanelValid\\\": 100,\",\"sourceParagraph\":1503},{\"blockId\":\"CH07-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessOversampleValid\\\": 0,\",\"sourceParagraph\":1504},{\"blockId\":\"CH07-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedRepresentation\\\": 0.88\",\"sourceParagraph\":1505},{\"blockId\":\"CH07-MB0047\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1506},{\"blockId\":\"CH07-MB0048\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1507},{\"blockId\":\"CH07-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"languageId\\\": \\\"SYN-L4\\\",\",\"sourceParagraph\":1508},{\"blockId\":\"CH07-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"LS-5\\\",\",\"sourceParagraph\":1509},{\"blockId\":\"CH07-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"populationWeight\\\": 0.06,\",\"sourceParagraph\":1510},{\"blockId\":\"CH07-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanelValid\\\": 60,\",\"sourceParagraph\":1511},{\"blockId\":\"CH07-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessOversampleValid\\\": 140,\",\"sourceParagraph\":1512},{\"blockId\":\"CH07-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedRepresentation\\\": 0.65,\",\"sourceParagraph\":1513},{\"blockId\":\"CH07-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"promptEquivalence\\\": \\\"PASSED\\\",\",\"sourceParagraph\":1514},{\"blockId\":\"CH07-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"referenceParity\\\": \\\"PASSED\\\",\",\"sourceParagraph\":1515},{\"blockId\":\"CH07-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationParity\\\": \\\"PASSED\\\"\",\"sourceParagraph\":1516},{\"blockId\":\"CH07-MB0058\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1517},{\"blockId\":\"CH07-MB0059\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1518},{\"blockId\":\"CH07-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"languageId\\\": \\\"SYN-L5\\\",\",\"sourceParagraph\":1519},{\"blockId\":\"CH07-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"LS-5\\\",\",\"sourceParagraph\":1520},{\"blockId\":\"CH07-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"populationWeight\\\": 0.04,\",\"sourceParagraph\":1521},{\"blockId\":\"CH07-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"populationPanelValid\\\": 40,\",\"sourceParagraph\":1522},{\"blockId\":\"CH07-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessOversampleValid\\\": 160,\",\"sourceParagraph\":1523},{\"blockId\":\"CH07-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"estimatedRepresentation\\\": 0.58,\",\"sourceParagraph\":1524},{\"blockId\":\"CH07-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"promptEquivalence\\\": \\\"PASSED\\\",\",\"sourceParagraph\":1525},{\"blockId\":\"CH07-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"referenceParity\\\": \\\"PASSED\\\",\",\"sourceParagraph\":1526},{\"blockId\":\"CH07-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationParity\\\": \\\"PASSED\\\"\",\"sourceParagraph\":1527},{\"blockId\":\"CH07-MB0069\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1528},{\"blockId\":\"CH07-MB0070\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1529},{\"blockId\":\"CH07-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"globalPopulationWeightedScore\\\": 0.9027,\",\"sourceParagraph\":1530},{\"blockId\":\"CH07-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"equalLanguageDiagnosticScore\\\": 0.796,\",\"sourceParagraph\":1531},{\"blockId\":\"CH07-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"languageFloor\\\": 0.58,\",\"sourceParagraph\":1532},{\"blockId\":\"CH07-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"maxLanguageGap\\\": 0.37,\",\"sourceParagraph\":1533},{\"blockId\":\"CH07-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"oversamplesReweightedToPopulation\\\": true,\",\"sourceParagraph\":1534},{\"blockId\":\"CH07-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"allLanguagesClaimAllowed\\\": false\",\"sourceParagraph\":1535},{\"blockId\":\"CH07-MB0077\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1536},{\"blockId\":\"CH07-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":1537},{\"blockId\":\"CH07-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"panelRolesKeptSeparate\\\": true,\",\"sourceParagraph\":1538},{\"blockId\":\"CH07-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"duplicateFullWeightingProhibited\\\": true,\",\"sourceParagraph\":1539},{\"blockId\":\"CH07-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"accountableHumanRole\\\": \\\"COUNTRY_LANGUAGE_PANEL_OWNER\\\",\",\"sourceParagraph\":1540},{\"blockId\":\"CH07-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"lockedAt\\\": \\\"2026-08-18T09:00:00Z\\\"\",\"sourceParagraph\":1541},{\"blockId\":\"CH07-MB0083\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1542},{\"blockId\":\"CH07-MB0084\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1543},{\"blockId\":\"CH07-MB0085\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1544},{\"blockId\":\"CH07-MB0086\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1545},{\"blockId\":\"CH07-MB0087\",\"type\":\"paragraph\",\"text\":\"this is not real country or language data,\",\"sourceParagraph\":1546},{\"blockId\":\"CH07-MB0088\",\"type\":\"paragraph\",\"text\":\"this is not actual AI performance,\",\"sourceParagraph\":1547},{\"blockId\":\"CH07-MB0089\",\"type\":\"paragraph\",\"text\":\"it is a machine-readable synthetic representation of the panel distinction.\",\"sourceParagraph\":1548},{\"blockId\":\"CH07-MB0090\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RULE OF SECTION 48\",\"sourceParagraph\":1550},{\"blockId\":\"CH07-MB0091\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH07-R01\",\"sourceParagraph\":1551},{\"blockId\":\"CH07-MB0092\",\"type\":\"paragraph\",\"text\":\"The GEO-1000 Population Panel, Country Observer Panel, and Language\",\"sourceParagraph\":1553},{\"blockId\":\"CH07-MB0093\",\"type\":\"paragraph\",\"text\":\"Fairness Panel MUST remain distinct in purpose, sampling, weighting, and\",\"sourceParagraph\":1554},{\"blockId\":\"CH07-MB0094\",\"type\":\"paragraph\",\"text\":\"public claims.\",\"sourceParagraph\":1555},{\"blockId\":\"CH07-MB0095\",\"type\":\"paragraph\",\"text\":\"A Country Observer observation MAY establish that an access condition,\",\"sourceParagraph\":1557},{\"blockId\":\"CH07-MB0096\",\"type\":\"paragraph\",\"text\":\"response, anomaly, or critical incident occurred in a defined\",\"sourceParagraph\":1558},{\"blockId\":\"CH07-MB0097\",\"type\":\"paragraph\",\"text\":\"country-product-language context.\",\"sourceParagraph\":1559},{\"blockId\":\"CH07-MB0098\",\"type\":\"paragraph\",\"text\":\"A Country Observer observation MUST NOT be represented as a country prevalence\",\"sourceParagraph\":1561},{\"blockId\":\"CH07-MB0099\",\"type\":\"paragraph\",\"text\":\"estimate or country GEO score.\",\"sourceParagraph\":1562},{\"blockId\":\"CH07-MB0100\",\"type\":\"paragraph\",\"text\":\"Country observation coverage, population-weighted observation coverage,\",\"sourceParagraph\":1564},{\"blockId\":\"CH07-MB0101\",\"type\":\"paragraph\",\"text\":\"and country estimation coverage MUST be reported separately.\",\"sourceParagraph\":1565},{\"blockId\":\"CH07-MB0102\",\"type\":\"paragraph\",\"text\":\"Every eligible country SHOULD receive a positive probability of\",\"sourceParagraph\":1567},{\"blockId\":\"CH07-MB0103\",\"type\":\"paragraph\",\"text\":\"observation within a predeclared surveillance cycle, unless it is\",\"sourceParagraph\":1568},{\"blockId\":\"CH07-MB0104\",\"type\":\"paragraph\",\"text\":\"explicitly outside the product's lawful Native Reach.\",\"sourceParagraph\":1569},{\"blockId\":\"CH07-MB0105\",\"type\":\"paragraph\",\"text\":\"Language fairness MUST evaluate materially comparable representation\",\"sourceParagraph\":1571},{\"blockId\":\"CH07-MB0106\",\"type\":\"paragraph\",\"text\":\"under semantically equivalent prompts and comparable contexts. It MUST\",\"sourceParagraph\":1572},{\"blockId\":\"CH07-MB0107\",\"type\":\"paragraph\",\"text\":\"NOT require word-for-word identical responses or ignore legitimate local\",\"sourceParagraph\":1573},{\"blockId\":\"CH07-MB0108\",\"type\":\"paragraph\",\"text\":\"differences.\",\"sourceParagraph\":1574},{\"blockId\":\"CH07-MB0109\",\"type\":\"paragraph\",\"text\":\"Oversampled languages MUST be reweighted to their target population\",\"sourceParagraph\":1576},{\"blockId\":\"CH07-MB0110\",\"type\":\"paragraph\",\"text\":\"shares for global population estimates.\",\"sourceParagraph\":1577},{\"blockId\":\"CH07-MB0111\",\"type\":\"paragraph\",\"text\":\"Population-weighted scores, equal-language diagnostic scores, language\",\"sourceParagraph\":1579},{\"blockId\":\"CH07-MB0112\",\"type\":\"paragraph\",\"text\":\"floors, critical-error rates, unknown rates, and untested languages MUST\",\"sourceParagraph\":1580},{\"blockId\":\"CH07-MB0113\",\"type\":\"paragraph\",\"text\":\"remain separately visible.\",\"sourceParagraph\":1581},{\"blockId\":\"CH07-MB0114\",\"type\":\"paragraph\",\"text\":\"A language disparity MUST NOT be attributed solely to the AI product\",\"sourceParagraph\":1583},{\"blockId\":\"CH07-MB0115\",\"type\":\"paragraph\",\"text\":\"before prompt equivalence, reference parity, measurement parity, and\",\"sourceParagraph\":1584},{\"blockId\":\"CH07-MB0116\",\"type\":\"paragraph\",\"text\":\"adjudication parity are assessed.\",\"sourceParagraph\":1585},{\"blockId\":\"CH07-MB0117\",\"type\":\"paragraph\",\"text\":\"A single observation MAY carry multiple diagnostic panel tags, but MUST\",\"sourceParagraph\":1587},{\"blockId\":\"CH07-MB0118\",\"type\":\"paragraph\",\"text\":\"NOT receive duplicate full population weight.\",\"sourceParagraph\":1588},{\"blockId\":\"CH07-MB0119\",\"type\":\"paragraph\",\"text\":\"Critical Observer incidents MAY trigger confirmatory sampling and\",\"sourceParagraph\":1590},{\"blockId\":\"CH07-MB0120\",\"type\":\"paragraph\",\"text\":\"conformity review without being treated as evidence of country\",\"sourceParagraph\":1591},{\"blockId\":\"CH07-MB0121\",\"type\":\"paragraph\",\"text\":\"prevalence.\",\"sourceParagraph\":1592},{\"blockId\":\"CH07-MB0122\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1593},{\"blockId\":\"CH07-MB0123\",\"type\":\"paragraph\",\"text\":\"GEO-1000 Population Panel, Country Observer Panel, and Language Fairness Panel should be kept separate in terms of purpose, sample, weighting, and public claim. A single Country Observer observation can show that a certain access condition, response, anomaly, or critical event has occurred in a specific country–product–language context; it cannot be presented as a country prevalence estimate or country GEO score. Country coverage, population-weighted coverage, and country estimate coverage should be reported separately. Language fairness should assess material representation in semantically equivalent prompts and comparable conditions; it should not demand word-for-word identical responses and should not ignore legitimate local differences. Over-sampled languages should be reweighted to reflect the actual target population shares in global population estimates.\",\"sourceParagraph\":1594}]}}","text":"## Chapter Boundary\n\nSection 5 established the following provision:\n\n> For each AI product, a product-eligible user population must be defined separately.\n\nSection 6 converted this population into sampling:\n\n> The main Population Panel should provide predetermined probabilities of selection and weights to people in the target population before results are seen.\n\nNow we need to solve two different fairness problems. The first problem:\n\n> Population weighting may make very small countries invisible in the main sample.\n\nThe second problem:\n\n> The global population average can hide severe representation distortions in low-resource or low-population languages.\n\nA pure population sample is strong for measuring the average human experience in the world population. However, the same sample:\n\n- cannot observe each country individually,\n\n- cannot provide enough samples for each language,\n\n- cannot produce a country score for small countries,\n\nmay not catch rare but substantive errors in low-resource languages. Therefore, GEO-1000 cannot consist of a single panel alone. Alongside the main population panel, two complementary structures are needed:\n\n#### Country Observer Panel\n\nCanonical panel label:\n\n#### Country Observer Panel\n\nand:\n\n#### Language Fairness Panel\n\nCanonical panel label:\n\n#### Language Fairness Panel\n\nThe Country Observer Panel answers the following question:\n\n> In countries that are underrepresented in the main population panel, is there at least real user observation, access verification, or early risk signal?\n\nThe Language Fairness Panel, on the other hand, answers the following question:\n\n> Even if the global average appears high, are users in certain languages being represented more incorrectly, incompletely, or riskily?\n\nThese two panels do not replace the main Population Panel. They complement it. Giving a Country Observer observation to a country does not measure that country's score. Over-sampling a language does not make that language larger in the global population. The task of this section is:\n\n- country visibility versus population weight,\n\n- to separate the global average with language fairness,\n\n- to separate population estimation with early warning,\n\n- to separate statistical sufficiency with observation scope\n\nfrom each other. This section has not yet:\n\n- defined how prompts should be translated in all languages,\n\n- the back-translation and semantic equivalence test,\n\n- the detailed selection mechanism of participants,\n\n- how atomic claims will be scored by adjudicators,\n\n- the final conformity thresholds\n\nexactly. These issues will be further regulated in the following sections. The fundamental question of Section 7 is:\n\n> How do we measure while producing a global score without artificially inflating small countries and languages, but also without leaving them completely invisible?\n\n## NOMOS Challenge\n\nYou set up a 1,000-person population panel for an AI product. As a result of a population-proportional sample:\n\n- hundreds of users from a few large countries,\n\n- dozens of users from some medium-sized countries,\n\n- no users from most small countries\n\ncame. Global score:\n\n#### 94\n\nIt was calculated as such. Then you said: 'This AI product represents the brand worldwide with 94 per cent accuracy.' However, there is not a single real user observation in 70 small countries. These countries may constitute a small portion of the world's population. Still, the following questions remain unanswered: Is the product really accessible? Does the intent work in the correct language? Does the system confuse the country with another country? Is the local company confused with the global brand? Are the price, service, and legal context wrong? Does the system fail to respond? Are there low-frequency critical errors? The global population score does not answer these questions. Now imagine that you added one person to each country. You have obtained 70 additional observations.\n\nYou added each of these countries to the main score with a weight of 1/1,070, making small countries many times larger than their true share of the population. Geographic visibility increased, but the population estimate was distorted. Then one user in one country received an incorrect response, and you declared: ‘This country's GEO score is zero.’ One person's response is not a country score. It is a single event observed under specified conditions in that country. Now consider language. The English result is 98 per cent, German 96 per cent and Turkish 91 per cent. A low-resource language has only eight users and a raw result of 50 per cent. You say, ‘Eight people are not enough; remove the language from the report.’ The global score rises, but the experience of users in that language disappears.\n\nIn another approach, you give an additional 200 users to a low-resource language. The raw sample mean decreases. Then you say: \"The global score dropped to 84.\" However, the actual population share of the language may be 2 per cent. Oversampling increases statistical power. It does not increase population weight. The first ruling of this section is as follows:\n\n> Being observed is not the same as having weight.\n\nIts second provision states:\n\n> A single country observation is not a country score.\n\nIts third provision states:\n\n> Language fairness is not giving equal weight to all languages in the global score.\n\nIts fourth provision states:\n\n> A high global average does not eliminate severe degradation in a particular language.\n\nIts fifth provision states:\n\n> Additional observations taken to increase coverage cannot be added to the main score without being reweighted according to the true population.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to establish complementary panels that increase country and language visibility without disrupting the statistical population representation of the main Population Panel. The section normalises the following distinctions:\n\n- Geographic visibility with population weight\n\n- Country observation with country score\n\n- Single event detection with prevalence estimation\n\n- Main Population Panel with Country Observer Panel\n\n- Country Observer Panel with country-specific validation study\n\n- Country coverage with population coverage\n\n- Number of languages coverage with language population coverage\n\n- Language fairness with equal global weighting of languages\n\n- Language global share with excessive sampling\n\n- Native language and prompt language\n\n- Language and locale\n\n- Language and writing system\n\n- Supported language and language that can generate responses alone\n\n- Prompt translation defect and AI language defect\n\n- Resource shortage and AI representation defect\n\n- Adjudicator insufficiency and product failure\n\n- Legitimate local difference and unjust language inequality\n\n- Global average and sub-group base\n\n- Country access coverage and country representation accuracy\n\n- Observer alert and definitive conformity decision\n\n- Single-wave surveillance cycle with multi-wave geographical coverage\n\nAt the end of this section, each GEO-1000 audit should be able to provide clear answers to the following questions:\n\n> In which countries is there at least one actual observation?\n\n> In which countries is there only early warning observation, and in which ones are there statistical predictions?\n\n> Which languages were measured, which ones reached sufficient sampling, and which ones were not tested?\n\n> If low-resource languages were oversampled, how was this reweighted in the global score?\n\n> Is the source of a poor result in a language truly AI-generated, or is it due to prompt, resource, locale, or adjudication issues?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\n> GEO-1000 should manage the main population estimate with the Population Panel; the small country observation with the Country Observer Panel; language inequality and low-resource language identification with the Language Fairness Panel separately.\n\nThese three structures:\n\n- differ in purposes,\n\n- differ in sample targets,\n\n- differ in weights,\n\n- differ in public claims\n\nhave. An observation from one panel cannot be converted into the meaning of another panel without explanation. The default structure:\n\n## 3. WHAT IS THE Country Observer PANEL?\n\nCountry Observer Panel, in countries that are underrepresented or not represented enough in the main population panel:\n\n- product access,\n\n- prompt applicability,\n\n- basic entity analysis,\n\n- language and locale behaviour,\n\n- early Critical error signal\n\nis a supplementary audit panel set up to observe. The primary task of the Country Observer Panel is not to produce a score.\n\n> It is to make the invisible country visible for the first time.\n\nA Country Observer observation can show: “In the specified country, under the specified AI product, user surface, language, and time condition, this result has been observed at least once.” It cannot show: “X% of users in this country see the same result.”\n\n## 4. Country Observer UNIT\n\nThe basic observation unit of the Country Observer Panel is not the country alone. The preferred unit:\n\nS_{c,a,l,s,w}\n\nwhere:\n\n- c: country or jurisdiction\n\n- a: AI product and user surface\n\n- l: language, writing system, and locale\n\n- s: session or plan condition\n\n- w: measurement wave\n\nThis distinction is necessary. In the same country:\n\n- more than one language,\n\n- different plans,\n\n- web and mobile surfaces,\n\n- different legal or technical access conditions\n\nmay exist. Adding a user to a country does not mean monitoring all language and user surfaces of the country.\n\n## 5. DUTIES OF THE Country Observer PANEL\n\n### 5.1. Access Verification\n\nAlthough the product appears to be available in the country in the official registration, the actual user:\n\n- may not be able to log in,\n\n- may not be able to reach the specified surface,\n\nIt may encounter a plan or verification obstacle. The Country Observer observation tests the official access record at the field level.\n\n### 5.2. Basic Identity Check\n\nAI product supervising entity:\n\n- the right company,\n\n- the right brand,\n\n- correct country relationship\n\nCan it analyse inside?\n\n### 5.3. Detection of Locale Deviation\n\nIn different country locales in the same language:\n\n- price,\n\n- company,\n\n- service,\n\n- law,\n\n- advice\n\nIs it changing? If it is changing, can this difference be financial and explainable?\n\n### 5.4. Low Frequency Heavy Fault Signal\n\nIn the response of a single observer:\n\n- wrong licence,\n\n- wrong legal entity,\n\n- dangerous health or safety information,\n\n- heavy identity fusion\n\nIt can be observed. A single event does not indicate the prevalence in the country. However, it can trigger verification work.\n\n### 5.5. New Research Hypothesis\n\nThe observer panel may raise the question: 'Is this event singular, or is it a recurring pattern among users in the country?' This question is tested with a targeted additional sample.\n\n### 5.6. Geographic Scope Record\n\nIt shows in how many suitable countries the product has real user observation. This is not the number of country scores. It is observational coverage.\n\n## 6. THINGS THE Country Observer PANEL CANNOT DO\n\nCountry Observer Panel alone:\n\n- reliable country score,\n\n- country population rate,\n\n- country ranking,\n\n- country conformity decision,\n\n- country-based model superiority\n\ncannot produce. If there is a single observer in a country and the answer is correct: “The country score is 100” cannot be said. If it is wrong: “The country score is zero” cannot be said. Correct status:\n\n### OBSERVED_PASS\n\n### Observed Issue\n\n### OBSERVED_CRITICAL_CANDIDATE\n\n### ACCESS_FAILURE\n\n### INSUFFICIENT_FOR_COUNTRY_ESTIMATE\n\nare observational records.\n\n## 7. ELIGIBLE COUNTRY REGISTRY\n\nAn Eligible Country Register should be created for each AI product. Each country can hold one of the following statuses:\n\n### CE-0 — UNKNOWN\n\nThe product and audit access status could not be determined.\n\n### CE-1 — OUTSIDE NATIVE REACH\n\nThe product remains outside the official or legitimate area of use. It is not taken into account for the country representation accuracy score. It is shown as an access scope gap.\n\n### CE-2 — RESTRICTED OR SPECIAL ACCESS\n\nOnly:\n\n- institution,\n\n- invitation,\n\n- plan,\n\n- device,\n\n- user type\n\naccess is available. A separate panel may be required.\n\n### CE-3 — ELIGIBLE FOR Country Observer observation\n\nThe product is suitable for Country Observer observation in terms of the relevant user surface and at least one audit language.\n\n### CE-4 — ELIGIBLE FOR POPULATION ESTIMATION\n\nIt has a product-eligible population share in the country's main population panel.\n\n### CE-5 — ELIGIBLE FOR COUNTRY ESTIMATION\n\nA country estimate can be produced under sufficient sample, effective sample, language coverage, and quality conditions. A country can be CE-3 but not CE-5. This means the country is observed but a reliable country score could not be produced.\n\n## 8. THE LADDER BETWEEN Country Observer observation AND COUNTRY ESTIMATION\n\n### CS-0 — NOT OBSERVED\n\nThere is no valid Country Observer observation for the eligible country.\n\n### CS-1 — SINGLE Country Observer observation\n\nThere is one valid country–product–language observation. Generates only an event record.\n\n### CS-2 — REPEATED Country Observer observation\n\nThere are several observations from different waves or users. May raise a pattern suspicion. Not sufficient for the country rate.\n\n### CS-3 — EXPLORATORY COUNTRY SAMPLE\n\nThe exploratory sample band in section 6 has been reached. An exploratory result can be given with wide uncertainty.\n\n### CS-4 — PROVISIONAL COUNTRY ESTIMATE\n\nA provisional country estimate can be produced with sufficient and balanced sampling.\n\n### CS-5 — COUNTRY ESTIMATION CANDIDATE\n\nSufficient active sample, language/locale coverage, repeat waves, and quality controls are in place. It can be a country score candidate. Exact thresholds should be calibrated with a pilot and external review.\n\n## 9. OBSERVER CYCLE\n\nObserving every wave in all eligible countries may not be possible in terms of cost and operations. Therefore, the Observer Cycle can be used. Measurement waves:\n\nw=1,2,…,K\n\nand eligible countries: let c∈CS. Assignment of an observer for country c in wave w:\n\nz_{c,w} = 1 (if the country will be observed in the wave); otherwise z_{c,w} = 0\n\ncan be represented as follows. Basic cycle condition:\n\n\\sum_{w=1}^K z_{c,w} \\geq 1\n\nmust be satisfied. That is, each eligible country should be observed at least once within the defined surveillance cycle. This provision:\n\n- does not mean that each country will be observed in every wave,\n\n- or that a single observation will constitute the country score\n\nNeither condition follows from the cycle requirement.\n\n## 10. OBSERVER FREQUENCY CLASSES\n\nCountries can be divided into predefined frequency classes.\n\n### SF-0 — OUTSIDE REACH\n\nNo product access. No user-representative observation is conducted. Access status is monitored.\n\n### SF-1 — ONCE PER SURVEILLANCE CYCLE\n\nLow-risk countries, small countries, or countries that already have sufficient indirect coverage in the Population Panel.\n\n### SF-2 — PERIODIC OBSERVER\n\nObserved again at regular intervals.\n\n### SF-3 — EVERY PRINCIPAL WAVE\n\nCan be observed in every principal wave for the following reasons:\n\n- high user population,\n\n- material target market,\n\n- previous serious error,\n\n- rapid access or regulatory change,\n\nhigh-risk service area.\n\n### SF-4 — INCIDENT MONITORING\n\nIntensified observation due to verified Critical event or continuous representation disruption. The frequency class cannot be retroactively changed after results are reviewed. A new event can create a new version for the next wave.\n\n## 11. OBSERVER PRIORITY\n\nThe frequency of Country Observer activity should not be determined solely based on population. As relevant, the following areas may be considered:\n\n- Product-eligible population\n\n- The actual service or market coverage of the entity\n\n- Previous erroneous representation records\n\n- High-risk user impact\n\n- Changes in product access\n\n- Language and locale uniqueness\n\n- Low-resource language usage\n\n- Legal or operational differences\n\n- Time elapsed since the last observation\n\nThese fields must be used according to the predefined governance rule. The approach of “The result of this country is bad, let's measure less.” is prohibited.\n\n## 12. Country Observer COVERAGE METRICS\n\nCountry coverage is not a single number. At least three separate metrics must be used.\n\n### 12.1. Country Observation Coverage\n\nThe proportion of eligible countries with at least one valid observer or population observation.\n\nCOC_count = |{c ∈ C_S: n_c^obs > 0}| / |C_S|\n\nThis metric shows the coverage in terms of the number of countries.\n\n### 12.2. Population-Weighted Country Observation Coverage\n\nThe proportion of the eligible population in countries with at least one valid observation to the total target population.\n\nCOC_pop = [Σ_{c:n_c^obs>0} N_c] / [Σ_{c∈C_S} N_c]\n\nThis metric shows whether large populations are observed.\n\n### 12.3. Country Prediction Coverage\n\nIt is the proportion of countries that reach a sufficient country sample.\n\nCEC = |{c: n_{c,eff} ≥ m_c}| / |C_S|\n\nHere, mc is the predefined adequacy threshold for the country prediction. An audit:\n\n- 100 per cent country observation coverage,\n\n- 99 per cent population observation coverage,\n\n- may carry only 10 per cent country prediction coverage\n\nThese results cannot be converted into each other.\n\n## 13. OBSERVER ALARM\n\nWhen a material issue is found in a watcher's observation, a Watcher Alarm is created. Types of alarms:\n\n### SA-1 — ACCESS ANOMALY\n\nThe official access record conflicts with actual user access.\n\n### SA-2 — ENTITY ANOMALY\n\nThe entity merges with another company, product, or country operation.\n\n### SA-3 — LANGUAGE OR LOCALE ANOMALY\n\nIncorrect language, wrong country context, or obvious locale deviation occurs.\n\n### SA-4 — MATERIAL FACT ANOMALY\n\nThe material price, service, capacity, partnership, or time is incorrect.\n\n### SA-5 — CRITICAL INCIDENT CANDIDATE\n\nHealth, law, finance, security, licensing, or severe identity error exists.\n\n### SA-6 — REFERENCE OR PROMPT ANOMALY\n\nThe source of the problem is not the AI response:\n\n- prompt,\n\n- translation,\n\n- official source,\n\n- adjudication\n\nmight be at fault. The alarm is not an automatic country failure. It is the beginning of verification and escalation.\n\n## 14. OBSERVER ESCALATION PROTOCOL\n\n### Step 1 — Validity Verification\n\nIs the observation really in accordance with the protocol? Correct country Correct product Correct prompt Correct language First appropriate output Full evidence checked.\n\n### Step 2 — Re-coding the Incident\n\nProblem:\n\n- AI product,\n\n- reference record,\n\n- prompt,\n\n- participant,\n\n- access\n\nWhich of the sources could it belong to?\n\n### Step 3 — Independent Repeat\n\nBefore selecting the result, verification is done with predefined new users or sessions.\n\n### Step 4 — Targeted Sampling\n\nIf the event continues, country-specific sampling may be applied. Sample size:\n\n- event severity,\n\n- expected prevalence,\n\n- country population,\n\n- language diversity\n\nare predefined in advance.\n\n### Step 5 — Decision Separation\n\nThe following results are kept separate:\n\n- Single verified event\n\n- Repeated pattern\n\n- Country rate\n\n- Critical suitability effect\n\nA single Critical event can affect the conformity review even without an estimate of prevalence.\n\n## 15. WHAT IS LANGUAGE FAIRNESS?\n\nLanguage fairness:\n\n> It is not about all languages receiving exactly the same word-for-word answer.\n\nIt means:\n\n> In semantically equivalent and materially comparable user questions, apart from real local differences, users should not systematically receive representations that are more incorrect, more incomplete, less evidenced, or riskier due to their language.\n\nLanguage fairness does not seek this equality:\n\nR_Turkish = R_English = R_Japanese\n\nword by word. It searches for this:\n\n> Comparable representation in terms of identity, material reality, boundary, time, source status, and user suitability.\n\n## 16. LANGUAGE DIFFERENCE IS NOT ALWAYS UNFAIRNESS\n\nLanguage outcomes can legitimately vary for the following reasons: The scope of local services is different. Price and contract conditions vary by country. Legal knowledge depends on the jurisdiction. User intent is different. Prompts are not semantically identical. Official sources carry different content across languages. The AI product does not officially support the relevant language. Each of these differences should be recorded. Explainable local differences should not be labelled as injustice. However, even though the reality is the same, in a low-resource language:\n\n- the company identity is corrupted,\n\n- borders are removed,\n\n- the wrong licence is added,\n\n- the response is constantly given in another language\n\nit may be a language-based representation problem.\n\n## 17. LANGUAGE UNIT\n\nIn language assessment, just the general language name may not be sufficient. The preferred language unit:\n\nL=(λ,σ,ℓ,d)\n\nand can be considered as. Here:\n\n- λ: language\n\n- σ: writing system\n\n- ℓ: locale or country context\n\n- d: task or domain\n\nFor example, the same language:\n\n- in a different writing system,\n\n- in a different country,\n\n- in legal or healthcare field\n\nmay require a separate measurement cell. Not all users of a language are a single homogeneous group.\n\n## 18. LANGUAGE SUPPORT STATUSES\n\nEach AI product can use one of the following statuses for language and user surface:\n\n### LS-0 — UNKNOWN\n\nSupport status could not be determined.\n\n### LS-1 — NOT SUPPORTED\n\nThe product does not officially support the language in question. It is not evaluated within the main supported language score. Experimental testing may be conducted.\n\n### LS-2 — OBSERVED BUT UNSUPPORTED\n\nThe product can generate responses. However, there is no official support or quality guarantee. Reported as a separate experimental result.\n\n### LS-3 — LIMITED SUPPORT\n\nSupport exists for specific tasks, interfaces, or features.\n\n### LS-4 — OFFICIALLY SUPPORTED\n\nThe product language generally claims to be supported.\n\n### LS-5 — AUDIT-ELIGIBLE\n\nLanguage:\n\n- official support,\n\n- prompt equivalence,\n\n- local adjudicator,\n\n- reference proficiency,\n\n- valid capture\n\nis fully audit-compliant. The presence of official support alone does not constitute LS-5.\n\n## 19. LOW-RESOURCE LANGUAGE\n\nThe term “low-resource language” does not describe the value of the language or the quality of its speakers. In certain technology and audit contexts, it indicates one or more of the following deficiencies:\n\n- Limited digital content\n\n- Limited localised official source\n\n- Limited evaluation dataset\n\n- Restricted adjudicator access\n\n- Low product support\n\n- Limited terminology resource\n\n- Limited machine translation quality\n\n- Low user panel access\n\nA language can be low-resource in one field and strong in another. For example, while it has strong resources in everyday conversation:\n\n- law,\n\n- medicine,\n\n- specific technical field\n\nmay have limited resources. Therefore, language resource level:\n\n- product,\n\n- task,\n\n- time,\n\n- field\n\nshould be defined together.\n\n## 20. TASKS OF THE Language Fairness Panel\n\n### 20.1. Oversampling Low-Resource Languages\n\nProvides additional users for languages that do not receive enough observations in the main population panel.\n\n### 20.2. Measuring Language-Based Material Error Distribution\n\nNot only overall accuracy:\n\n- identity,\n\n- scope,\n\n- time,\n\n- source,\n\n- recommendation,\n\n- Critical error\n\nexamines the results separately.\n\n### 20.3. Verifying Response Language\n\nDoes the AI respond in alignment with the prompt language? If it switches to another language, what is the reason?\n\n### 20.4. Examining Language and Locale Distinction\n\nDoes the same language carry different local realities in different countries? Does AI manage this difference correctly?\n\n### 20.5. Examining Source and Official Representation Parity\n\nDo the official records of existence carry the same material reality in all languages? A low AI result may arise from a language gap in the source itself.\n\n### 20.6. Testing Prompt Equivalence\n\nDo translations truly carry the same user intent?\n\n### 20.7. Testing Adjudicator Parity\n\nCan adjudicators in different languages apply the same normative standard?\n\n## 21. FOUR GATES FOR LANGUAGE FAIRNESS\n\nA language outcome should be evaluated through four gates before attributing it to the AI product’s language performance.\n\n#### Gate 1 — Intent Equivalence\n\nAre the prompts semantically comparable? If it fails:\n\n### PROMPT_EQUIVALENCE_FAILURE\n\nstatus is given.\n\n#### Gate 2 — Reference Parity\n\nAre official and verified reality records sufficient for the relevant language? If it fails:\n\n### REFERENCE_LANGUAGE_GAP\n\nstatus is given.\n\n#### Gate 3 — Measurement Parity\n\nAre capture, session, interface, and user conditions comparable? If it fails:\n\n### MEASUREMENT_PARITY_FAILURE\n\nstatus is given.\n\n#### Gate 4 — Adjudication Parity\n\nAre there adjudicators in the native language or with sufficient local expertise and the same evaluation rules? If it fails:\n\n### ADJUDICATION_LANGUAGE_GAP\n\nis assigned. Unless the four doors are sufficiently met: “The AI product is worse in this language.” cannot be definitively established. The language gap is still a finding of the GEO ecosystem. However, its source must be correctly classified.\n\n## 22. ESTABLISHMENT OF THE LANGUAGE UNIVERSE\n\nThe phrase “all languages” cannot be used without limitation. Each audit must create a Language Register within the Scope. Languages can be selected from the following groups:\n\n### 22.1. Global Core Language Set\n\nNative languages used by a large human population appropriate for the product.\n\n### 22.2. Entity Market Language Set\n\nThe languages targeted by the audited entity:\n\n- service,\n\n- in terms of content,\n\n- customer,\n\n- contract\n\nand maintenance.\n\n### 22.3. Product Support Language Set\n\nLanguages officially supported by the AI product.\n\n### 22.4. Low-Resource Observer Language Set\n\nLow-resource languages that are weak in the population panel or carry risk in terms of product behaviour.\n\n### 22.5. High-Risk Language Set\n\nLanguages that have material user impact in the context of health, law, finance, or security.\n\n### 22.6. Experimental Language Set\n\nLanguages that the product does not officially support but are tested for observational or research purposes. These groups may overlap. The reason each language is included should be recorded.\n\n## 23. LANGUAGE FAIRNESS ALLOCATION\n\nFor language l, let the target in the main population panel be nlP. The additional target in the fairness panel is nlF. The total collected language observation:\n\nn_l^T = n_l^P + n_l^F\n\nis obtained. The total weight of the language in the global population calculation is preserved as Wl. The normalised weight per observation:\n\nw_{i,l} = W_l / n_l^T\n\ncan be considered in this way. Thus, the language is oversampled. However, it does not exceed the actual population share in the global score.\n\n## 24. CONTRIBUTION OF THE Language Fairness Panel TO THE GLOBAL SCORE\n\nLanguage fairness observations can be used in two ways.\n\n### 24.1. Separate Diagnostic Analysis\n\nThe fairness panel only produces language scores and error types. The main Population Panel does not change the global score. This method is simpler.\n\n### 24.2. Combined Reweighted Analysis\n\nPopulation and Language Fairness observations are used together. Each observation is reweighted according to the actual target language population weight. This method can increase the volume of information. However:\n\n- double counting,\n\n- multilingual user,\n\n- panel dependency,\n\n- weight variability\n\nIt should be taken into consideration. The method used must be defined before the results are seen.\n\n## 25. LANGUAGE COVERAGE METRICS\n\n### 25.1. Language Observation Coverage\n\nThe proportion of languages in the scope that have at least one valid observation.\n\nLOC = |{l: n_l^obs > 0}| / |L_scope|\n\n### 25.2. Language Prediction Coverage\n\nThe proportion of languages that meet sufficient sample and quality requirements.\n\nLEC = |{l: n_{l,eff} ≥ m_l}| / |L_scope|\n\n### 25.3. Population-Weighted Language Coverage\n\nThe proportion of the target language population that is adequately observed. Requires a primary audit language or divided person weighting to prevent double-counting of multilingual users.\n\n### 25.4. Product Support Coverage\n\nIndicates in how many of the covered languages the product receives official support. Language support and representation accuracy are kept separate.\n\n## 26. LANGUAGE INEQUALITY METRICS\n\n### CANDIDATE METRICS — PROPOSED METRICS\n\nThe exact metric set should be calibrated through pilot and external review.\n\n### 26.1. Language Outcome\n\nFor each sufficient language cell: θl is calculated.\n\n### 26.2. Highest–Lowest Difference\n\nDG_max = max_l θ_l − min_l θ_l\n\nEasy to understand. However, sensitive to small and volatile single cells.\n\n### 26.3. Robust Percentile Difference\n\nDG_{90−10} = Q_{0.90}(θ_l) − Q_{0.10}(θ_l)\n\nIt is less sensitive to extreme cells.\n\n### 26.4. Language floor\n\nIt is the lowest language result with a sufficient sample.\n\nLF_min = min_{l∈L_est} θ_l\n\nThis value should be provided along with small cell uncertainty.\n\n### 26.5. Robust Language floor\n\nIt is the lower percentile or subgroup average of sufficient languages. Example:\n\nLF_10 = Q_{0.10}(θ_l)\n\n### 26.6. Language Critical Error Rate\n\nFor each language:\n\nCR_l = valid response with verified Critical error / valid language response\n\nIt is published separately from the global average.\n\n### 26.7. Unknown Language Rate\n\nU_l = UNKNOWN or unsolvable response / relevant language observation\n\nA high unknown rate can be as important as a low score.\n\n## 27. EQUAL LANGUAGE SCORE AND POPULATION-WEIGHTED SCORE\n\nDiagnostic results that weight all languages equally can be produced:\n\nG_equal-language = (1/L)Σ_{l=1}^L θ_l\n\nThis result approximates the question: \"Average tested language experience.\" It does not represent the average human experience in the world population. Population-weighted language result:\n\nG_population-language = Σ_l W_lθ_l\n\nIt preserves the population distribution. The two scores are different. They cannot be used interchangeably. Both can be shown together on the language fairness card.\n\n## 28. LANGUAGE FAIRNESS IS NOT AN OBLIGATION OF EQUAL OUTCOMES\n\nHaving the same score in two languages is not an automatic violation. The difference can be explained for the following reasons:\n\n- Actual local service difference\n\n- Country-specific law\n\n- Official support coverage of the product\n\n- Resource access\n\n- User intent\n\n- Prompt area\n\nHowever, if the source of the difference cannot be explained or low-resource languages systematically experience:\n\n- incorrect identity,\n\n- incorrect authorisation,\n\n- material deficiency,\n\n- high Critical error\n\nif it carries, there is a language fairness issue. NOMOS defines language fairness as follows:\n\n> In equivalent material conditions, the language itself should not be the cause of systematic misrepresentation.\n\n## 29. LANGUAGE AND COUNTRY INTERSECTION\n\nLanguage score is not independent of countries. The same language:\n\n- different countries,\n\n- different product access,\n\n- different local resources,\n\n- different law,\n\n- different price\n\ncan be used in the context of. Therefore, the country–language cell: (c,l) should be examined separately when necessary. Example:\n\n- Germany × Turkish\n\n- Turkey × Turkish\n\n- United Kingdom × Turkish\n\nIt is the same language but may carry different locale and market context. Language fairness cannot be measured solely by the official language of the country.\n\n## 30th Country Observer AND Language Fairness Panel INTERSECTION\n\nAn observation simultaneously:\n\n- small Country Observer observation,\n\n- low-resource language observation\n\nmay occur. In this case, the record must contain the following fields:\n\n### PRIMARY_PANEL_ROLE\n\n### SECONDARY_DIAGNOSTIC_TAGS\n\npopulation weight Observer status Language fairness status The same observation cannot be fully counted in two separate panels. It is a single observation. It has more than one diagnostic tag.\n\n## 31. CHANGING PARTICIPANTS IN OBSERVER AND LANGUAGE PANEL\n\nA participant can only be changed due to protocol violation. The following outcomes are not reasons for change:\n\n- Incorrect AI response\n\n- Rejection\n\n- Response in another language\n\n- Critical error\n\n- Entity confusion\n\n- Negative recommendation\n\nThese are measurement results. The purpose of the Observer panel is specifically to identify unexpected events. Changing the user because an unexpected event occurred defeats the purpose of the panel.\n\n## 32. LANGUAGE ADJUDICATION\n\nAdjudication in the Language Fairness Panel must at minimum meet the following conditions:\n\n- Being able to understand the prompt and response naturally or at a high proficiency level\n\n- Knowledge of local terminology\n\n- Being able to evaluate sources of evidence\n\n- Using the same normative evaluation guide\n\n- Being blind to the model and, if possible, brand interests\n\n- Disclosing conflict of interest\n\n- Third adjudicator path in case of disagreement\n\nLanguage where no adjudicator can be found:\n\n### ADJUDICATION_LANGUAGE_GAP\n\nmust have the status. It cannot be assumed that all nuances are correct if translated into English by automatic translation. Machine translation can be a helpful tool. It is not the final local meaning verification.\n\n## 33. SOURCE PARITY\n\nOf the entity on the English site:\n\n- scope,\n\n- price,\n\n- licence,\n\n- limits,\n\n- evidence\n\nexists; if these are not present in another language version, the low AI result may not only be a model problem. Three separate results are needed:\n\n- Entity Language Source Gap\n\n- AI Product Language Representation Gap\n\n- Combined User-Facing Gap\n\nThe ultimate problem from the user's perspective is still real. However, the responsibility for correction may vary:\n\n- entity,\n\n- AI provider,\n\nboth.\n\n## 34. CLAIM PARITY\n\nPrompts that are not semantically equivalent produce different answers. Example: In one language: “What does this company do?” In another language: “Is this company trustworthy and recommendable?” If asked, the results cannot be compared. The Prompt Constitution will be detailed in the following section. The provision of this section is as follows:\n\n> Language model performance differences cannot be attributed to AI language performance without verifying prompt equivalence.\n\n## 35. RESPONSE LANGUAGE DEVIATION\n\nThe user gives a prompt in Turkish. AI generates an answer in English. This result:\n\n- even if the factual information is correct,\n\n- may affect user experience,\n\n- accessibility,\n\n- and comprehension.\n\nResponse language deviation must be recorded separately:\n\n### MATCHED_LANGUAGE\n\n### PARTIAL_CODE_SWITCH\n\n### UNREQUESTED_LANGUAGE_SWITCH\n\n### UNUSABLE_LANGUAGE_OUTPUT\n\n### UNKNOWN\n\nCode-switching is not always an error. For example, using technical terms in English may be normal. It is material if it disrupts the main usability of the response.\n\n## 36. WRITING SYSTEM AND TRANSLITERATION\n\nThe same language can be used in different writing systems. For an entity's:\n\n- name,\n\n- product name,\n\n- legal identity\n\nit may be confused with another entity during transliteration. The Language Fairness Panel, to the extent relevant:\n\n- writing system,\n\n- transliteration,\n\n- official local name,\n\n- name used by the user\n\nmust record their fields. Writing system differences should only be considered formal. If it affects entity resolution, it is material.\n\n## 37. SYNTHETIC Country Observer CYCLE\n\nSYNTHETIC METHODOLOGY PRESENTATION / The following product, countries, numbers, and results are fictional. In the registry of the synthetic product called Orion AI Native Reach: there are 120 eligible countries. The main Population Panel has produced valid observations in the first wave from only 38 countries due to population weighting.\n\n### 37.1. First Wave Scope\n\nNumber of eligible countries: 120 Countries with at least one observation: 38\n\nCOC_count = 38/120 = 31.7%\n\nLet these 38 countries carry 97% of the eligible population.\n\nCOC_pop = 97%\n\nResult: The number of countries is low in coverage, population coverage is high.\n\n### 37.2. Four-Wave Observer Cycle\n\nDistribute the remaining 82 countries into four waves using a previously frozen random and risk-based programme. Each country is observed at least once within the cycle. At the end of the fourth wave:\n\nCOC_count = 120/120 = 100%\n\nHowever, sufficient sampling for country estimation has been reached in only 14 countries.\n\n### CEC = 14/120 = 11.7%\n\nCorrect public record: Country observation coverage: 100% / Population observation coverage: close to 100% / Country prediction coverage: 11.7% Incorrect public record: “All 120 countries have a reliable country score.”\n\n### 37.3. Small Country Critical Alarm\n\nIn a Country Observer record, the AI response observed in small country J misidentifies the audited entity as a licensed law firm. In reality, the entity provides no legal services. A single observation does not estimate country J's error rate; it establishes a candidate Critical event. Protocol: validate the observation; verify the prompt and reference record; repeat with new independent users; if the event persists, open a targeted country sample; and assess the Critical conformity impact separately. The observed event may be rare; the potential harm need not be small.\n\n## 38. SYNTHETIC LANGUAGE FAIRNESS DEMONSTRATION\n\n### SYNTHETIC SAMPLE — NOT REAL AI OR COMPANY PERFORMANCE\n\nInclude five languages in the Synthetic Apple.com Core Identity testing. Main Population Panel allocation:\n\nSynthetic language results:\n\nPopulation-weighted global result:\n\n0.55(95)+0.25(92)+0.10(88)+0.06(65)+0.04(58)=90.27\n\nGlobal score is high. However, there is severe degradation in two languages.\n\n### 38.1. Language Fairness Oversampling\n\nAdd 140 additional observations to L4, 160 to L5. New numbers of observations:\n\nThe crude mean across all 1,300 observations is lower than the population-weighted result shown above. It is not a global-population score, because L4 and L5 were intentionally oversampled. Once target-population weights are restored, the global result returns to the population-weighted figure above. Oversampling improves sensitivity for L4 and L5 without changing the population-weighted estimate.\n\n### 38.2. Language Fairness Scorecard\n\nSynthetic result. Population-weighted global score: 90.27. Equal-language score:\n\n(95+92+88+65+58)/5=79.6\n\nHighest-to-lowest language difference: 37 points; lowest language: 58; mean of the bottom two languages: 61.5; Critical-error rate in lower-resource languages: reported separately; prompt equivalence: assumed verified; reference parity: assumed verified. Correct interpretation:\n\n> The global population representation appears high; however, language fairness is seriously weak.\n\nIncorrect interpretation: “Since the global score is 90, GEO is successful in all languages.”\n\n## 39. SYNTHETIC CASE DISTINGUISHING THE SOURCE OF LANGUAGE DETERIORATION\n\nLet's consider three languages.\n\n#### Language A\n\nPrompt equivalent Official source complete Local adjudicator sufficient AI response incorrect Result: Candidate for language representation problem of AI product.\n\n#### Language B\n\nPrompt translation materially different AI response appropriate to prompt Result: Prompt Constitution flaw.\n\n#### Language C\n\nOn the official language page, the old price AI carries over the old price English source is current Result: The entity's language source parity issue and the combined representation gap reaching the user. Even if the scores of the three languages appear the same, the correction owner is different.\n\n## 40. COUNTRY AND LANGUAGE PANELS PUBLIC RESULT CARD\n\nThe candidate public result card must include the following fields:\n\n#### Country Fields\n\nNative Reach appropriate number of countries Number of countries with at least one observation Country observation coverage Population-weighted observation coverage Country prediction coverage Country Observer cycle time Open Observer alerts OUTSIDE_NATIVE_REACH countries NOT OBSERVED countries Measurement date\n\n#### Language Fields\n\nNumber of languages covered Observed number of languages Number of languages for which predictions can be made Officially supported languages Experimental languages Population-weighted global language result Equal language recognition result Language floor Language difference Critical error rates\n\n### REFERENCE_LANGUAGE_GAP\n\n### PROMPT_EQUIVALENCE_FAILURE\n\nNOT TESTED languages Language panel version This card makes visible the reality that a single \"multilingual\" badge cannot carry.\n\n## 41. MANDATORY NORMATIVE PROVISIONS\n\n**CH07-N01**\n\nThe Country Observer and Language Fairness panels must be defined separately from the main Population Panel in terms of purpose, sample, and weighting.\n\n**CH07-N02**\n\nA Country Observer observation cannot be used as a country population score.\n\n**CH07-N03**\n\nFor a country, having an observation and being able to produce a country estimate are separate statuses.\n\n**CH07-N04**\n\nThe number of countries covered and the population-weighted country coverage should be reported together.\n\n**CH07-N05**\n\nCountry estimate coverage should be shown separately from observation coverage.\n\n**CH07-N06**\n\nEach eligible country must have a probability of being observed greater than zero in the defined surveillance cycle; out-of-scope countries should also be shown.\n\n**CH07-N07**\n\nThe surveillance cycle and country frequency classes should be versioned before AI responses are viewed.\n\n**CH07-N08**\n\nA low-scoring country cannot be removed from the Observer cycle after results are seen.\n\n**CH07-N09**\n\nA Country Observer alert is not an automatic country failure; it triggers verification and escalation.\n\n**CH07-N10**\n\nA single Critical Observer event can also initiate a separate conformity review without prevalence estimation.\n\n**CH07-N11**\n\nObservations from the Country Observer Panel cannot be used in the main Population Panel with raw equal weighting.\n\n**CH07-N12**\n\nIf Country Observer observations are included in the main score, their selection probabilities and weights must be explicitly modelled.\n\n**CH07-N13**\n\nLanguage fairness cannot be defined as the obligation to give the same answer word for word.\n\n**CH07-N14**\n\nLanguage fairness requires comparable accuracy, limits, source, and user suitability under equivalent material conditions.\n\n**CH07-N15**\n\nLanguage and country, language and locale, and language and writing system cannot be used interchangeably.\n\n**CH07-N16**\n\nThe language universe covered and the reason for including each language should be explained before the measurement.\n\n**CH07-N17**\n\nThe claim of 'all languages' can only be used if indeed all relevant language and user surfaces have been measured; otherwise, 'languages within scope' should be said.\n\n**CH07-N18**\n\nObservations in a language that is not officially supported cannot be added to the main supported product performance without explanation.\n\n**CH07-N19**\n\nOver-sampling of low-resource languages should not increase the global population weight.\n\n**CH07-N20**\n\nIf observations from the Language Fairness Panel are included in the global result, they should be returned to the actual target language population weights.\n\n**CH07-N21**\n\nThe same multilingual user cannot be counted as more than one full human in the global population weight.\n\n**CH07-N22**\n\nWithout verifying prompt equivalence, the difference in language score cannot be definitively attributed to AI language performance.\n\n**CH07-N23**\n\nA low language score cannot be loaded onto the AI product alone without evaluating the reference parity.\n\n**CH07-N24**\n\nA language accuracy decision cannot be finalised without sufficient local adjudicator competence.\n\n**CH07-N25**\n\nResponse language deviation and code switching must be recorded separately.\n\n**CH07-N26**\n\nLegitimate local reality differences cannot be classified as language injustice.\n\n**CH07-N27**\n\nUnexplained material language differences cannot be closed solely as personalisation or style differences.\n\n**CH07-N28**\n\nGlobal population score cannot replace the result of language fairness.\n\n**CH07-N29**\n\nEqual language score cannot be presented like a population-weighted global score.\n\n**CH07-N30**\n\nLanguage-based Critical error rates must be shown separately from the overall average.\n\n**CH07-N31**\n\nPROMPT_EQUIVALENCE_FAILURE, REFERENCE_LANGUAGE_GAP, MEASUREMENT_PARITY_FAILURE, and ADJUDICATION_LANGUAGE_GAP must be maintained as separate statuses.\n\n**CH07-N32**\n\nIn Country Observer and Language Fairness panels, the decision to change participants cannot be based on the correctness of the AI response or for the benefit of the brand.\n\n**CH07-N33**\n\nThe same observation can have multiple diagnostic labels; however, it cannot receive more than one full weight in the global prediction without explanation.\n\n**CH07-N34**\n\nCountry and language panel records must have an accountable human or institutional owner.\n\n## 42. FORMS OF FAILURE\n\n**CH07-F01 — SINGLE-PERSON COUNTRY SCORE**\n\nA single observer’s observation is published as the score GEO for the country.\n\n**CH07-F02 — SYMBOLIC COUNTRY COVERAGE**\n\nOne person is assigned to each country and it is called 'proportional to the world population.'\n\n**CH07-F03 — CONFUSING POPULATION COVERAGE WITH NUMBER OF COUNTRIES**\n\nAlthough a small number of large countries are observed, only population coverage is shown; although a large number of small countries are observed, only the number of countries is shown.\n\n**CH07-F04 — COUNTING OBSERVATION SCOPE AS ESTIMATION SCOPE**\n\nAll countries with at least one observation are considered scored countries.\n\n**CH07-F05 — COUNTING A Country Observer observation TWICE IN THE MAIN SCORE**\n\nThe same observation receives full weight in both Observer and Population Panel.\n\n**CH07-F06 — CONVERTING A Country Observer alert INTO A CLAIM OF PREVALENCE**\n\nA single event is generalised to all users in the country.\n\n**CH07-F07 — DISREGARDING A CRITICAL OBSERVER EVENT**\n\nThe high risk of harm is not examined by saying “only one user”.\n\n**CH07-F08 — REMOVING A LOW-SCORED COUNTRY FROM THE CYCLE**\n\nThe observer schedule is changed according to the results.\n\n**CH07-F09 — OBSERVING ONLY COMMERCIAL MARKETS**\n\nCountries that have access to products but are considered commercially unimportant are completely excluded; The result is presented globally.\n\n**CH07-F10 — CONSIDERING THE COUNTRY AND THE LANGUAGE THE SAME**\n\nThe official language of the country is accepted as the language of all users.\n\n**CH07-F11 — ALL LANGUAGES CLAIM**\n\nA limited few languages are measured; The result is published as \"in all languages\".\n\n**CH07-F12 — WIPING LOW-WELD TONGUE**\n\nBecause the sample is small, language is completely excluded from the report.\n\n**CH07-F13 — OVERGROWING LOW-RESOURCE LANGUAGE GLOBALLY**\n\nOversampled language enters the global score with its raw sample share.\n\n**CH07-F14 — COUNTING THE EQUAL LANGUAGE SCORE AS THE POPULATION SCORE**\n\nThe diagnostic score that gives equal weight to every language is presented as the average experience of the world population.\n\n**CH07-F15 — COUNTING POPULATION SCORE AS LANGUAGE FAIRNESS**\n\nThe global result, boosted by the weight of large languages, conceals the distortion in small languages.\n\n**CH07-F16 — COUNTING PROMPT TRANSLATION ERROR AS AI ERROR**\n\nThe difference arising from semantically different prompts is attributed to the model.\n\n**CH07-F17 — COUNTING SOURCE LANGUAGE GAP AS AI ERROR**\n\nEven though there is no localised official information, all responsibility is given to the AI product.\n\n**CH07-F18 — DEFENDING AI ERROR WITH RESOURCE LACK**\n\nEven though the same and sufficient resources are available, errors in the low-resource language are closed solely as missing content.\n\n**CH07-F19 — CONSIDERING ADJUDICATOR LANGUAGE INADEQUACY AS PRODUCT FAILURE**\n\nEven if the adjudicator misunderstood, the result is finalised.\n\n**CH07-F20 — CONSIDERING MACHINE TRANSLATION AS FINAL ADJUDICATOR**\n\nLocal meaning loss is not checked.\n\n**CH07-F21 — CONSIDERING CODE CHANGES AS AUTOMATIC ERROR**\n\nFailures occur with natural technical terms or acceptable language mixing.\n\n**CH07-F22 — CONSIDERING UNUSABLE LANGUAGE DEVIATION AS STYLE**\n\nEven if the system responds in another language that the user cannot understand, the result is considered successful.\n\n**CH07-F23 — IGNORING THE WRITING SYSTEM DIFFERENCE**\n\nA transliteration error causing entity confusion is considered formal.\n\n**CH07-F24 — COUNTING A MULTILINGUAL USER TWICE**\n\nThe same person is counted as a separate population unit in all languages.\n\n**CH07-F25 — CONSIDERING LEGITIMATE LOCAL DIFFERENCE AS UNFAIR**\n\nEven if the price or legal difference is based on actual local conditions, it is declared as language bias.\n\n**CH07-F26 — CONSIDERING UNEXPLAINED DIFFERENCE AS PERSONAL**\n\nMaterial inconsistency under equivalent conditions is not questioned.\n\n**CH07-F27 — DEFINITE SCORE FROM INSUFFICIENT LANGUAGE CELL**\n\nLanguage ranking is done with a few observations.\n\n**CH07-F28 — SETTING THE LANGUAGE PANEL AFTER THE RESULT**\n\nOnly languages that scored low or high are selected afterward.\n\n**CH07-F29 — MEASURING ONLY HIGH-SOURCE LANGUAGES**\n\nThe result is presented as \"multilingual\".\n\n**CH07-F30 — PRESERVING UNTESTED LANGUAGES**\n\nThe NOT TESTED field does not appear on the public result card.\n\n## 43. AUDIT PROCEDURE\n\n### Step 1 — Freeze Product-Country Compliance Registry\n\nFor each country:\n\n- Native Reach\n\n- product access\n\n- audit language\n\n- user surface\n\nstatus is recorded.\n\n### Step 2 — Extract Main Population Panel Country Coverage\n\nIt is determined in which countries the main panel observation exists.\n\n### Step 3 — Define the Observer Country Universe\n\nCountries that are insufficient in the main panel and have CE-3 status are identified.\n\n### Step 4 — Establish the Observer Cycle\n\nCountries:\n\n- are assigned in terms of frequency class,\n\n- wave,\n\n- language/locale,\n\n- user surface\n\nand maintenance.\n\n### Step 5 — Preselect Observer Participants\n\nMain and backup lists are created before seeing the results.\n\n### Step 6 — Classify Observer Results\n\nAccess, identity, language, material fact, Critical event statuses are assigned.\n\n### Step 7 — Apply Alarm and Escalation\n\nBy alarm type:\n\n- again,\n\n- targeted sample\n\n- reference review,\n\n- conformity review\n\nis started.\n\n### Step 8 — Calculate Country Coverage Metrics\n\nThe scope of the number of countries, the scope of the population, and the scope of the country estimate are calculated separately.\n\n### Step 9 — Freeze Language Record\n\nFor each language:\n\n- for reasons of scope,\n\n- support status,\n\n- writing system,\n\n- locale,\n\n- user universe\n\nis recorded.\n\n### Step 10 — Establish Over-Sampling for Language fairness\n\nAdditional targets are set for languages that are insufficient on the main population panel.\n\n### Step 11 — Check the Four Language Gates\n\nSystem equivalence Reference parity Measurement parity Adjudication parity is evaluated.\n\n### Step 12 — Score Language Observations\n\nFor each language:\n\n- truthfulness,\n\n- material deficiency,\n\n- Critical error,\n\n- unknown,\n\n- response language deviation\n\nis recorded.\n\n### Step 13 — Separate Global and Diagnostic Weights\n\nThe result weighted by population and the result with equal language recognition are calculated separately.\n\n### Step 14 — Calculate the Language Inequality\n\nlanguage floor, solid language difference, Critical error difference, unknown rate is calculated.\n\n### Step 15 — Examine Country–Language Intersections\n\nMaterial locale differences or conflicts are identified.\n\n### Step 16 — Create Public Result Card\n\nScope, prediction sufficiency, alarms, and untested areas are made visible.\n\n## 44. NECESSARY EVIDENCE\n\nAI-product and user-surface record; Native Reach country register; number of eligible countries; country-access status; principal Population Panel country allocations; Country Observer universe; Observer cycle; Observer frequency classes; country–wave assignment matrix; country–language assignments; primary and reserve Observer lists; country-observation results; Observer alerts; escalation records; targeted country samples; country-observation coverage; population-weighted country coverage; country-estimation coverage; in-scope language register; language-support status; script and locale records; principal language sample counts; Language Fairness oversample counts; language-population weights; multilingual-user records; prompt-equivalence records; reference-parity records; measurement-parity records; local-adjudicator qualification records; and language results.\n\nLanguage Critical error rates Response language deviations Language inequality metrics Untested languages Equal language and population-weighted results Panel weights Double count checks Public results card Panel versions Responsible person or institution\n\n## 45. AUDIT CHECKLIST\n\nHas the purpose of the Country Observer Panel been separated from the main population panel? Which countries are included in the Observer? Is the eligibility and access status of the countries up to date? Is there a probability of positive observation in each eligible country's surveillance cycle? Was the surveillance cycle frozen before the results? Has the Country Observer observation been used like a country score? Have country observation coverage and country estimate coverage been separated? Has the population-weighted country coverage been shown separately? Have Country Observer observations been double-counted in the main score? Has the Critical Observer alert been confirmed? Has a single alarm been converted into a prevalence claim? Was targeted sampling opened due to the alarm? Was the universe of languages in scope predefined? Is the claim of 'all languages' consistent with the actual coverage? Have language, writing system, and locale been separated?\n\nHas the product's language support status been recorded? Why were low-resource languages chosen? Were examples of Language fairness reweighted in the global score? Was the same multilingual user double-counted? Has prompt equivalence been verified? Are official sources materially equal across languages? Do the adjudicators know the relevant language at a sufficient level? Has machine translation been used as the final judge? Have response language deviations been recorded? Has legitimate local variation been separated from language injustice? Has the language corpus and language variance been published? Are language-based critical error rates visible? Have REFERENCE_LANGUAGE_GAP and other measurement flaws been separated? Are untested languages visible? Does the public score hide the language fairness issue?\n\nHas the same observation received full weight across multiple panels? Is the accountable owner of the panel records identified?\n\n## 46. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “Isn’t giving one person to every small country still better than giving nothing?”\n\nFor early observation and geographic coverage, yes. But this person does not measure the country's proportion. The correct term:\n\n#### Country Observer observation\n\nmust be.\n\n### Objection 2 — “If one person does not produce a country score, why are we sending them?”\n\nBecause a single observation can:\n\n- reveal access issues,\n\n- clarify entity confusion,\n\n- identify the wrong language,\n\n- and make Critical events\n\nvisible for the first time. The Observer panel is an early warning system.\n\n### Objection 3 — “Do we have to observe every country in every wave?”\n\nNo. A defined surveillance cycle can be used. High-risk or significant countries can be monitored more frequently; small and low-risk countries can be observed alternately. The schedule should be determined before the results.\n\n### Objection 4 — \"If small countries receive a low weight in the global score, why are we giving it so much attention?\"\n\nBecause their population share may be low. However:\n\n- law,\n\n- language,\n\n- local user harm,\n\n- geographical integrity\n\nare actual human communities. Low weight should not mean invisibility.\n\n### Objection 5 — \"Isn't weighting all languages equally the fairest method?\"\n\nEqual language weighting can be a valuable diagnostic perspective. It does not represent the average human experience in the world population. The population score and the language fairness score should be published separately.\n\n### Objection 6 — “If a small language only slightly affects the global result, why does a low score matter?”\n\nBecause the representation problem is real for people who use that language. Also, severe errors in licences, health, or identity are not automatically trivialised by population share.\n\n### Objection 7 — “Isn’t it unfair to blame AI if a low-resource language has few resources?”\n\nIt may only be unfair to blame AI. Therefore:\n\n- resource parity,\n\n- prompt equivalence,\n\n- adjudicator competence\n\nexamined separately. The error that reaches the user is still factually correct. The owner of the correction should be determined correctly.\n\n### Objection 8 — “Isn’t it normal to have different answers in the same language in different countries?”\n\nIt may be normal if the real local conditions are different. If the same material fact changes differently and without explanation, it may be a problem. Language and locale should be examined together.\n\n### Objection 9 — “If AI gives the correct answer in another language, why would it fail?”\n\nIf the user cannot understand the response, representation cannot be used. Language deviation should be evaluated separately along with accuracy.\n\n### Objection 10 — “Finding 200 users for language scores can be very expensive.”\n\nA smaller sample can be used. Result:\n\n- discovery,\n\n- temporary,\n\n- insufficient sample\n\nshould be published with this status. Cost does not grant false certainty.\n\n### Objection 11 — “If the global score is already high, doesn't a separate language card unnecessarily scare the customer?”\n\nHiding material language deterioration does not protect the customer. It produces false confidence. A short and clear results card can be used.\n\n### Objection 12 — “Does a single Critical Observer event reduce all eligibility?”\n\nThe automatic response depends on the type of event. A single event does not indicate national prevalence. However, if it carries severe damage or false authority, it can initiate Critical review and temporary measures. The decision will be defined in Section 15.\n\n## COMMON RULE OF SECTION 49\n\nCreating a world score is easy. You fill the world population with large countries. You lose small countries in fractions. You measure high-resource languages. Low-resource languages: you remove by saying \"Sample insufficient.\" Then you publish a single number. This number may be mathematically calculated correctly. But it may not show all the representation flaws of the world. NOMOS does not give equal weight to every country. Because people's population shares are not equal. NOMOS does not ignore small countries either. Because low population does not mean nonexistence. NOMOS does not weight all languages equally in the global score. Because this could distort the average experience of the world population. NOMOS does not make small languages invisible either. Because even if their global weight is low, the representation error for the people using that language can be 100% real.\n\nCountry Observer Panel:\n\n> Shows where we have never looked.\n\nLanguage Fairness Panel:\n\n> Shows who is below the global average.\n\nPopulation Panel:\n\n> Estimates the average experience in the world population.\n\nWhen these three work together, the global score becomes both faithful to the population and auditable in terms of justice. A single user in a country is not the whole country. But that user is a real person. A language may be a small part of the global population. But the error in that language is not less wrong. A Critical event may be rare. But harm is not measured solely by frequency. Therefore, NOMOS's seventh measurement law is:\n\n> The right to visibility is not the same as population weight.\n\nIts eighth law is as follows:\n\n> Observing a country is not the same as estimating a country.\n\nThe ninth law is as follows:\n\n> Language fairness is not making all languages equal; it is ensuring that no one is systematically distanced from material reality because of their language.\n\nThe tenth law is:\n\n> If the global average makes the language and country that are behind invisible, it alone is not a sufficient GEO result.\n\n## NOMOS’s Section 7 Order\n\n> Do not hide countries you never look at while showing me the world score.\n\n> Do not send one user to a small country and make that person the whole country.\n\n> But also do not ignore a small country forever just because its population is small.\n\n> Separate country observation, country prediction, and country appropriateness.\n\n> Do not judge the whole country based on a single severe mistake. / But do not ignore human harm by saying \"only one answer.\"\n\n> Do not base success in major languages on the silence of minor languages.\n\n> You can sample a low-resource language a lot. / But do not make it appear larger than it is in the world population.\n\n> Do not make the equal language score a population score; do not make the population score a language fairness score.\n\n> Do not blame me if the translation is wrong. / If the source is incomplete, also say this. / If the adjudicator doesn't understand the language, do not make a definitive judgement.\n\n> Do not classify a legitimate local difference as unfairness, or dismiss an unexplained material disparity as mere style.\n\n> First, observe the countries. / Then state where you can support an estimate. / Then oversample the languages that require diagnostic visibility. / Then return them to their true population weights. / Keep visible the people whom the global average leaves behind.\n\n## The Chapter's Closing Sentence\n\nA genuinely global GEO measurement must show more than what most of the world sees. It must also show who remains unobserved in smaller countries and languages, who is misrepresented and which result is only a single alert.\n\n## Normative Core\n\n> The GEO-1000 Population Panel, Country Observer Panel and Language Fairness Panel MUST remain distinct in purpose, sampling, weighting and public claims. A Country Observer observation MAY establish that an access condition, response, anomaly or Critical incident occurred in a defined country–product–language context. A Country Observer observation MUST NOT be represented as a country-prevalence estimate or country GEO score. Country-observation coverage, population-weighted observation coverage and country-estimation coverage MUST be reported separately. Every eligible country SHOULD receive a positive probability of observation within a predeclared surveillance cycle unless it falls outside the product's lawful Native Reach. Language fairness MUST evaluate materially comparable representation under semantically equivalent prompts and comparable contexts. It MUST NOT require word-for-word identical responses or disregard legitimate local differences. Oversampled languages MUST be returned to their target-population weights for global population estimates. Population-weighted scores, equal-language diagnostics, language floors, Critical-error rates, unknown rates and untested languages MUST remain separately visible. A language disparity MUST NOT be attributed solely to the AI product until prompt equivalence, reference parity, measurement parity and adjudication parity have been assessed. One observation MAY carry multiple diagnostic panel tags, but MUST NOT receive duplicate full population weight. Critical Observer incidents MAY trigger confirmatory sampling and conformity review without being treated as evidence of country prevalence.","character_count":64015,"record_sha256":"d250d8f01cc535b9f27e3bbe357f686a9e9f7fd0668c0175151119b5e70053df"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-08","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":10,"chapter_number":8,"item_number":null,"title":"Participant Selection, Weighting and Sampling Bias","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K05","K06","K07","K08","K22"],"source_word_count":8675,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":8,\"id\":\"NOMOS-GEO-AUDIT-CH08\",\"title\":\"Participant Selection, Weighting and Sampling Bias\",\"subtitle\":null,\"sourceFile\":\"8.ci bölüm.docx\",\"sourceSha256\":\"37EF84DC45A748DD5560417EF00D220C5B2B369918BD66A20DF507281DE5BB19\",\"sourceWordCount\":8675,\"sourceIds\":[\"K04\",\"K05\",\"K06\",\"K07\",\"K08\",\"K22\"],\"machine\":{\"chapter\":8,\"chapterId\":\"NOMOS-GEO-AUDIT-CH08\",\"title\":\"Participant Selection, Weighting and Sampling Bias\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K05\",\"K06\",\"K07\",\"K08\",\"K22\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH08-R01\",\"normativeRuleEnglish\":\"Every GEO-1000 observation MUST retain an auditable participant-selection path from the target population and sampling frame through selection, invitation, consent, assignment, completion, protocol validation, exclusion, and final weighting. Within-panel random selection MUST NOT be represented as random selection from the global target population unless the full population inclusion mechanism supports that claim. Participant recruitment, assignment, compensation, validation, replacement, exclusion, and weighting MUST remain independent of whether the AI response is positive, negative, correct, incorrect, refused, critical, or commercially favourable. Persons, accounts, devices, sessions, languages, panel memberships, and AI-product observations MUST remain distinct units. Probability, opt-in, open-call, client-supplied, institutional, and hybrid participant sources MUST be separately identified. Weighting MAY reduce observed imbalance, but MUST NOT be represented as creating observations for populations absent from the sampling frame or correcting unmeasured selection with certainty. Unweighted results, weighted results, weight diagnostics, coverage gaps, nonresponse, effective sample size, and sensitivity analyses MUST remain visible. Fraud and duplication decisions MUST use proportionate multi-signal review and MUST NOT automatically treat VPN use, shared devices, institutional networks, or accessibility tools as fabrication. Participant evidence MUST remain verifiable without unnecessarily disclosing identity, account credentials, private conversations, or precise location. Every participant-selection and weighting decision MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Her GEO-1000 gözlemi hedef nüfus ve örnekleme çerçevesinden seçim, davet, rıza, görev ataması, tamamlama, protokol doğrulaması, dışlama ve nihai ağırlığa kadar denetlenebilir katılımcı seçim yolu taşımalıdır. Panel içindeki rastgele seçim, tam nüfus katılım mekanizması bunu desteklemedikçe küresel hedef nüfustan rastgele seçim olarak sunulamaz. Katılımcının seçilmesi, ödüllendirilmesi, doğrulanması, değiştirilmesi, dışlanması ve ağırlıklandırılması AI cevabının olumlu, olumsuz, doğru, yanlış, ret, Critical veya ticari bakımdan elverişli olmasına bağlı olamaz. Ağırlıklandırma gözlenen dengesizliği azaltabilir; örnekleme çerçevesinde hiç bulunmayan insanları gözlenmiş hâle getiremez.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH08-MB0001\",\"type\":\"paragraph\",\"text\":\"57. MACHINE-READABLE PARTICIPANT RECORD\",\"sourceParagraph\":1804},{\"blockId\":\"CH08-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1805},{\"blockId\":\"CH08-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1806},{\"blockId\":\"CH08-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"adultBaseline\\\": true,\",\"sourceParagraph\":1807},{\"blockId\":\"CH08-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"consentStatus\\\": \\\"GRANTED\\\",\",\"sourceParagraph\":1808},{\"blockId\":\"CH08-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"country\\\": {\",\"sourceParagraph\":1809},{\"blockId\":\"CH08-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"residenceCountry\\\": \\\"TR\\\",\",\"sourceParagraph\":1810},{\"blockId\":\"CH08-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"measurementCountry\\\": \\\"TR\\\",\",\"sourceParagraph\":1811},{\"blockId\":\"CH08-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"verificationLevel\\\": \\\"GV-3\\\",\",\"sourceParagraph\":1812},{\"blockId\":\"CH08-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"vpnStatus\\\": \\\"NOT_DETECTED\\\"\",\"sourceParagraph\":1813},{\"blockId\":\"CH08-MB0011\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1814},{\"blockId\":\"CH08-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": {\",\"sourceParagraph\":1815},{\"blockId\":\"CH08-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"assignedLanguage\\\": \\\"tr\\\",\",\"sourceParagraph\":1816},{\"blockId\":\"CH08-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"tr-TR\\\",\",\"sourceParagraph\":1817},{\"blockId\":\"CH08-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"competenceStatus\\\": \\\"VERIFIED_FUNCTIONAL\\\"\",\"sourceParagraph\":1818},{\"blockId\":\"CH08-MB0016\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1819},{\"blockId\":\"CH08-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"aiAccess\\\": {\",\"sourceParagraph\":1820},{\"blockId\":\"CH08-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"productId\\\": \\\"SYNTH-ORION-WEB\\\",\",\"sourceParagraph\":1821},{\"blockId\":\"CH08-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"plan\\\": \\\"free\\\",\",\"sourceParagraph\":1822},{\"blockId\":\"CH08-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"surface\\\": \\\"web\\\",\",\"sourceParagraph\":1823},{\"blockId\":\"CH08-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"accessVerified\\\": true,\",\"sourceParagraph\":1824},{\"blockId\":\"CH08-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"sharedAccount\\\": false\",\"sourceParagraph\":1825},{\"blockId\":\"CH08-MB0023\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1826},{\"blockId\":\"CH08-MB0024\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1827},{\"blockId\":\"CH08-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"relationships\\\": {\",\"sourceParagraph\":1828},{\"blockId\":\"CH08-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"auditedEntityRelationship\\\": \\\"GENERAL_PUBLIC\\\",\",\"sourceParagraph\":1829},{\"blockId\":\"CH08-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"aiProviderRelationship\\\": \\\"NONE_DECLARED\\\",\",\"sourceParagraph\":1830},{\"blockId\":\"CH08-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"brandFamiliarity\\\": \\\"AWARE_NON_CUSTOMER\\\",\",\"sourceParagraph\":1831},{\"blockId\":\"CH08-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"aiUsageFrequency\\\": \\\"OCCASIONAL\\\"\",\"sourceParagraph\":1832},{\"blockId\":\"CH08-MB0030\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1833},{\"blockId\":\"CH08-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"assignment\\\": {\",\"sourceParagraph\":1834},{\"blockId\":\"CH08-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"W1\\\",\",\"sourceParagraph\":1835},{\"blockId\":\"CH08-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"stratumId\\\": \\\"TR-TR-FREE-WEB\\\",\",\"sourceParagraph\":1836},{\"blockId\":\"CH08-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"productAssignmentMethod\\\": \\\"BLOCK_RANDOMIZED\\\",\",\"sourceParagraph\":1837},{\"blockId\":\"CH08-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"promptId\\\": \\\"CORE-IDENTITY-001\\\",\",\"sourceParagraph\":1838},{\"blockId\":\"CH08-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"promptOrder\\\": 1,\",\"sourceParagraph\":1839},{\"blockId\":\"CH08-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"priorBenchmarkExposure\\\": false\",\"sourceParagraph\":1840},{\"blockId\":\"CH08-MB0038\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1841},{\"blockId\":\"CH08-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"incentive\\\": {\",\"sourceParagraph\":1842},{\"blockId\":\"CH08-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"type\\\": \\\"FIXED_COMPLETION_PAYMENT\\\",\",\"sourceParagraph\":1843},{\"blockId\":\"CH08-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"responseContentDependent\\\": false,\",\"sourceParagraph\":1844},{\"blockId\":\"CH08-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"speedBonus\\\": false\",\"sourceParagraph\":1845},{\"blockId\":\"CH08-MB0043\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1846},{\"blockId\":\"CH08-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"completion\\\": {\",\"sourceParagraph\":1847},{\"blockId\":\"CH08-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"invited\\\": true,\",\"sourceParagraph\":1848},{\"blockId\":\"CH08-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"consented\\\": true,\",\"sourceParagraph\":1849},{\"blockId\":\"CH08-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"started\\\": true,\",\"sourceParagraph\":1850},{\"blockId\":\"CH08-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"completed\\\": true,\",\"sourceParagraph\":1851},{\"blockId\":\"CH08-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"protocolValidity\\\": \\\"VALID\\\",\",\"sourceParagraph\":1852},{\"blockId\":\"CH08-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"semanticOutcomeUsedForValidityDecision\\\": false\",\"sourceParagraph\":1853},{\"blockId\":\"CH08-MB0051\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1854},{\"blockId\":\"CH08-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":1855},{\"blockId\":\"CH08-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"fraudStatus\\\": \\\"FI-1\\\",\",\"sourceParagraph\":1856},{\"blockId\":\"CH08-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"duplicateCheck\\\": \\\"PASSED\\\",\",\"sourceParagraph\":1857},{\"blockId\":\"CH08-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"crossProviderDuplicateCheck\\\": \\\"PASSED\\\",\",\"sourceParagraph\":1858},{\"blockId\":\"CH08-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"humanReviewRequired\\\": false\",\"sourceParagraph\":1859},{\"blockId\":\"CH08-MB0057\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1860},{\"blockId\":\"CH08-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"weights\\\": {\",\"sourceParagraph\":1861},{\"blockId\":\"CH08-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"baseWeightType\\\": \\\"MODEL_CALIBRATION_WEIGHT\\\",\",\"sourceParagraph\":1862},{\"blockId\":\"CH08-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"designProbabilityKnown\\\": false,\",\"sourceParagraph\":1863},{\"blockId\":\"CH08-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"nonresponseAdjustment\\\": 1.08,\",\"sourceParagraph\":1864},{\"blockId\":\"CH08-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"calibrationAdjustment\\\": 1.24,\",\"sourceParagraph\":1865},{\"blockId\":\"CH08-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"finalWeight\\\": 1.3392,\",\"sourceParagraph\":1866},{\"blockId\":\"CH08-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"outcomeUsedInWeighting\\\": false\",\"sourceParagraph\":1867},{\"blockId\":\"CH08-MB0065\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1868},{\"blockId\":\"CH08-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"privacy\\\": {\",\"sourceParagraph\":1869},{\"blockId\":\"CH08-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"publicIdentityDisclosure\\\": false,\",\"sourceParagraph\":1870},{\"blockId\":\"CH08-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"screenshotRedactionRequired\\\": true,\",\"sourceParagraph\":1871},{\"blockId\":\"CH08-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"publicEvidenceManifest\\\": true,\",\"sourceParagraph\":1872},{\"blockId\":\"CH08-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"rawEvidenceAccess\\\": \\\"CONTROLLED_AUDIT_ACCESS\\\"\",\"sourceParagraph\":1873},{\"blockId\":\"CH08-MB0071\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1874},{\"blockId\":\"CH08-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":1875},{\"blockId\":\"CH08-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"selectionOwnerRole\\\": \\\"PARTICIPANT_SELECTION_LEAD\\\",\",\"sourceParagraph\":1876},{\"blockId\":\"CH08-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"validityReviewerRole\\\": \\\"BLINDED_PROTOCOL_VALIDATOR\\\"\",\"sourceParagraph\":1877},{\"blockId\":\"CH08-MB0075\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1878},{\"blockId\":\"CH08-MB0076\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1879},{\"blockId\":\"CH08-MB0077\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1880},{\"blockId\":\"CH08-MB0078\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1881},{\"blockId\":\"CH08-MB0079\",\"type\":\"paragraph\",\"text\":\"It does not belong to a real participant,\",\"sourceParagraph\":1882},{\"blockId\":\"CH08-MB0080\",\"type\":\"paragraph\",\"text\":\"It is not real panel data,\",\"sourceParagraph\":1883},{\"blockId\":\"CH08-MB0081\",\"type\":\"paragraph\",\"text\":\"It is a synthetic machine diagram of the participant selection chain.\",\"sourceParagraph\":1884},{\"blockId\":\"CH08-MB0082\",\"type\":\"paragraph\",\"text\":\"Chapter 58 machine-readable rule\",\"sourceParagraph\":1886},{\"blockId\":\"CH08-MB0083\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH08-R01\",\"sourceParagraph\":1887},{\"blockId\":\"CH08-MB0084\",\"type\":\"paragraph\",\"text\":\"Every GEO-1000 observation MUST retain an auditable participant-selection\",\"sourceParagraph\":1889},{\"blockId\":\"CH08-MB0085\",\"type\":\"paragraph\",\"text\":\"path from the target population and sampling frame through selection,\",\"sourceParagraph\":1890},{\"blockId\":\"CH08-MB0086\",\"type\":\"paragraph\",\"text\":\"invitation, consent, assignment, completion, protocol validation,\",\"sourceParagraph\":1891},{\"blockId\":\"CH08-MB0087\",\"type\":\"paragraph\",\"text\":\"exclusion, and final weighting.\",\"sourceParagraph\":1892},{\"blockId\":\"CH08-MB0088\",\"type\":\"paragraph\",\"text\":\"Within-panel random selection MUST NOT be represented as random selection\",\"sourceParagraph\":1894},{\"blockId\":\"CH08-MB0089\",\"type\":\"paragraph\",\"text\":\"from the global target population unless the full population inclusion\",\"sourceParagraph\":1895},{\"blockId\":\"CH08-MB0090\",\"type\":\"paragraph\",\"text\":\"mechanism supports that claim.\",\"sourceParagraph\":1896},{\"blockId\":\"CH08-MB0091\",\"type\":\"paragraph\",\"text\":\"Participant recruitment, assignment, compensation, validation,\",\"sourceParagraph\":1898},{\"blockId\":\"CH08-MB0092\",\"type\":\"paragraph\",\"text\":\"replacement, exclusion, and weighting MUST remain independent of whether\",\"sourceParagraph\":1899},{\"blockId\":\"CH08-MB0093\",\"type\":\"paragraph\",\"text\":\"the AI response is positive, negative, correct, incorrect, refused,\",\"sourceParagraph\":1900},{\"blockId\":\"CH08-MB0094\",\"type\":\"paragraph\",\"text\":\"critical, or commercially favourable.\",\"sourceParagraph\":1901},{\"blockId\":\"CH08-MB0095\",\"type\":\"paragraph\",\"text\":\"Persons, accounts, devices, sessions, languages, panel memberships, and\",\"sourceParagraph\":1903},{\"blockId\":\"CH08-MB0096\",\"type\":\"paragraph\",\"text\":\"AI-product observations MUST remain distinct units.\",\"sourceParagraph\":1904},{\"blockId\":\"CH08-MB0097\",\"type\":\"paragraph\",\"text\":\"Probability, opt-in, open-call, client-supplied, institutional, and hybrid\",\"sourceParagraph\":1906},{\"blockId\":\"CH08-MB0098\",\"type\":\"paragraph\",\"text\":\"participant sources MUST be separately identified.\",\"sourceParagraph\":1907},{\"blockId\":\"CH08-MB0099\",\"type\":\"paragraph\",\"text\":\"Weighting MAY reduce observed imbalance, but MUST NOT be represented as\",\"sourceParagraph\":1909},{\"blockId\":\"CH08-MB0100\",\"type\":\"paragraph\",\"text\":\"creating observations for populations absent from the sampling frame or\",\"sourceParagraph\":1910},{\"blockId\":\"CH08-MB0101\",\"type\":\"paragraph\",\"text\":\"correcting unmeasured selection with certainty.\",\"sourceParagraph\":1911},{\"blockId\":\"CH08-MB0102\",\"type\":\"paragraph\",\"text\":\"Unweighted results, weighted results, weight diagnostics, coverage gaps,\",\"sourceParagraph\":1913},{\"blockId\":\"CH08-MB0103\",\"type\":\"paragraph\",\"text\":\"nonresponse, effective sample size, and sensitivity analyses MUST remain\",\"sourceParagraph\":1914},{\"blockId\":\"CH08-MB0104\",\"type\":\"paragraph\",\"text\":\"visible.\",\"sourceParagraph\":1915},{\"blockId\":\"CH08-MB0105\",\"type\":\"paragraph\",\"text\":\"Fraud and duplication decisions MUST use proportionate multi-signal\",\"sourceParagraph\":1917},{\"blockId\":\"CH08-MB0106\",\"type\":\"paragraph\",\"text\":\"review and MUST NOT automatically treat VPN use, shared devices,\",\"sourceParagraph\":1918},{\"blockId\":\"CH08-MB0107\",\"type\":\"paragraph\",\"text\":\"institutional networks, or accessibility tools as fabrication.\",\"sourceParagraph\":1919},{\"blockId\":\"CH08-MB0108\",\"type\":\"paragraph\",\"text\":\"Participant evidence MUST remain verifiable without unnecessarily\",\"sourceParagraph\":1921},{\"blockId\":\"CH08-MB0109\",\"type\":\"paragraph\",\"text\":\"disclosing identity, account credentials, private conversations, or\",\"sourceParagraph\":1922},{\"blockId\":\"CH08-MB0110\",\"type\":\"paragraph\",\"text\":\"precise location.\",\"sourceParagraph\":1923},{\"blockId\":\"CH08-MB0111\",\"type\":\"paragraph\",\"text\":\"Every participant-selection and weighting decision MUST be versioned and\",\"sourceParagraph\":1925},{\"blockId\":\"CH08-MB0112\",\"type\":\"paragraph\",\"text\":\"attributable to an accountable human or organisation.\",\"sourceParagraph\":1926},{\"blockId\":\"CH08-MB0113\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1927},{\"blockId\":\"CH08-MB0114\",\"type\":\"paragraph\",\"text\":\"Every GEO-1000 observation must carry a verifiable participant selection path from target population and sampling frame through selection, invitation, consent, assignment, completion, protocol verification, exclusion, and final weighting. Random selection within the panel cannot be presented as random selection from the global target population unless the full population participation mechanism supports it. The participant's selection, reward, verification, modification, exclusion, and weighting cannot depend on whether the AI response is positive, negative, correct, incorrect, rejected, Critical, or commercially viable. Weighting can reduce observed imbalance; it cannot make people who are completely absent from the sampling frame appear observed.\",\"sourceParagraph\":1928}]}}","text":"## Chapter Boundary\n\nChapter 5 defined the target user population. Chapter 6 transformed this population into a sample design distributed across countries, languages, and user surfaces. Chapter 7 established complementary observation systems for small countries and low-resource languages that the main population panel might leave invisible. Now the numbers in the sample plan need to turn into real people:\n\n> Where will these participants be found, how will they be selected, under what conditions will they be admitted to the panel, and with what biases will they enter the GEO score?\n\nFrom an online panel of a research organisation? From an open social media call? From universities? From the client's own user list? From local research partners? Only from people who frequently use AI products? From those accessing the product for the first time? From the brand's customers? From competitors' employees? From paid plan subscribers? How will it be verified that a person actually lives in that country? How many times will the same person be counted if they use multiple accounts? How much will the participant be paid? Will the payment encourage completing the task quickly or submitting a specific outcome? Will the observation be preserved if the participant uploads an incorrect AI response? Or will it be replaced with a new user by saying: 'The task was done incorrectly.'? A thousand valid screenshots may have been collected. But the participants:\n\n- 60% are heavy AI users,\n\n- 20% are technology students,\n\n- 10% are customers of the audited brand,\n\n- 5% are company employees,\n\n- the rest are general users\n\nWhich population, then, will the result represent? Can demographic weighting fully correct the difference? Matching age, gender, country and language totals does not necessarily correct unmeasured differences in technological interest, brand awareness, paid-plan access or patterns of AI use. Weighting is valuable, but it cannot erase every defect in participant selection. This chapter applies the earlier requirement—that each audit control be testable by model, date, country, language, query set, repetition count and measurement record—to the selection of actual participants. It also preserves NOMOS's central demand: evidence, boundary, context and time. This chapter:\n\n- participant selection sources,\n\n- the distinction between probabilistic and non-probabilistic panels,\n\n- eligibility and identity verification,\n\n- country, language, and product access control,\n\n- participant uniqueness,\n\n- rewarding and redirection risk,\n\n- panel conditioning,\n\n- task assignment,\n\n- non-response and panel attrition,\n\n- fraudulent or automated participation,\n\n- selection and coverage bias,\n\n- the power and limit of weighting,\n\n- sample quality statuses\n\ndefines. This chapter does not yet:\n\n- all technical areas of AI product registry,\n\n- the full translation and version system of prompts,\n\n- The detailed architecture of the NOMOS Capture infrastructure,\n\n- atomic claim adjudication,\n\n- the final score formula\n\nis not fully defined. The key question of Section 8 is:\n\n> When we collect a thousand valid records, how do we prove which real people these records belong to and which human population they can represent?\n\n## NOMOS Challenge\n\nYou purchase 1,000 people from an online research panel. The panel provider tells you: “1,000 participants ready according to your country and age quotas.” The participants complete the task. They upload screenshots. Raw results:\n\n> 90 per cent accurate representation.\n\nIn your report, you write: “90% of users worldwide are accurately represented.” However, the actual distribution of participants is as follows:\n\nHeavy AI users:\n\n- better prompt habits,\n\n- more paid accounts,\n\n- newer product surfaces,\n\n- personalisation knowledge,\n\n- higher familiarity with technology\n\ncan have. Even if the same locked prompt is used:\n\n- account plan,\n\n- interface,\n\n- previous usage,\n\n- product access\n\nmay be different. Let the synthetic group pass rates be as follows:\n\nThe raw result of the panel:\n\n0.10(72)+0.30(84)+0.60(96)=90\n\nthat is:\n\n#### 90 per cent\n\nWith target-population weights, however, the result is:\n\n0.50(72)+0.30(84)+0.20(96)=80.4\n\nThe same 1,000 responses therefore produce two different results:\n\n- Raw panel result: 90 per cent\n\n- Target-population result: 80.4 per cent\n\nNow imagine that you have weighted it. Each of the 100 people in the low-usage group receives a high weight. Each of the 600 people in the high-usage group receives a low weight. The raw sample size is still 1,000. However, the effective sample size drops to approximately 349. So, you may have collected 1,000 files. Statistically, you may not have the information volume of 1,000 equal and independent people. Now a more difficult question: In the low-usage group, people who have never been included in the panel:\n\n- low-connected,\n\n- using assistive technology,\n\n- living in a small city,\n\n- not having a specific device\n\nIf there are people, can weighting bring them back? No. An observation of a person who is not in the sampling frame at all cannot be produced by weighting. Now the audited company sends connections to its own customers: 'Please participate in the survey.' The company's loyal customers participate at a high rate. People who know competitors but do not use the company participate less. Age and country weights may be correct. Brand loyalty bias may continue. Now another problem: The participant is told: 'You will earn a bonus if you upload a correct and high-quality AI response.' The participant revises the response. They choose the best result. They crop the screenshot. This is no longer the natural response of the AI product. It is the selected response created by the reward system. The first rule of this section is:\n\n> The number of participants is not the representative power of the participants.\n\nIts second provision states:\n\n> Random assignment is not random population selection.\n\nIts third provision states:\n\n> Weighting can reduce observed imbalance; it cannot create an unobserved person.\n\nIts fourth provision states:\n\n> A participant's inclusion in the panel, retention in the task, or replacement cannot depend on the content of the AI response.\n\nIts fifth provision states:\n\n> The reward should be given for completing the protocol honestly, not for choosing a positive or desired response.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to make the entire selection chain from the target population to actual user observation visible, unbiased, and auditable. The section normatively distinguishes:\n\n- Target population and sampling frame\n\n- Sampling frame versus participant source\n\n- Selected person versus invited person\n\n- The invited person and the person who agrees to participate\n\n- The participant and the person who completes the task\n\n- The person who completes the task and the valid observation\n\n- Human and account\n\n- Human and device\n\n- Human and session\n\n- Probabilistic panel and volunteer panel\n\n- Within-panel random selection and random selection from the population\n\n- Quota compliance and true representation\n\n- Demographic balance and behavioural balance\n\n- Personal familiarity and target population suitability\n\n- Customer, employee, and general user\n\n- Participant reward and outcome incentive\n\n- Task speed and task integrity\n\n- Invalid observation with non-response\n\n- Population change with panel loss\n\n- Fake participant and real low-quality observation\n\n- Data fraud and incorrect AI response\n\n- Bias correction with weighting\n\n- Evidence production with calibration\n\n- Raw sample size vs. effective sample size\n\n- Participant privacy and audit evidence\n\n- General population panel and expert or client panel\n\nAt the end of this section, each GEO-1000 audit should be able to answer the following questions:\n\n> From which source and by what rule were the participants selected?\n\n> Who had no chance of joining the panel?\n\n> Who was more likely to participate and why?\n\n> Which known differences do the weights correct, and which do they not correct?\n\n> Was an invalid or fake participation decision made without seeing the content of the AI response?\n\n> Did the participants' reward affect their behaviour and choice of answers?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEach GEO-1000 observation must carry a traceable participant selection chain from the target population to a valid analytic record; selection, invitation, acceptance, task assignment, completion, verification, exclusion, and weighting decisions must be independent of the positive or negative content of the AI response. [K04–K06] A participant panel only:\n\n- country,\n\n- does not imply any suitability in terms of age,\n\n- language\n\nIt cannot be considered suitable for population representation because it meets the quotas. The panel should also carry the following information to the extent relevant:\n\n- Participant source\n\n- Selection mechanism\n\n- Method of participation in the panel\n\n- Eligibility criteria\n\n- Unique human verification\n\n- Country and language verification\n\n- AI product access verification\n\n- Account and plan information\n\n- Device and interface\n\n- Frequency of AI usage\n\n- Brand familiarity\n\n- Relationship with the audited entity\n\n- Incentive structure\n\n- Task assignment method\n\n- Non-response\n\n- Invalidity\n\n- Repetition and panel conditioning\n\n- Weighting\n\n- Lack of coverage\n\n- Privacy and consent\n\nMeasuring some of this information may affect the participant. Therefore, the order of data collection is also part of the protocol.\n\n## 3. PARTICIPANT SELECTION CHAIN\n\nThe chain from the target population to the final observation consists of the following stages:\n\n### U→F→S→I→E→A→C→V\n\nHere:\n\n- U: target user population\n\n- F: individuals included in the sampling frame\n\n- S: selected people\n\n- I: invited persons\n\n- E: individuals whose eligibility has been confirmed and who have been included in the study\n\n- A: people assigned to the task\n\n- C: people who complete the task\n\n- V: people producing valid analytical observations\n\nAt each transition, people may be lost or new bias may emerge. For example:\n\n- Those in the target population who are not on the panel\n\n- Those on the panel who are not invited\n\n- Those invited who do not accept\n\n- Those who accept but do not complete the task\n\n- Those who complete but produce invalid records\n\nare not the same people. Therefore, only the last 1,000 valid observations cannot be reported. The entire selection funnel must be visible.\n\n## 4. PARTICIPANT SOURCES\n\nGEO-1000 does not have to be tied to a single participant source. Each source carries its own bias and scope.\n\n### 4.1. Probability Population Panel\n\nParticipants have a known and non-zero probability of being selected from a defined population. Sources:\n\n- household,\n\n- address,\n\n- population or eligible record,\n\n- probability telephone or multi-stage selection\n\nmay be. Advantages:\n\n- design weight can be established,\n\n- population inference is stronger,\n\nselection process can be more transparent. Limitations:\n\n- can be expensive on a global scale,\n\n- In some countries, there may not be a suitable frame,\n\n- It may also be necessary to find the subpopulation with access to AI products,\n\nAgain, there will be nonresponse. Probabilistic selection is not automatically a perfect representation.\n\n### 4.2. Professional Volunteer Research Panel\n\nParticipants have previously registered voluntarily for the research panel. Advantages:\n\n- it is fast,\n\n- can access multiple countries,\n\n- can provide profile and language knowledge,\n\nrepeat measurement can be performed. Limitations:\n\n- participants are prone to research,\n\n- the level of technology use may be high,\n\n- heavy panel users may be present,\n\n- probabilities of entry into the population may be unknown,\n\nthe same person may be registered on more than one panel. This panel:\n\n#### cannot be presented as a probabilistic world population sample\n\nIt must not be represented as a probability sample of the global population.\n\n### 4.3. Open Online Call\n\nParticipation is obtained through social media, a website, or an open registration form. Advantages:\n\n- wide visibility,\n\n- low cost,\n\n- access to small countries and language communities\n\ncan be provided. Limitations:\n\n- strong self-selection,\n\n- concentration of brand fans or critics,\n\n- fake and repeat participation,\n\n- over-representation of technology enthusiasts\n\ncan be created. An open call may be much more suitable than the main Population Panel for:\n\n- discovery,\n\n- Country Observer,\n\n- Language Fairness,\n\n- replication challenge\n\nThese sources may be better suited to those purposes.\n\n### 4.4. University or Institution Affiliated Panel\n\nParticipants:\n\n- are selected through universities,\n\n- public institution,\n\n- research centres,\n\n- professional organisations\n\nAdvantages:\n\n- local access,\n\n- language expertise,\n\n- ethical oversight,\n\n- reaching low-resource communities\n\ncan be provided. Limitations:\n\n- student or professional intensity,\n\n- institution effect,\n\n- age and education bias,\n\n- corporate participation pressure\n\ncan create.\n\n### 4.5. Customer or User List\n\nThe actual customers or users of the audited institution are selected. This source:\n\n- customer experience,\n\n- suitability for service,\n\n- brand awareness\n\nIt is valuable for measurement. It is not a general population panel. Customers:\n\n- can get to know the brand better,\n\n- may have a more positive or more negative experience,\n\nmay be more likely to ask the relevant query. The customer panel should be defined separately.\n\n### 4.6. Participants Suggested by the Audited Institution\n\nCompany:\n\n- employee,\n\n- customer,\n\n- partner,\n\n- follower\n\ncan suggest. These users cannot be included in the main independent Population Panel without explanation. Separately:\n\n### ENTITY-NOMINATED PANEL\n\nshould be labelled. If the institution selects participants, the result is not an independent population measurement.\n\n### 4.7. Multi-Source Hybrid Panel\n\nIt is the combination of multiple sources. Example:\n\n- probability-based national panel,\n\n- professional online panel,\n\n- small country local partner,\n\n- Language fairness community\n\ncan be used together. A hybrid panel can increase coverage. However, among the resources:\n\n- repeat individuals,\n\n- selection probabilities,\n\n- incentives,\n\n- task experience,\n\n- user profile\n\nshould also be modelled.\n\n## 5. PARTICIPANT SELECTION INTEGRITY STATUSES\n\nThe participant selection quality of a panel can be classified with the following statuses.\n\n### PSI-0 — UNKNOWN SELECTION\n\nIt is not sufficiently known where and how the participants were selected. A population generalisation cannot be made.\n\n### PSI-1 — CONVENIENCE OR OPEN OPT-IN\n\nIt is a convenience or open volunteer sample. It can be used for exploration and event detection.\n\n### PSI-2 — QUOTA-CONTROLLED OPT-IN\n\nThere are country, language, or demographic quotas. The probability of entry into the population is unknown.\n\n### PSI-3 — CALIBRATED MULTI-SOURCE PANEL\n\nThere are multiple sources, predefined quotas, and population calibration. Unmeasured selection bias may persist.\n\n### PSI-4 — PROBABILITY OR STRONG HYBRID DESIGN\n\nThere is a known probability of selection or a strong probability–non-probability hybrid design.\n\n### PSI-5 — REPLICATED AND EXTERNALLY AUDITED SELECTION\n\nSelection design:\n\n- with different providers,\n\n- in different waves,\n\n- under independent audit\n\nrepeated and bias behaviour has been examined. These statuses do not guarantee the accuracy of the score. It indicates the confidence level resulting from participant selection.\n\n## 6. RANDOM SELECTION WITHIN PANEL AND RANDOM SELECTION FROM THE POPULATION\n\nThere may be 100,000 registered users in a volunteer panel. From this panel, 1,000 people can be randomly selected. Selection:\n\n> is random within the panel.\n\nHowever, the panel itself is not randomly formed from the world population. Therefore, one cannot say: “1,000 people randomly from the world population.” The correct expression is:\n\n> “1,000 participants randomly selected within a pre-registered volunteer panel and calibrated to the target population.”\n\nRandomness should be limited to the stage at which it is applied.\n\n## 7. PARTICIPANT ELIGIBILITY RECORD\n\nEach participant must meet the following conditions to the extent relevant before being included in the main panel:\n\n- Age eligibility\n\n- Volunteer consent\n\n- Country and locale suitability\n\n- Language proficiency\n\n- Access to AI product\n\n- Correct plan or user surface\n\n- Necessary device and technical access\n\n- Unique human status\n\n- Task independence\n\n- Declaration of interest or relationship\n\n- Research security\n\n- Approval for data and evidence collection\n\nConformity decision cannot be retroactively changed after the AI response is generated.\n\n## 8. COUNTRY VERIFICATION\n\nThe question “Which country does the participant belong to?” does not have a single answer. The following fields should be separated:\n\n- Usual country of residence\n\n- Physical country at the time of measurement\n\n- Country or market region of the AI account\n\n- Payment or plan country\n\n- Citizenship\n\n- Target country of the prompt\n\n- Locale\n\nCitizenship may not be the primary variable for most GEO measurements. What is important most of the time:\n\n> It is the actual geographic and product condition in which the user experiences the product.\n\n### 8.1. Travelling User\n\nThe participant may be in a country other than their usual country of residence. This person:\n\n- residing population,\n\n- measurement location,\n\n- travelling user\n\nshould hold separate status as.\n\n### 8.2. VPN or Location Changer\n\nUse of VPN:\n\n- privacy,\n\n- institution network,\n\n- normal user habit\n\nmay be the reason. It should not be considered automatic fraud. However, it makes verifying the country layer difficult. There may be a rule in the main controlled country panel to turn off or log VPN usage. Location changes used to bypass restrictions are not suitable for the main Native Reach panel.\n\n### 8.3. Country Verification Levels\n\n### GV-0 — Unverified\n\nOnly self-declaration is available.\n\n### GV-1 — Panel-Verified\n\nThe panel provider has an existing country registration.\n\n### GV-2 — Session-Consistent\n\nDeclaration, locale, and session signals are consistent.\n\n### GV-3 — Multi-Signal Verified\n\nPrivacy-protecting compatible with multiple signals.\n\n### GV-4 — Independently Audited\n\nThe country verification method has passed external quality control. Collecting an exact street address or sensitive coordinates is not a default requirement. Country verification should be conducted with data minimisation.\n\n## 9. LANGUAGE PROFICIENCY AND NATURAL USAGE\n\nA participant's knowledge of the language may rely solely on the statement: \"Yes, I know.\" This can be a low-risk initial registration. In a high-security language panel, additionally:\n\n- statement of natural usage,\n\n- brief comprehension check,\n\n- panel history,\n\n- local adjudicator evaluation,\n\n- neutral trial before task\n\nmay be used. The language test should not turn into a heavy academic exam that selects only highly educated people. The goal:\n\n> For the participant to understand the prompt and response as a real user.\n\n## 10. VERIFICATION OF AI PRODUCT ACCESS\n\nParticipant:\n\n- should be able to access the correct AI product,\n\n- the correct user surface,\n\n- the correct plan or account condition\n\nshould be able to access. The following situations should be recorded separately:\n\n- Free plan\n\n- Paid individual plan\n\n- Enterprise plan\n\n- Training Plan\n\n- Web\n\n- Mobile\n\n- Desktop\n\n- Embedded Product\n\n- Visible Model\n\n- Web/retrieval feature\n\nA participant cannot use someone else's account. Password sharing cannot be requested. Proof of product access must be verified without publicly disclosing the account identity.\n\n## 11. RELATIONSHIP BETWEEN PARTICIPANT AND AUDITED ENTITY\n\nThe participant's relationship with the monitored organisation may affect response capture behaviour or brand familiarity. Relationship statuses:\n\n### GENERAL_PUBLIC\n\n### CURRENT_CUSTOMER\n\n### FORMER_CUSTOMER\n\n### PROSPECT\n\n### EMPLOYEE\n\n### FORMER_EMPLOYEE\n\n### CONTRACTOR\n\n### PARTNER\n\n### INVESTOR\n\n### COMPETITOR_AFFILIATE\n\n### AI_PROVIDER_AFFILIATE\n\n### RESEARCHER\n\n### UNKNOWN\n\nDirect employee and decision-maker relationships in the main general population panel:\n\n- can be excluded,\n\n- can be put in a separate layer,\n\ncan be protected with a low true population weight. Automatically excluding all related individuals may also be wrong. If they are part of the actual customer or employee population, they can be measured separately and explicitly.\n\n## 12. BRAND FAMILIARITY\n\nAudited participant entity:\n\n- has never heard of it,\n\n- knows the name,\n\n- knows the product,\n\n- is a customer,\n\n- may have a strong positive or negative opinion.\n\nThis information:\n\n- can affect participant behaviour,\n\n- natural user panel,\n\n- prompt response\n\ndesired. If familiarity is measured before the prompt, it may prepare the user for the brand. Therefore, the order of preference:\n\n- Previously existing panel information\n\n- Separate question after the task\n\n- If necessary, a neutral pre-measurement that does not guide the brand\n\ncan be conducted. Familiarity should be visible in the main sample; it should not be selected as a weighting variable after the results are seen.\n\n## 13. AI USAGE FAMILIARITY\n\nParticipants:\n\n- first-time user,\n\n- infrequent user,\n\n- regular user,\n\n- intense or professional user\n\nindividuals can be. AI usage frequency:\n\n- product access,\n\n- type of plan,\n\n- interface knowledge,\n\n- personalisation,\n\n- likelihood of completing the task correctly\n\ncan be affected. However, selecting only experienced AI users can reduce protocol errors. But it may distort the natural population estimate. The correct solution:\n\n- measure the level of use,\n\n- compare it with the target population,\n\n- stratify or weight if necessary,\n\nand preserve the valid real experience of the inexperienced user.\n\n## 14. UNIQUE HUMAN RULE\n\nThe population unit of GEO-1000 is a human. The same person:\n\n- in multiple panels,\n\n- with multiple email accounts,\n\n- with multiple AI accounts,\n\n- from different devices\n\ncan participate. Uniqueness checks must be performed in a privacy-preserving manner. Signals that can be used:\n\n- Anonymous participant identity\n\n- Cryptographic or hash-based rechecking\n\n- Device and session compliance\n\n- Payment or incentive record\n\n- Time and task pattern\n\n- Same screenshot or response file\n\n- Secure rechecking across multiple panel providers\n\nNo single signal needs to be sufficient for automatic exclusion. In particular:\n\n- Families sharing a common device,\n\n- institution network,\n\n- Accessibility tools,\n\n- low-income shared device users\n\nmay be mistakenly considered fake. Suspicious records should undergo human review and an appeal process.\n\n## 15. PARTICIPANT REWARD\n\nThe payment or reward given to the participant may influence their data behaviour.\n\n### 15.1. Payment Independent of Outcome\n\nPreferred method:\n\n> A fixed and pre-disclosed payment for completing the task according to protocol.\n\nPayment:\n\n- cannot depend on the answer being positive,\n\n- can be forced into the categories of\n\ntrue,\n\n- long,\n\n- favourable to the brand,\n\n- or accompanied by citations.\n\nSelection must remain independent of all such outcomes.\n\n### 15.2. Quality Bonus\n\nIf a quality bonus is to be used, only:\n\n- complete evidence,\n\n- correct prompt,\n\n- correct product,\n\n- timely delivery,\n\n- personal data cleaning\n\ncan be tied to protocol conditions. It cannot be tied to the content of the AI response.\n\n### 15.3. Speed Incentive\n\nBonus solely for speed:\n\n- rush,\n\n- incomplete capture,\n\n- wrong product,\n\n- cutting the answer\n\nmay increase the risk. A reasonable time should be allowed within the field window.\n\n### 15.4. Payment by Country\n\nPayment may vary according to local fees and living conditions. This situation:\n\n- may affect participation likelihood,\n\n- panel composition\n\nand should be recorded along with the payment approach and country differences.\n\n### 15.5. Coercion and Excessive Incentive\n\nThe reward should not create pressure that the person cannot reasonably refuse. In particular:\n\n- low-income,\n\n- student,\n\n- dependent employees,\n\n- vulnerable groups\n\nshould have their voluntariness protected.\n\n## 16. INVITATION TEXT AND PRE-ORIENTATION\n\nWhen inviting a participant, the purpose of the research may prepare them for a specific outcome. A wrong invitation: “We are testing how incorrectly AI systems interpret Apple.” This statement may lead the participant to look for errors. Another wrong invitation: “Help us prove the AI success of our brand.” This statement creates an expectation of a positive result. Preferred neutral invitation: “Study of standardised user experience and representation observation in specific AI products.” The audited entity may become visible later due to the nature of the task. However, the desired outcome should not be disclosed.\n\n## 17. TASK ASSIGNMENT\n\nParticipants:\n\n- AI product,\n\n- prompt,\n\n- language,\n\n- wave,\n\n- session condition,\n\n- task order\n\nshould be assigned in advance using a predefined method. The participant’s:\n\n- favorite AI product,\n\n- the plan he/she knows best,\n\n- the language with a high probability of receiving a positive response\n\nif allowed to choose by themselves, assignment bias may occur.\n\n### 17.1. Random Assignment Within Layer\n\nAfter country, language, and product access are verified, tasks can be assigned randomly within the layer.\n\n### 17.2. Block Assignment\n\nBlocks can be used to balance free/paid plans, mobile/web, or AI usage frequency.\n\n### 17.3. Product Order\n\nIf a person is testing more than one product, the order should be randomised. The first product's response may affect the evaluation of the next response or task behaviour.\n\n### 17.4. Order of Prompts\n\nIf a participant receives more than one Prompt:\n\n- learning,\n\n- acquiring new information about the brand,\n\n- carrying the previous answer to the next question\n\nrisk arises. The order of prompts should be recorded and balanced if necessary.\n\n## 18. PANEL CONDITIONING\n\nUsers who frequently participate in research panels may learn the expectations of the task. A person participating repeatedly in the same test:\n\n- may remember the correct answer,\n\n- may choose the screen capture more carefully,\n\n- may review the site of the audited entity in advance,\n\nwhich errors are being searched for can be learned. This situation is called:\n\n#### Panel conditioning\n\nPanel conditioning can change the real natural user experience.\n\n### 18.1. Sources of Conditioning\n\nPrevious GEO waves Re-measurement of the same entity Same prompt bank Peer review training Task sharing in participant communities Previous payments and quality feedback\n\n### 18.2. Conditioning Controls\n\nNew and repeat participant status First participation date Previous exposure to the same entity Previous exposure to the same prompt family Rotating panel Task order randomness Specified rest period New user subsample Comparison of conditioned and new user results Reusing the same participant is not prohibited. It is valuable for longitudinal measurement. Cannot be presented as a natural cross-sectional panel.\n\n## 19. TASK CONTAMINATION AND PARTICIPANT COMMUNICATION\n\nParticipants:\n\n- prompts,\n\n- responses,\n\n- screenshots,\n\n- expected results\n\ncan be shared with each other. This situation may disrupt independent observations. Controls:\n\n- Limit wave duration\n\n- Randomising tasks\n\n- Obligation not to share results with participants\n\n- Looking for repeated answers and visuals\n\n- Examining unusual text and time similarities\n\n- Monitoring community or panel provider warnings\n\nParticipant communication is not automatically fraud. Its material effect should be examined.\n\n## 20. TYPES OF NON-RESPONSE\n\nNon-response is not a single event.\n\n### 20.1. Not Responding to Invitations\n\nThe participant does not respond at all to the research invitation.\n\n### 20.2. Declining Participation\n\nThe person sees the invitation and chooses not to participate.\n\n### 20.3. Not Starting the Task After Qualification\n\nThe person is found suitable but does not start the task.\n\n### 20.4. Task Interruption\n\nThe participant abandons the task halfway.\n\n### 20.5. Not Uploading Evidence\n\nAn AI response may have been obtained, but the required record is not uploaded.\n\n### 20.6. Invalid Completion\n\nThe task is completed, but:\n\n- wrong prompt,\n\n- wrong product,\n\n- selected answer,\n\n- missing screenshot\n\nare reasons for invalidity. The ratios of these stages should be reported separately.\n\n## 21. RESPONSE FUNNEL\n\nFor layer h:\n\n- nhsel: selected person\n\n- nhinv: invited person\n\n- nhcon: person who agreed to participate\n\n- nhsta: person who started the task\n\n- nhcom: person who completed\n\n- nhval: valid observation\n\nLet's be. Basic rates:\n\nRR_h^invite = n_h^con/n_h^inv; RR_h^complete = n_h^com/n_h^sta; RR_h^valid = n_h^val/n_h^com\n\nIt can be shown as. Each of these rates indicates a different problem. Giving only the final valid count hides field bias.\n\n## 22. NONRESPONSE BIAS\n\nPeople who do not respond may be different from those who do. Example: A user with low internet cannot complete a task. A free plan user cannot access the product. An inexperienced user finds the instructions difficult. A heavy AI user completes faster. In certain countries, payment remains low. A user with a negative brand experience may be more willing to participate. Even if the non-response rate is high, bias may be low. Even if the non-response rate is low, a small missing group may be financially significant. Therefore, not just the rate alone:\n\n#### Which characteristics differ between respondents and non-respondents\n\nshould be examined.\n\n## 23. NON-RESPONSE WEIGHTS\n\nIf a non-response setting will be used, classes should be determined before results are seen. Potential variables:\n\n- Country\n\n- Language\n\n- Age\n\n- Frequency of AI usage\n\n- Plan\n\n- Interface\n\n- Panel source\n\n- Accessibility\n\n- Brand familiarity\n\nNonresponse coefficient:\n\na_r^{NR} = (Σ_{j∈S_r} d_j) / (Σ_{i∈R_r} d_i)\n\ncan be calculated as. However, this adjustment only relies on the assumption within the same class: that respondents sufficiently represent non-respondents. It does not resolve unmeasured differences.\n\n## 24. MISSING DATA MECHANISMS\n\nMissingness can conceptually be considered in three classes.\n\n### 24.1. Random Missingness\n\nMissingness is not materially related to the outcome and user characteristics. In reality, it may be rare.\n\n### 24.2. Missingness Explained by Observed Features\n\nDeficiency:\n\n- country,\n\n- does not imply any suitability in terms of age,\n\n- plan,\n\n- language\n\nis related to well-known areas such as. Weighting can partially help.\n\n### 24.3. Deficiency Related to Unobserved Features\n\nFor example:\n\n- Distrust in the AI product,\n\n- low digital skills,\n\n- brand opposition,\n\n- privacy concern\n\nif not measured, weighting cannot completely correct the deficiency. In this case, sensitivity analysis is needed.\n\n## 25. PANEL LOSS AND INTER-WAVE SEPARATION\n\nIn longitudinal or rotating panels where the same users are measured again, some individuals do not participate in the subsequent wave. This is panel loss. Those remaining in the panel:\n\n- more disciplined,\n\n- more reward-motivated,\n\n- more experienced,\n\n- more accustomed to research\n\nmay be. The change in score over time can partially stem from the panel composition.\n\n### 25.1. Panel Attrition Record\n\nIn each wave:\n\n- re-invited,\n\n- participated,\n\n- dropped out,\n\n- newly added\n\nuser numbers should be shown.\n\n### 25.2. Panel Attrition Weight\n\nThe remaining participants can be weighted according to the probability of dropout. This method does not completely resolve the bias from unobserved panel attrition.\n\n### 25.3. Refresh Panel\n\nNew users can be added to replace the lost users. Results for new and returning users should be reported separately.\n\n## 26. PARTICIPANT FRAUD AND AUTOMATION\n\nThere is a risk of fraud when GEO-1000 is an incentivised and global panel. Possible behaviours:\n\n- A person participating with multiple accounts\n\n- Use of automatic bots\n\n- Uploading someone else’s screenshot\n\n- Editing images\n\n- Uploading the same answer on behalf of different users\n\n- Declaring the wrong country or language\n\n- Sharing an account\n\n- Having someone else complete the task\n\n- Rejoining among panel providers\n\n- Creating bulk task farms\n\nThis risk does not allow all low-income or fast participants to be considered suspicious. Fraud detection:\n\n- debates,\n\n- multiple signals,\n\n- proportional human review\n\nshould be based on.\n\n## 27. FRAUD AND INTEGRITY STATUSES\n\n### FI-0 — NOT ASSESSED\n\nThe record has not been evaluated in terms of integrity.\n\n### FI-1 — NO MATERIAL ANOMALY\n\nNo clear signs of fraud have been found. This is not an absolute guarantee of accuracy.\n\n### FI-2 — REVIEW REQUIRED\n\nThere is one or more suspicious signal(s).\n\n### FI-3 — DUPLICATE OR COLLUSION SUSPECTED\n\nThere is suspicion of repeated person, visual, response, or coordination.\n\n### FI-4 — PROTOCOL MANIPULATION CONFIRMED\n\nThe participant has deliberately altered the protocol.\n\n### FI-5 — FABRICATED OBSERVATION CONFIRMED\n\nThe observation is not a genuine AI user experience. FI-2 is not automatic exclusion. Review required.\n\n## 28. FRAUD SIGNALS\n\nSame screenshot hash Same rare file footprint Impossible completion time Multiple identity repetitions from the same device or session Material conflict in country, locale, and product access Mismatch between response text and image Trace of image editing Same person in primary and backup list Same anonymous person across multiple panel providers Serial and mechanical completion of tasks Same unusual error pattern A single signal may not be conclusive proof of fraud.\n\n## 29. FAIRNESS IN FRAUD DETECTION\n\nThe following users may be mistakenly marked as fraudulent:\n\n- Families sharing a common device\n\n- Employees using an institutional network\n\n- University laboratory participants\n\n- Privacy-conscious individuals using VPNs\n\n- Disabled users using screen readers or automation helpers\n\n- Those uploading similar compressed images due to low bandwidth\n\nTherefore:\n\n- automatic tagging,\n\n- human review,\n\n- participant description if necessary,\n\n- appeal\n\nmust work together.\n\n## 30. PARTICIPANT VALIDITY SHOULD BE SEPARATED FROM AI RESPONSE ACCURACY\n\nA participant's record may be protocol-compliant. The AI response may be incorrect. This is a valid observation. Another participant's AI response may be correct. However, the user:\n\n- has refreshed the answer three times,\n\n- only uploaded the best answer,\n\n- may have changed the prompt\n\nThis is an invalid observation. Therefore:\n\n> The correct AI answer does not mean valid data. / An incorrect AI answer does not mean invalid data.\n\nValidity assessment should be done before substantive scoring and as blind as possible.\n\n## 31. CONFLICT OF INTEREST AND PARTICIPANT SOURCE\n\nA participant or panel provider may benefit from the following outcomes:\n\n- The audited company receiving a high score\n\n- The AI product appearing superior\n\n- A competitor receiving a low score\n\n- Selling more audit engagements\n\n- Award or employment relationship\n\nTherefore:\n\n- participant relations,\n\n- panel provider agreement,\n\n- result-independent wage,\n\n- Customer intervention\n\nshould be explained. Audited institution:\n\n- which users to invite,\n\n- who will be considered invalid,\n\n- Who will be replaced with a backup\n\ncannot decide.\n\n## 32. PANEL PROVIDER EFFECT\n\nUsers of two panel providers in the same country may be different. A provider:\n\n- Technology enthusiasts,\n\n- students,\n\n- intensive research participants\n\nmay have a higher rate in terms of care. The panel provider should be kept as a cluster or source variable in the analysis. If possible:\n\n- provider-based results,\n\n- source effect,\n\n- provider comparison within the same country\n\nshould be examined. A single-provider global panel cannot be presented as the world population itself.\n\n## 33. REPEAT INDIVIDUAL IN MULTI-SOURCE PANEL\n\nThe same individual:\n\n- two professional panels,\n\n- in an open call,\n\n- in the university list\n\ncan be found. A multi-source panel can increase the number of records without increasing uniqueness. Cross-source duplication check:\n\n- privacy-preserving,\n\n- not requiring central identity disclosure,\n\n- allowing objection to mismatches\n\nshould be established in a manner.\n\n## 34. SAMPLE BIAS MAP\n\nGEO-1000 should separately monitor the following types of bias.\n\n### SB-1 — Coverage Bias\n\nSome segments of the target population are not in the sampling frame.\n\n### SB-2 — Selection Bias\n\nThe probability of individuals in the frame being selected for the panel does not conform to the intended design.\n\n### SB-3 — Self-Selection Bias\n\nVolunteers for the study are systematically different from the population.\n\n### SB-4 — Invitation and Guidance Bias\n\nThe invitation text or source attracts people with a specific viewpoint.\n\n### SB-5 — Nonresponse Bias\n\nPeople who do not participate or do not complete are different in terms of outcomes.\n\n### SB-6 — Panel Attrition Bias\n\nUsers who remain between waves are different from those who drop out.\n\n### SB-7 — Incentive Bias\n\nPayment or reward affects the participant’s speed, selection behaviour, or response.\n\n### SB-8 — Panel Conditioning\n\nThe participant learns the expected behaviour from previous studies.\n\n### SB-9 — Task Assignment Bias\n\nParticipants choose the product, language, or prompts themselves or are assigned unevenly.\n\n### SB-10 — Country and Language Classification Bias\n\nThe participant is assigned to the wrong country or language cell.\n\n### SB-11 — Product, Plan, and Interface Bias\n\nOnly certain plan or device users determine the panel.\n\n### SB-12 — Brand Familiarity Bias\n\nCustomers, fans, critics, or employees are overrepresented.\n\n### SB-13 — Fraud and Repeat Bias\n\nThe same person or automatic participation gains too much influence on the result.\n\n### SB-14 — Validity Exclusion Bias\n\nNegative AI responses are more often considered invalid.\n\n### SB-15 — Weighting Bias\n\nPost-result selected variables or excessive weight model score change in the desired direction.\n\n### SB-16 — Surviving Sample Bias\n\nOnly users who can successfully complete the task are presented like the real product population.\n\n## 35. PURPOSE OF WEIGHTING\n\nWeighting can serve the following purposes:\n\n- Correcting unequal selection probabilities\n\n- Returning oversampling to the actual population proportion\n\n- Reducing differences in response rates\n\n- Calibrating the sample to known target totals\n\n- Combining multiple panel sources\n\nWeighting cannot serve the following purposes:\n\n- Creating a person who does not exist in the sampling frame\n\n- Correcting unmeasured variables precisely\n\n- Making a false or erroneous observation real\n\n- Producing the desired score after the result\n\n- Converting a single small cell into a reliable country score\n\n- Automatically turning a non-probability panel into a probability sample\n\n## 36. HEAVYWEIGHT ARCHITECTURE\n\nThe final weight can be shown as follows:\n\nw_i = d_i × a_i^{NR} × g_i^{CAL} × p_i^{SRC}\n\nHere:\n\n- di: basic design or starting weight\n\n- aiNR: response setting\n\n- giCAL: population calibration\n\n- piSRC: panel source or improbable sample adjustment\n\nIn the probabilistic panel:\n\nd_i = 1/π_i\n\nIt is possible. The true πi is unknown in the non-probability panel. In this case, the weight used is not the design weight, but the model-based or calibration weight. This distinction should be preserved in the report.\n\n## 37. WEIGHTS CANNOT DEPEND ON THE RESULT\n\nThe following variables cannot be used to create weights:\n\n- Whether the AI response is correct or not\n\n- Whether the brand is mentioned or not\n\n- Whether there is a citation\n\n- Whether a recommendation is made\n\n- Whether a critical error occurs\n\n- Whether the audited organisation likes the result\n\nWeight variables:\n\n- Before the result is seen,\n\n- Because it is related to the target population and the selection mechanism\n\nIt should be selected. If the response result determines the weight, the score shapes itself.\n\n## 38. CALIBRATION VARIABLES\n\nPossible calibration areas:\n\n- Country\n\n- Region\n\n- Language\n\n- Age\n\n- Gender, if necessary and ethical\n\n- Education\n\n- Urban/rural settlement\n\n- Type of internet access\n\n- Device\n\n- Frequency of AI usage\n\n- Free/paid plan\n\n- Interface\n\n- Accessibility status\n\nReliable target totals may not be available for each variable. Too many calibration variables:\n\n- can create excessive weight variability,\n\n- low effective sample size,\n\n- model overfitting\n\nIt should be frozen before seeing calibration set results.\n\n## 39. WEIGHTING BALANCE\n\nThe sample before and after weighting should be compared with the target population. For each field:\n\n- target share\n\n- raw sample share\n\n- weighted share\n\n- remaining difference\n\nshould be shown. Weighting is applied only to the number of results and the balance table cannot be saved.\n\n## 40. WEIGHT DIAGNOSES\n\nThe main report should include the following fields as far as they are relevant:\n\n- Minimum weight\n\n- Maximum weight\n\n- Median weight\n\n- Weight variation coefficient\n\n- Weight ratio\n\n- Number of trimmed weights\n\n- Raw sample size\n\n- Effective sample size\n\n- Design effect\n\n- Unweighted result\n\n- Main weighted result\n\n- Alternative reasonable weight results\n\nWeighted single number is not a full control record without weight identifiers.\n\n## 41. SYNTHETIC WEIGHTING DISPLAY\n\nSYNTHETIC METHODOLOGY DISPLAY / The participants, rates, and scores below are hypothetical. Target population and panel distribution:\n\nSynthetic pass rates:\n\nRaw result:\n\n900/1000=90%\n\n### 41.1. Calibration Coefficients\n\nLow usage:\n\n0.50/0.10=5\n\nMedium usage:\n\n0.30/0.30=1\n\nHigh usage:\n\n0.20/0.60 = 1/3\n\nWeighted response:\n\n72(5) + 252(1) + 576(1/3) = 804\n\nWeight total:\n\n100(5) + 300(1) + 600(1/3) = 1,000\n\nWeighted result:\n\n804/1000=80.4%\n\n### 41.2. Effective Sample Size\n\nn_eff = [100(5)+300(1)+600(1/3)]² / [100(5²)+300(1²)+600(1/3)²] ≈ 349\n\nResult card:\n\n- Raw valid observation: 1,000\n\n- Raw score: 90 per cent\n\n- Weighted score: 80.4 per cent\n\n- Effective sample: approximately 349\n\n- Main reason: overrepresentation of heavy AI users\n\nCorrect interpretation:\n\n> The panel contains 1,000 valid observations; however, as a result of target population calibration, the estimate has dropped to 80.4%, and due to weight variability, the effective sample size has become approximately 349.\n\nIncorrect comment: “Global accuracy has been proven to be 90% with 1,000 people.”\n\n## 42. WHAT CAN'T WEIGHTING CORRECT?\n\nThe following groups should not be present in the synthetic panel at all:\n\n- People who use screen readers\n\n- Low bandwidth rural users\n\n- People using the AI product for the first time\n\n- Small language community\n\nWeighting only changes the weight of the users in the panel. The unobserved group's:\n\n- AI your answer,\n\n- reveal access issues,\n\n- error distribution\n\ncannot produce. Therefore:\n\n> Zero coverage means zero information.\n\nWeighted result: it may be a prediction closer to the target population, not a full observation of the target population.\n\n## 43. SENSITIVITY ANALYSIS\n\nThe dependence of the weighted result on a single weight model should be examined. Candidate results:\n\n- Raw unweighted result\n\n- Design-focused result\n\n- Demographically calibrated result\n\n- Calibration with AI usage added\n\n- Weight trimmed result\n\n- Panel provider effect adjusted result\n\nIf the results change materially:\n\n#### Weight Sensitivity Warning\n\nshould be given. A single “correct” weight result cannot be presented as an absolute truth.\n\n## 44. PROPENSITY ADJUSTMENT IN NONPROBABILITY PANEL\n\nThe tendency of nonprobability panel members to enter the panel can be modelled according to the target population. The probability of a person being in the panel:\n\np̂_i = Pr(i ∈ panel | X_i)\n\ncan be estimated. The candidate inverse propensity weight:\n\nw_i^P = 1/p̂_i\n\ncan be used. This method:\n\n- uses only the measured Xi variables,\n\n- is affected by a misspecified model,\n\n- unobserved selection does not resolve,\n\nrequires strong target reference data. Therefore, the result:\n\n#### Should be labelled as:\n\nModel-calibrated non-probability panel estimate\n\n## 45. SMALL AREA AND MODEL-BASED ESTIMATES\n\nDirect sampling may be insufficient in some country or language cells. Model-based methods can borrow information from neighbouring or similar groups. This means that the result:\n\n- directly observed country score,\n\n- raw user rate\n\nis not. The correct status:\n\n### MODEL-ASSISTED ESTIMATE\n\n### SMALL-AREA ESTIMATE\n\n### NOT DIRECTLY OBSERVED\n\nmust. Model-based prediction does not turn a country without Country Observer observation into an observed country.\n\n## 46. PARTICIPANT PRIVACY AND VERIFIABILITY\n\nGEO-1000 requires proof for each observation. This does not mean that the participant's identity will be disclosed publicly. A screenshot may include the following information:\n\n- Name\n\n- Email\n\n- Profile picture\n\n- Previous chat\n\n- Account plan\n\n- Institution name\n\n- Sensitive personal information\n\nNOMOS Capture or equivalent system:\n\n- should not capture unnecessary personal fields,\n\n- should perform automatic or controlled redaction,\n\n- raw evidence must be kept in a secure repository,\n\nonly anonymous and verifiable records should be presented to the public.\n\n### 46.1. Public Evidence Manifest\n\nPublic record:\n\n- observer ID,\n\n- country/language cell,\n\n- AI product,\n\n- time,\n\n- evidence hash,\n\n- verification status\n\nmay be carried. Full screenshot:\n\n- edited,\n\n- restricted access,\n\n- open to independent auditor\n\nmay be. The existence of a thousand pieces of evidence does not require the public disclosure of the identities of a thousand people.\n\n### 46.2. Sensitive Small Country Risk\n\nIf there is only one participant in the small country:\n\n- device,\n\n- time,\n\n- language,\n\n- plan\n\nWhen gathered together, it may become easier to re-identify the person. The level of detail in public reports should be reduced. The integrity of the evidence can be maintained in an access-controlled system.\n\n## 47. PARTICIPANT CONSENT\n\nThe participant should at least know the following:\n\n- Which task they will perform\n\n- Which data and screenshots will be collected\n\n- Which personal information will not be collected\n\n- How long the data will be stored\n\n- Who will have access\n\n- How the result will be published\n\n- How the reward will be given\n\n- That participation is voluntary\n\n- Right to withdraw or object\n\n- Account password will not be requested\n\n- It is not expected to violate AI product terms\n\nConsent should not be hidden in a long and incomprehensible text. [K06; K22]\n\n## 48. PARTICIPANT SAFETY\n\nFrom the participant:\n\n- to provide personal health, legal, or financial information to AI,\n\n- to use real customer data,\n\n- to share employer confidential information,\n\n- to weaken account security,\n\n- to exceed usage limits,\n\n- to violate the law or product terms\n\ncannot be requested. Test prompts should carry as much as possible a public and non-personal context.\n\n## 49. THE PARTICIPANT CARRIES HUMAN DIVERSITY, NOT QUALITY\n\nInexperienced user:\n\n- can complete the task more slowly,\n\n- more difficult,\n\n- with more support\n\nThis person is not automatically a low-quality participant. They are a real part of the target population. If the research design is so complex that only people with high digital skills can complete it: the problem may not be with the participant, but with the measurement tool. The protocol:\n\n- should be understandable,\n\n- accessible,\n\n- linguistically natural,\n\n- technical load limited\n\nmust be.\n\n## 50. PUBLIC RESULT CARD OF PARTICIPANT SELECTION\n\nThe candidate public report should include the following areas:\n\n#### Panel Resources\n\nProbability panel share Professional volunteer panel share Open call share University/institution share Client or affiliated panel share\n\n#### Participation Funnel\n\nPerson in frame Selected Invited Accepted Started Completed Producing valid observation\n\n#### Eligibility and Integrity\n\nCountry verification levels Language verification levels Repeat person rate Suspicious registration rate Verified fraud rate Invalid observation rate\n\n#### Panel Composition\n\nAI usage frequency Plan and interface Brand familiarity Customer/employee relationship New and returning panel users Accessibility coverage\n\n#### Weight Diagnoses\n\nThe following may be published without revealing participant identities: raw and weighted scores; minimum and maximum weights; the coefficient of variation of weights; effective sample size; sensitivity to weighting; and coverage warnings.\n\n## 51. MANDATORY NORMATIVE PROVISIONS\n\n**CH08-N01**\n\nEach valid GEO-1000 observation must carry a traceable participant selection chain from the target population to the analytic record.\n\n**CH08-N02**\n\nRandom selection within the panel cannot be presented as random selection from the target world population.\n\n**CH08-N03**\n\nThe source of participants, selection method, and panel status should be explained in the public method record.\n\n**CH08-N04**\n\nIt cannot be defined as an improbable or voluntary panel probability sample of the population.\n\n**CH08-N05**\n\nParticipants proposed by the audited institution cannot be included in the main independent population panel without explanation.\n\n**CH08-N06**\n\nCustomer, employee, partner, competitor, and AI provider relationships should be recorded separately.\n\n**CH08-N07**\n\nParticipant eligibility criteria must be verified before the AI response is seen.\n\n**CH08-N08**\n\nCountry, language, product, plan, and interface eligibility must be determined before the related task assignment.\n\n**CH08-N09**\n\nThe same person cannot be duplicated as a population unit due to multiple accounts, devices, languages, panels, or AI products.\n\n**CH08-N10**\n\nUnique person verification must be conducted according to data minimisation and privacy principles.\n\n**CH08-N11**\n\nA single technical signal alone cannot be considered sufficient for an automatic definite decision that a participant is fake.\n\n**CH08-N12**\n\nA participant's reward cannot be tied to the AI response being positive, correct, long, containing references, or beneficial to the brand.\n\n**CH08-N13**\n\nThe quality bonus can only be tied to pre-specified protocol integrity.\n\n**CH08-N14**\n\nThe invitation and information text cannot direct the participant to seek a positive or negative outcome.\n\n**CH08-N15**\n\nAI product should be assigned or randomised before seeing the prompt and task sequence result.\n\n**CH08-N16**\n\nProduct and prompt selection based on participant preference cannot be used unexplainedly in the main population panel.\n\n**CH08-N17**\n\nPanel conditioning, previous test exposure, and repeat participation must be recorded.\n\n**CH08-N18**\n\nThe results of new and conditioned participants should be analysed separately when necessary.\n\n**CH08-N19**\n\nA participant's inclusion in the panel, retention, exclusion, or replacement with a backup cannot be based on the content of the AI response.\n\n**CH08-N20**\n\nIncorrect, negative, rejection, or Critical AI output should be preserved under a valid protocol.\n\n**CH08-N21**\n\nParticipant validity should be evaluated blind as much as possible and before assessing the substantive accuracy of the response.\n\n**CH08-N22**\n\nInvitation, acceptance, initiation, completion, and valid observation rates should be reported separately.\n\n**CH08-N23**\n\nThe response setting should only be based on predefined and relevant variables.\n\n**CH08-N24**\n\nWeighting cannot make users who are not present in the sampling frame represented.\n\n**CH08-N25**\n\nWeight variables cannot depend on the AI response result, the score, or the commercial interest of the regulated institution.\n\n**CH08-N26**\n\nNon-probability panel weights cannot be presented as design weights.\n\n**CH08-N27**\n\nCalibration variables and target totals must be released before the results are seen.\n\n**CH08-N28**\n\nUnweighted and main weighted results must be stored together.\n\n**CH08-N29**\n\nWeighting diagnoses and effective sample size should be visible in the main report.\n\n**CH08-N30**\n\nIf weight trimming is used, the untrimmed result and sensitivity analysis must be retained.\n\n**CH08-N31**\n\nPanel provider and participant source effects should be kept as separate variables.\n\n**CH08-N32**\n\nIn a multi-source panel, inter-source repeated person checks should be conducted.\n\n**CH08-N33**\n\nFraud detection should include multiple signals, proportional human review, and an appeal process.\n\n**CH08-N34**\n\nUse of VPN, shared device, institutional network, or assistive technology should not automatically be considered fraud.\n\n**CH08-N35**\n\nModel-based small area estimation cannot be presented directly like an observed country or language score.\n\n**CH08-N36**\n\nParticipant screenshots and evidence must carry personal data minimisation, redaction, and access control.\n\n**CH08-N37**\n\nAccount passwords, private chat history, or unnecessary sensitive information cannot be requested from participants.\n\n**CH08-N38**\n\nParticipants cannot be asked to bypass product, age, identity, country, or account restrictions.\n\n**CH08-N39**\n\nChildren or users requiring special protection cannot be enrolled in the panel without separate ethical and security protocols.\n\n**CH08-N40**\n\nParticipant selection and weighting records must have an accountable human or institution owner.\n\n## 52. FORMS OF FAILURE\n\n**CH08-F01 — COUNTING RANDOMS WITHIN THE PANEL AS RANDOMS OF THE WORLD**\n\nRandom selection within the voluntary panel is presented as if it were random selection from the world population.\n\n**CH08-F02 — COUNTING QUOTA COMPLIANCE AS REPRESENTATIVE**\n\nIt is claimed that all behavioural and technological biases are resolved because age and country quotas are met.\n\n**CH08-F03 — COUNTING CUSTOMERS AS GENERAL POPULATION**\n\nThe monitored institution's user list is used as if it were a world user panel.\n\n**CH08-F04 — HIDING EMPLOYEES AND PARTNERS**\n\nUsers associated with the institution are shown as independent general participants.\n\n**CH08-F05 — COUNTING A SINGLE PANEL PROVIDER AS GLOBAL**\n\nProvider effect and registered panel behaviour are not examined.\n\n**CH08-F06 — COUNTING AN OPEN CALL AS PROBABLE SAMPLE**\n\nSelf-selection is preserved.\n\n**CH08-F07 — DETERMINING THE COUNTRY ONLY BY IP**\n\nTravel, VPN, institutional network, and account region differences are ignored.\n\n**CH08-F08 — COUNTING CITIZENSHIP AS THE COUNTRY OF USE**\n\nThe actual geographic condition of the product experience is misclassified.\n\n**CH08-F09 — COUNTING LANGUAGE DECLARATION AS DEFINITE WITHOUT TESTING**\n\nThe participant is kept in the language cell even though they do not understand the prompt or the response.\n\n**CH08-F10 — TURNING THE LANGUAGE TEST INTO ELITE SELECTION**\n\nOnly highly educated people remain on the panel.\n\n**CH08-F11 — COUNTING THE NUMBER OF ACCOUNTS AS PEOPLE**\n\nOne person increases the population weight with multiple accounts.\n\n**CH08-F12 — COUNTING A SHARED ACCOUNT AS ONE PERSON**\n\nAn account used by multiple people is interpreted as a single user.\n\n**CH08-F13 — PAYMENT DEPENDING ON THE RESULT**\n\nA bonus is given for a positive, correct, or referenced answer.\n\n**CH08-F14 — BREAKING THE PROTOCOL WITH SPEED BONUS**\n\nThe participant takes a screenshot or rushes through the task before completing the response.\n\n**CH08-F15 — EXPLAINING THE RESULT IN THE INVITATION**\n\nThe participant is asked to find an error or success.\n\n**CH08-F16 — LETTING THE PARTICIPANT CHOOSE THE PRODUCT**\n\nUsers who select only the AI product they can access or like determine the panel.\n\n**CH08-F17 — HIDING THE EFFECT OF THE ORDER OF PROMPTS**\n\nPrevious answers affect subsequent tasks.\n\n**CH08-F18 — CONSIDERING PANEL CONDITIONING AS NATURAL USE**\n\nA user who sees the same test multiple times is presented as a new user.\n\n**CH08-F19 — IGNORING RESULT SHARING**\n\nParticipants transferred tasks and responses to each other.\n\n**CH08-F20 — COUNTING NON-RESPONSES AS MISSING DATA**\n\nThe systematic difference of non-participants is not examined.\n\n**CH08-F21 — COUNTING PANEL LOSS AS IMPROVEMENT IN RESULTS**\n\nScores increase because weak or inexperienced users drop out in subsequent waves.\n\n**CH08-F22 — INVALIDATING NEGATIVE RESPONSES**\n\nThe participant is replaced because AI provided an incorrect response.\n\n**CH08-F23 — VALIDATING THE CORRECT RESPONSE**\n\nEven if the participant refreshed the response, the record is kept because the result is correct.\n\n**CH08-F24 — OVERLOOKING BOTS AND REPEATED PARTICIPATION**\n\nThe same person or automation produces a large number of observations.\n\n**CH08-F25 — MASS EXCLUSION WITH A SINGLE SIGNAL**\n\nVPN, device, or speed causes real users to be considered fake.\n\n**CH08-F26 — CONDUCTING FRAUD REVIEW BASED ON RESULTS**\n\nLow-scoring records are examined more frequently.\n\n**CH08-F27 — CREATING A HEAVILY UNSEEN GROUP**\n\nIt is claimed that users not present in the sample are represented.\n\n**CH08-F28 — SELECTING POST-RESULT CALIBRATION**\n\nWeight variables that produce the most appropriate score are selected afterward.\n\n**CH08-F29 — COUNTING RAW PANEL SHARE AS POPULATION SHARE**\n\nHeavy AI users determine the global result with raw numbers.\n\n**CH08-F30 — COUNTING UNUSUAL WEIGHT AS DESIGN WEIGHT**\n\nExact population inference is made even though the true selection probability is unknown.\n\n**CH08-F31 — HIDING THE UNWEIGHTED RESULT**\n\nThe effect of the weight model is not visible.\n\n**CH08-F32 — HIDING THE EFFECTIVE SAMPLE**\n\nEven though 1,000 raw records carry 300–400 information units, full precision is claimed.\n\n**CH08-F33 — REMOVING WEAK GROUPS BY WEIGHT TRIMMING**\n\nHigh-weighted small groups are trimmed because they lower the result.\n\n**CH08-F34 — COUNTING MODEL-BASED ESTIMATED SCORES**\n\nA country without an example is given a definite score.\n\n**CH08-F35 — REVEALING IDENTITY FOR EVIDENCE**\n\nScreenshots are published publicly with personal information.\n\n**CH08-F36 — MAKING A SMALL COUNTRY PARTICIPANT REIDENTIFIABLE**\n\nCountry, time, device, and plan information reveal a single person.\n\n**CH08-F37 — ASKING PARTICIPANT FOR PASSWORD OR PRIVATE CHAT**\n\nCollecting evidence violates account security.\n\n**CH08-F38 — COUNTING LOW DIGITAL SKILL AS INAPPROPRIATENESS**\n\nActual target users are removed from the panel.\n\n**CH08-F39 — DISRUPTING VOLUNTARINESS WITH REWARD**\n\nThe participant is under pressure that they reasonably cannot refuse.\n\n**CH08-F40 — LEAVE THE PARTICIPANT'S CHOICE TO THE CLIENT**\n\nThe audited entity selects the people who can determine the outcome of the panel.\n\n## 53. AUDIT PROCEDURE\n\n### Step 1 — Map Participant Sources\n\nEach panel is recorded separately for partner, open call, and client source.\n\n### Step 2 — Assign Selection Status\n\nAn appropriate status between PSI-0 and PSI-5 is determined for each source.\n\n### Step 3 — Compare Coverage with Target Population\n\nDetermine which target users are not present in the panel.\n\n### Step 4 — Save Participant Selection Probability or Model\n\nKnown probability, quota or voluntary selection is explained.\n\n### Step 5 — Freeze Conformity Rules\n\nAge Country Language Product Plan Interface Uniqueness Extraction relationship conditions are versioned.\n\n### Step 6 — Review the Invitation Text\n\nIt is evaluated whether there is positive or negative guidance.\n\n### Step 7 — Examine the Reward Structure\n\nIt is confirmed that the reward is completely independent of the outcome.\n\n### Step 8 — Verify Participant Country, Language, and Product Access\n\nPrivacy-protective verification levels are assigned.\n\n### Step 9 — Perform Unique Human Check\n\nIntra-source and inter-source repetitions are examined.\n\n### Step 10 — Record Relationship and Familiarity Areas\n\nCustomer, employee, competitor, AI provider, and brand familiarity are determined.\n\n### Step 11 — Distribute Tasks According to Assignment Plan\n\nProduct, prompt, queue, and wave assignments are verified.\n\n### Step 12 — Record Panel Conditioning\n\nNew, repeated, and previously participated users in the same test are separated.\n\n### Step 13 — Set Up Participation Funnel\n\nInvitation, acceptance, start, completion, and validity numbers are calculated.\n\n### Step 14 — Blind the Validity Review\n\nProtocol validity is determined without the AI response score being seen as much as possible.\n\n### Step 15 — Examine Fraud Signals\n\nMultiple signals and human evaluation are used.\n\n### Step 16 — Analyse My Non-Response and Panel Loss\n\nThe known characteristics of those who participate and those who do not participate are compared.\n\n### Step 17 — Generate Design and Calibration Weights\n\nProbabilistic and non-probabilistic sources are handled with separate methods.\n\n### Step 18 — Calculate the Weight Diagnoses\n\nWeight distribution Design effect Effective sample Trimming effect is calculated.\n\n### Step 19 — Perform Sensitivity Analysis\n\nRaw, main heavily and alternative reasonable results are compared.\n\n### Step 20 — Create the Privacy and Evidence Manifest\n\nFull proof is in a secure repository; anonymous proof identities are kept in a public record.\n\n### Step 21 — Publish the Public Election Card\n\nPanel sources, participation funnel, biases, and weighting effect are made visible.\n\n## 54. REQUIRED EVIDENCE\n\nTarget-population record; sampling frame; panel providers; participant sources; selection status; probability or non-probability panel class; panel-registration method; invitation text; participant-admission rules; participant-country verification; language-proficiency record; AI-product access record; plan and user surface; unique-person verification; cross-source duplication checks; brand familiarity; frequency of AI use; relationship with the audited entity; relationship with the AI provider; participant compensation; bonus conditions; task-assignment method; randomisation and blocking record; product and prompt order; panel conditioning; prior pilot participation; number invited; number consenting; number starting the task; number completing it; number of valid observations; reasons for nonresponse; panel attrition; reasons for invalidity; and fraud indicators.\n\nReview and objection records Design weights Non-response settings Calibration variables Propensity model, if any Weight trimming record Effective sample size Sensitivity results Panel provider effect Unweighted and weighted scores Consent record Data storage and access policy Redaction rules Public evidence manifest Responsible person or institution\n\n## 55. AUDIT CHECKLIST\n\nFrom which sources did the participants come? Was each source probabilistic, voluntary, or open call? Was in-panel randomness presented like world population randomness? Did the supervising institution intervene in participant selection? Are customer and employee participants separate? Were AI provider or competitor relationships documented? How was country eligibility verified? Is language proficiency at the level of actual use? Was the correct AI product, plan, and interface used? How many accounts or panels did the same person participate in? Was cross-checking between sources conducted? Did the invitation text guide the outcome? Is the reward dependent on the content of the AI response? Did speed incentives reduce protocol quality? Was task assignment random or pre-rules based? Did the participant choose the product or prompt themselves?\n\nWas the product and prompt order balanced? Was panel conditioning recorded? Did participants share task results? Is the invite–accept–start–complete–validity funnel visible? Are non-responders different in known characteristics? Does panel attrition affect the time trend? Was the validity decision made before the AI response was scored? Were negative or critical responses more often considered invalid? Does fraud detection carry multiple signals? Did VPN or shared devices cause automatic exclusion? Was an appeal path given to the participant? Were weight variables defined before the result? Were improbable panel weights correctly labelled? Are target totals reliable? Are unweighted and weighted results presented together? Is the effective sample size clear? Was the impact of weight trimming shown?\n\nHave groups that are never observed been shown as largely represented? Were model-based predictions presented as if they were direct observations? Are there personal details in participant screenshots? Can users from small countries be re-identified? Were account passwords or private chats requested? Is participant consent clear and understandable? Is there a clearly accountable owner for participant choice and weighting records?\n\n## 56. OBJECTIONS AND RESPONSES\n\n### Objection 1 — \"If we use a professional research panel, why are we still talking about bias?\"\n\nA professional panel can strengthen data quality and field management. Panel members are also:\n\n- voluntary,\n\n- accustomed to research,\n\n- different in technology use\n\nIt is possible. Professionalism does not eliminate bias. It makes measuring and managing easier.\n\n### Objection 2 — “If the age and country quotas are met, why shouldn't the panel be considered representative?”\n\nParticipants:\n\n- Frequency of AI usage,\n\n- device,\n\n- plan,\n\n- brand awareness,\n\n- digital skill\n\nIt may differ in terms of maintenance. Quotas only balance the areas they cover.\n\n### Objection 3 — “Can't weighting correct these differences?”\n\nIt can partially correct differences for which measured and reliable target totals exist. It cannot definitively correct differences that are never observed or measured.\n\n### Objection 4 — “The institution's clients are real users; why shouldn't they be included on the main panel?”\n\nThey can be taken. However, they should be taken with the actual proportion and relationship status they carry in the general population. The entire customer list is not a general population panel.\n\n### Objection 5 — \"Why would it be a problem if the participant knows the brand?\"\n\nIt doesn’t have to be a problem. They are part of the natural user population. If familiarity is unbalanced, the result may vary. Therefore, it is recorded and analysed if necessary.\n\n### Objection 6 — \"Heavy AI users perform the task more accurately; wouldn’t it be better to select them?\"\n\nIt may be easier for protocol testing. However, if we want to measure real population experience, inexperienced users are also important. If the task is too complex, the measurement tool should be corrected. The population should not be simplified.\n\n### Objection 7 — 'If the participant misunderstood the response, why should we keep the observation?'\n\nThe participant's task is not to evaluate the response, but to capture it correctly. If the capture protocol is applied correctly, the AI response is preserved. Material evaluation is the responsibility of the adjudicators.\n\n### Objection 8 — “Shouldn’t a strict automatic filter be used to prevent fake participation?”\n\nAutomatic filtering is useful. If it becomes the sole final decision, it may unfairly exclude real users. Multiple signals and proportional human review are needed.\n\n### Objection 9 — 'If we do not make a thousand screenshots public, can the method be considered proven?'\n\nFull screen images:\n\n- personal information,\n\n- account security,\n\n- redefinition\n\nmay carry a risk. To the public:\n\n- edited examples,\n\n- observation manifest,\n\n- integrity hashes,\n\n- independent audit access\n\ncan be provided. Provability is not identity disclosure.\n\n### Objection 10 — “If the weighted score is very different from the raw score, which one is correct?”\n\nTwo results answer different questions. The raw score is the result of the people in the panel. The weighted score is the target population estimate. A large difference indicates that the panel composition is important. Both should be published.\n\n### Objection 11 — “If the effective sample dropped to 349, was collecting 1,000 observations wasted?”\n\nThe oversampled groups may have strengthened the subgroup analysis. However, the sensitivity of the global forecast may be lower than the raw number of 1,000. This fact should not be hidden.\n\n### Objection 12 — \"Can the non-probability panel not be used at all?\"\n\nIt can be used. Specifically:\n\n- pilot,\n\n- low-resource language,\n\n- small country,\n\n- Quick discovery\n\nIt is valuable for. It should be published with the correct status and restrictions.\n\n### Objection 13 — \"Isn't it more reliable to choose all the participants from new people?\"\n\nIt may be strong for cross-sectional population estimation. Repeat users are valuable for measuring longitudinal change. New and repeat panel roles should be kept separate.\n\n### Objection 14 — \"Doesn't collecting data from so many participants increase privacy risk?\"\n\nIt can. Therefore:\n\n- data minimisation,\n\n- anonymous identity,\n\n- redaction,\n\n- access control,\n\n- storage duration\n\nare mandatory parts of the protocol. Unnecessary information should not be collected.\n\n## COMMON PROVISION OF SECTION 59\n\nThe easiest number in a GEO study:\n\n#### 1,000\n\nIt is possible. The hardest question is this:\n\n> Which 1,000?\n\nTwo panels with the same country and age quotas can produce completely different results. In one:\n\n- intense AI users,\n\n- paid plan subscribers,\n\n- technology workers\n\nare the majority. In the other:\n\n- infrequent users,\n\n- mobile users,\n\n- people with low digital skills\n\nIt is more visible. Both are 1,000 people. It is not the same population. If you hide where participants come from and only publish the weighted result, you turn weighting into a veil of trust rather than a method of control. You cannot weight a person who is not present in a sample. You cannot change a user's response because it is wrong. A wrong answer is not a failure of the research. It is the event that the research is trying to measure. If a participant has followed the protocol incorrectly despite giving the correct answer, you cannot preserve the record. A correct answer does not make a fake observation real. You can pay one person more. But not to produce the answer you want. You can pay according to a country's living conditions. But you cannot buy the participant's willingness. A thousand screenshots can serve as evidence. You are not obliged to disclose the identity of a thousand people to the public.\n\nEvidence integrity and human privacy can be protected together. Therefore, NOMOS's eighth measurement law is as follows:\n\n> The power of the sample is determined not only by how many people it consists of, but also by why and how these people were selected.\n\nThe ninth law is as follows:\n\n> Weighting rebalances the observed population; it does not create the unobserved human.\n\nThe tenth law is:\n\n> An incorrect AI answer can be valid data; a correct AI answer can be invalid data.\n\nThe eleventh law is as follows:\n\n> The participant's reward should be given not for finding the desired result, but for proving the observed result without changing it.\n\nThe twelfth law is as follows:\n\n> Provability does not require the participant to give up privacy.\n\n## NOMOS's Chapter 8 Commandment\n\n> Do not bring me a thousand people alone. / Show me which world these thousand people come from.\n\n> Do not treat a volunteer panel as a random population sample, a customer list as the general public or an employee as an independent user.\n\n> Don't think that all the differences you didn't measure have improved just because the country and age quotas were met.\n\n> Do not multiply one person by the number of their accounts, devices, languages or panel memberships.\n\n> Don't put those who love the brand, those who hate the brand, and those who don't know the brand at all in the same invisible box.\n\n> Pay the participant not to find the correct answer, but to save the first suitable answer they see without changing it.\n\n> Do not change the user when AI gives a wrong answer. / Do not check if the answer is correct when the user breaks the protocol.\n\n> Do not make a person who is not on the panel appear weighted as if they exist.\n\n> Do not hide the raw result. / Do not declare only the weighted result as the truth. / Show the effective sample. / Show its sensitivity.\n\n> Do not declare someone using a VPN as fake, someone using a shared device as a bot, or someone using assistive technology as automation.\n\n> Investigate fraud. / But do not turn the investigation itself into a new exclusion bias.\n\n> Keep a screenshot. / Protect the person's name, account, and private life.\n\n> First, define the population. / Then show the framework. / Then explain who you chose, who did not come, who could not complete it, and why you excluded someone. / And only after that, decide how much your score can speak for humanity.\n\n## The Chapter's Closing Sentence\n\n> In GEO-1000, a participant is more than a person carrying a response. A participant is a human sampling unit whose capacity to represent others in the target population must be justified through the selection path, weight, independence and privacy safeguards.\n\n## Normative Core\n\n> Every GEO-1000 observation MUST retain an auditable participant-selection path from the target population and sampling frame through selection, invitation, consent, assignment, completion, protocol validation, exclusion, and final weighting. Within-panel random selection MUST NOT be represented as random selection from the global target population unless the full population inclusion mechanism supports that claim. Participant recruitment, assignment, compensation, validation, replacement, exclusion, and weighting MUST remain independent of whether the AI response is positive, negative, correct, incorrect, refused, critical, or commercially favourable. Persons, accounts, devices, sessions, languages, panel memberships, and AI-product observations MUST remain distinct units. Probability, opt-in, open-call, client-supplied, institutional, and hybrid participant sources MUST be separately identified. Weighting MAY reduce observed imbalance, but MUST NOT be represented as creating observations for populations absent from the sampling frame or correcting unmeasured selection with certainty. Unweighted results, weighted results, weight diagnostics, coverage gaps, nonresponse, effective sample size, and sensitivity analyses MUST remain visible. Fraud and duplication decisions MUST use proportionate multi-signal review and MUST NOT automatically treat VPN use, shared devices, institutional networks, or accessibility tools as fabrication. Participant evidence MUST remain verifiable without unnecessarily disclosing identity, account credentials, private conversations, or precise location. Every participant-selection and weighting decision MUST be versioned and attributable to an accountable human or organisation.","character_count":74787,"record_sha256":"e6b4992a46eb5682c69999429d6637eeeab7e5a445c75f05850d06652db9b1a0"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-09","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":11,"chapter_number":9,"item_number":null,"title":"The AI System Register and Comparability","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K02","K03"],"source_word_count":8223,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":9,\"id\":\"NOMOS-GEO-AUDIT-CH09\",\"title\":\"The AI System Register and Comparability\",\"subtitle\":null,\"sourceFile\":\"9.cu bölüm.docx\",\"sourceSha256\":\"053EE2C8569DE504C5C2EC0DC22A7B91C8A52B8F4AC42660EF573A8D7EF922C5\",\"sourceWordCount\":8223,\"sourceIds\":[\"K02\",\"K03\"],\"machine\":{\"chapter\":9,\"chapterId\":\"NOMOS-GEO-AUDIT-CH09\",\"title\":\"The AI System Register and Comparability\",\"subtitle\":null,\"sourceIds\":[\"K02\",\"K03\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH09-R01\",\"normativeRuleEnglish\":\"Every GEO-1000 observation MUST be linked to a versioned AI System Register record describing the measured product instance, including as applicable: - provider, - product, - user surface, - plan, - displayed or verified model identity, - model-identity confidence, - retrieval and web configuration, - citation capability and use, - tool access and use, - session state, - memory, - custom instructions, - prior context, - personalisation, - country, - locale, - language, - deployment state, - measurement time, - and known or observed system changes. Provider name, product name, displayed model label, base model, retrieval system, and user-facing product MUST NOT be treated as interchangeable identities. Feature availability MUST remain distinct from feature use. Live-product results MUST NOT be represented as frozen-model results, and frozen-model results MUST NOT be represented as full consumer-product experience. Comparisons MUST define their object and receive a disclosed comparability level based on matched entity, prompt, population, language, country, time, session, plan, interface, retrieval, tools, reference records, and adjudication rules. The same provider or displayed model MUST NOT create an automatic assumption of comparability. Different providers MAY be compared when user-facing conditions are sufficiently aligned and the resulting claim is limited to the measured product behaviour. Material product changes during a measurement wave MUST create a separate system state, sub-wave, modelled change factor, or mixed-state warning. Unknown configuration fields MUST remain UNKNOWN and MUST NOT be imputed as controlled or equivalent. Every system-registry and comparability decision MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Her GEO-1000 gözlemi; sağlayıcı, ürün, kullanıcı yüzeyi, plan, görünen veya doğrulanmış model kimliği, model kimliği güveni, retrieval ve web yapılandırması, citation kapasitesi ve kullanımı, araç erişimi, oturum durumu, hafıza, özel talimatlar, önceki bağlam, kişiselleştirme, ülke, locale, dil, dağıtım durumu, ölçüm zamanı ve bilinen sistem değişikliklerini ilgili olduğu ölçüde taşıyan sürümlü bir AI Sistem Sicili kaydına bağlanmalıdır. Sağlayıcı adı, ürün adı, görünen model etiketi, temel model, retrieval sistemi ve kullanıcı ürünü birbirinin yerine kullanılamaz. Aynı sağlayıcı veya model etiketi otomatik karşılaştırılabilirlik oluşturmaz. Farklı sağlayıcılar, kullanıcı koşulları yeterince eşleştirildiğinde ölçülen ürün davranışı bakımından karşılaştırılabilir.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH09-MB0001\",\"type\":\"paragraph\",\"text\":\"54. MACHINE-READABLE AI SYSTEM RECORD\",\"sourceParagraph\":1758},{\"blockId\":\"CH09-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1759},{\"blockId\":\"CH09-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1760},{\"blockId\":\"CH09-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1761},{\"blockId\":\"CH09-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1762},{\"blockId\":\"CH09-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"aiSystemRecord\\\": {\",\"sourceParagraph\":1763},{\"blockId\":\"CH09-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"systemRecordId\\\": \\\"NGAI-SYNTH-ORION-WEB-FREE-W1-001\\\",\",\"sourceParagraph\":1764},{\"blockId\":\"CH09-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"provider\\\": {\",\"sourceParagraph\":1765},{\"blockId\":\"CH09-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"providerId\\\": \\\"SYNTH-ORION\\\",\",\"sourceParagraph\":1766},{\"blockId\":\"CH09-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"providerName\\\": \\\"Orion Systems\\\"\",\"sourceParagraph\":1767},{\"blockId\":\"CH09-MB0011\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1768},{\"blockId\":\"CH09-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"product\\\": {\",\"sourceParagraph\":1769},{\"blockId\":\"CH09-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"productId\\\": \\\"SYNTH-ORION-CHAT\\\",\",\"sourceParagraph\":1770},{\"blockId\":\"CH09-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"productName\\\": \\\"Orion Chat\\\",\",\"sourceParagraph\":1771},{\"blockId\":\"CH09-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"productType\\\": [\",\"sourceParagraph\":1772},{\"blockId\":\"CH09-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"PT-1\\\",\",\"sourceParagraph\":1773},{\"blockId\":\"CH09-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"PT-2\\\"\",\"sourceParagraph\":1774},{\"blockId\":\"CH09-MB0018\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1775},{\"blockId\":\"CH09-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"surface\\\": \\\"WEB\\\",\",\"sourceParagraph\":1776},{\"blockId\":\"CH09-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"plan\\\": \\\"FREE\\\",\",\"sourceParagraph\":1777},{\"blockId\":\"CH09-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"applicationVersion\\\": \\\"6.1\\\"\",\"sourceParagraph\":1778},{\"blockId\":\"CH09-MB0022\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1779},{\"blockId\":\"CH09-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"model\\\": {\",\"sourceParagraph\":1780},{\"blockId\":\"CH09-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"displayedLabel\\\": \\\"Orion Standard\\\",\",\"sourceParagraph\":1781},{\"blockId\":\"CH09-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"identityConfidence\\\": \\\"MI-1\\\",\",\"sourceParagraph\":1782},{\"blockId\":\"CH09-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"versionedDeploymentId\\\": null,\",\"sourceParagraph\":1783},{\"blockId\":\"CH09-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"dynamicRoutingPossible\\\": true,\",\"sourceParagraph\":1784},{\"blockId\":\"CH09-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"claimAttributionLevel\\\": \\\"PRODUCT_SURFACE_ONLY\\\"\",\"sourceParagraph\":1785},{\"blockId\":\"CH09-MB0029\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1786},{\"blockId\":\"CH09-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"knowledgeConfiguration\\\": {\",\"sourceParagraph\":1787},{\"blockId\":\"CH09-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"mode\\\": \\\"KG-3\\\",\",\"sourceParagraph\":1788},{\"blockId\":\"CH09-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"webFeatureAvailable\\\": true,\",\"sourceParagraph\":1789},{\"blockId\":\"CH09-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"webFeatureEnabled\\\": true,\",\"sourceParagraph\":1790},{\"blockId\":\"CH09-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"webUseForObservation\\\": \\\"USED_CONFIRMED\\\",\",\"sourceParagraph\":1791},{\"blockId\":\"CH09-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"citationFeatureAvailable\\\": true,\",\"sourceParagraph\":1792},{\"blockId\":\"CH09-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"citationsDisplayed\\\": true,\",\"sourceParagraph\":1793},{\"blockId\":\"CH09-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"privateKnowledgeSource\\\": false\",\"sourceParagraph\":1794},{\"blockId\":\"CH09-MB0038\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1795},{\"blockId\":\"CH09-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"tools\\\": {\",\"sourceParagraph\":1796},{\"blockId\":\"CH09-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"available\\\": [\",\"sourceParagraph\":1797},{\"blockId\":\"CH09-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"WEB_SEARCH\\\"\",\"sourceParagraph\":1798},{\"blockId\":\"CH09-MB0042\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1799},{\"blockId\":\"CH09-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"usedForObservation\\\": [\",\"sourceParagraph\":1800},{\"blockId\":\"CH09-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"WEB_SEARCH\\\"\",\"sourceParagraph\":1801},{\"blockId\":\"CH09-MB0045\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":1802},{\"blockId\":\"CH09-MB0046\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1803},{\"blockId\":\"CH09-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"session\\\": {\",\"sourceParagraph\":1804},{\"blockId\":\"CH09-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"sessionType\\\": \\\"ST-1\\\",\",\"sourceParagraph\":1805},{\"blockId\":\"CH09-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"newConversation\\\": true,\",\"sourceParagraph\":1806},{\"blockId\":\"CH09-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"persistentMemory\\\": \\\"OFF\\\",\",\"sourceParagraph\":1807},{\"blockId\":\"CH09-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"customInstructions\\\": \\\"OFF\\\",\",\"sourceParagraph\":1808},{\"blockId\":\"CH09-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"priorConversationContext\\\": false,\",\"sourceParagraph\":1809},{\"blockId\":\"CH09-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"enterpriseWorkspace\\\": false\",\"sourceParagraph\":1810},{\"blockId\":\"CH09-MB0054\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1811},{\"blockId\":\"CH09-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"localisation\\\": {\",\"sourceParagraph\":1812},{\"blockId\":\"CH09-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"country\\\": \\\"TR\\\",\",\"sourceParagraph\":1813},{\"blockId\":\"CH09-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"tr-TR\\\",\",\"sourceParagraph\":1814},{\"blockId\":\"CH09-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"promptLanguage\\\": \\\"tr\\\",\",\"sourceParagraph\":1815},{\"blockId\":\"CH09-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"responseLanguage\\\": \\\"tr\\\"\",\"sourceParagraph\":1816},{\"blockId\":\"CH09-MB0060\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1817},{\"blockId\":\"CH09-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"deployment\\\": {\",\"sourceParagraph\":1818},{\"blockId\":\"CH09-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"W1\\\",\",\"sourceParagraph\":1819},{\"blockId\":\"CH09-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"timestampUtc\\\": \\\"2026-08-18T12:15:00Z\\\",\",\"sourceParagraph\":1820},{\"blockId\":\"CH09-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"experimentalState\\\": \\\"UNKNOWN\\\",\",\"sourceParagraph\":1821},{\"blockId\":\"CH09-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"knownChangeEvent\\\": null,\",\"sourceParagraph\":1822},{\"blockId\":\"CH09-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"observedInterfaceVariant\\\": \\\"STANDARD\\\"\",\"sourceParagraph\":1823},{\"blockId\":\"CH09-MB0067\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1824},{\"blockId\":\"CH09-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"accessResult\\\": {\",\"sourceParagraph\":1825},{\"blockId\":\"CH09-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"AVAILABLE\\\",\",\"sourceParagraph\":1826},{\"blockId\":\"CH09-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"rateLimitEncountered\\\": false,\",\"sourceParagraph\":1827},{\"blockId\":\"CH09-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"technicalRetryUsed\\\": false,\",\"sourceParagraph\":1828},{\"blockId\":\"CH09-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"systemRefusal\\\": false\",\"sourceParagraph\":1829},{\"blockId\":\"CH09-MB0073\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1830},{\"blockId\":\"CH09-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"configurationFingerprint\\\": {\",\"sourceParagraph\":1831},{\"blockId\":\"CH09-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"algorithm\\\": \\\"SYNTHETIC-HASH\\\",\",\"sourceParagraph\":1832},{\"blockId\":\"CH09-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": \\\"SCF-SYNTH-001\\\"\",\"sourceParagraph\":1833},{\"blockId\":\"CH09-MB0077\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1834},{\"blockId\":\"CH09-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"evidence\\\": {\",\"sourceParagraph\":1835},{\"blockId\":\"CH09-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"productMetadataCaptured\\\": true,\",\"sourceParagraph\":1836},{\"blockId\":\"CH09-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"fullScreenshotCaptured\\\": true,\",\"sourceParagraph\":1837},{\"blockId\":\"CH09-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"systemSettingsCaptured\\\": true,\",\"sourceParagraph\":1838},{\"blockId\":\"CH09-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceConfidence\\\": \\\"MEDIUM\\\"\",\"sourceParagraph\":1839},{\"blockId\":\"CH09-MB0083\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1840},{\"blockId\":\"CH09-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":1841},{\"blockId\":\"CH09-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"registryOwnerRole\\\": \\\"AI_SYSTEM_REGISTRY_OWNER\\\",\",\"sourceParagraph\":1842},{\"blockId\":\"CH09-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"recordValidatedBy\\\": \\\"SYSTEM_CONFIGURATION_REVIEWER\\\"\",\"sourceParagraph\":1843},{\"blockId\":\"CH09-MB0087\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1844},{\"blockId\":\"CH09-MB0088\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1845},{\"blockId\":\"CH09-MB0089\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1846},{\"blockId\":\"CH09-MB0090\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":1847},{\"blockId\":\"CH09-MB0091\",\"type\":\"paragraph\",\"text\":\"to the actual provider,\",\"sourceParagraph\":1848},{\"blockId\":\"CH09-MB0092\",\"type\":\"paragraph\",\"text\":\"to the real model,\",\"sourceParagraph\":1849},{\"blockId\":\"CH09-MB0093\",\"type\":\"paragraph\",\"text\":\"to the real product\",\"sourceParagraph\":1850},{\"blockId\":\"CH09-MB0094\",\"type\":\"paragraph\",\"text\":\"it does not belong.\",\"sourceParagraph\":1851},{\"blockId\":\"CH09-MB0095\",\"type\":\"paragraph\",\"text\":\"AI product sample is a machine-readable synthetic representation.\",\"sourceParagraph\":1852},{\"blockId\":\"CH09-MB0096\",\"type\":\"paragraph\",\"text\":\"55. MACHINE-READABLE COMPARISON RECORD\",\"sourceParagraph\":1854},{\"blockId\":\"CH09-MB0097\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1855},{\"blockId\":\"CH09-MB0098\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1856},{\"blockId\":\"CH09-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1857},{\"blockId\":\"CH09-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1858},{\"blockId\":\"CH09-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonRecord\\\": {\",\"sourceParagraph\":1859},{\"blockId\":\"CH09-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonId\\\": \\\"NGCMP-SYNTH-001\\\",\",\"sourceParagraph\":1860},{\"blockId\":\"CH09-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonQuestion\\\": \\\"Which consumer AI product produces more accurate entity representation under matched clean-session conditions?\\\",\",\"sourceParagraph\":1861},{\"blockId\":\"CH09-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"systemA\\\": \\\"NGAI-SYNTH-ORION-WEB-PAID-W1-002\\\",\",\"sourceParagraph\":1862},{\"blockId\":\"CH09-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"systemB\\\": \\\"NGAI-SYNTH-HELIOS-WEB-FREE-W1-003\\\",\",\"sourceParagraph\":1863},{\"blockId\":\"CH09-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonLevel\\\": \\\"CP-4\\\",\",\"sourceParagraph\":1864},{\"blockId\":\"CH09-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"matchedDimensions\\\": {\",\"sourceParagraph\":1865},{\"blockId\":\"CH09-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"entityRecordVersion\\\": true,\",\"sourceParagraph\":1866},{\"blockId\":\"CH09-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackVersion\\\": true,\",\"sourceParagraph\":1867},{\"blockId\":\"CH09-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"promptVersion\\\": true,\",\"sourceParagraph\":1868},{\"blockId\":\"CH09-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"country\\\": true,\",\"sourceParagraph\":1869},{\"blockId\":\"CH09-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": true,\",\"sourceParagraph\":1870},{\"blockId\":\"CH09-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": true,\",\"sourceParagraph\":1871},{\"blockId\":\"CH09-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"measurementWindow\\\": true,\",\"sourceParagraph\":1872},{\"blockId\":\"CH09-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"populationFrame\\\": true,\",\"sourceParagraph\":1873},{\"blockId\":\"CH09-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"newConversation\\\": true,\",\"sourceParagraph\":1874},{\"blockId\":\"CH09-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"persistentMemoryOff\\\": true,\",\"sourceParagraph\":1875},{\"blockId\":\"CH09-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"customInstructionsOff\\\": true,\",\"sourceParagraph\":1876},{\"blockId\":\"CH09-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"webAccess\\\": true,\",\"sourceParagraph\":1877},{\"blockId\":\"CH09-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"citationMode\\\": true\",\"sourceParagraph\":1878},{\"blockId\":\"CH09-MB0121\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1879},{\"blockId\":\"CH09-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"unmatchedDimensions\\\": {\",\"sourceParagraph\":1880},{\"blockId\":\"CH09-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"provider\\\": true,\",\"sourceParagraph\":1881},{\"blockId\":\"CH09-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"plan\\\": true,\",\"sourceParagraph\":1882},{\"blockId\":\"CH09-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"displayedModel\\\": true,\",\"sourceParagraph\":1883},{\"blockId\":\"CH09-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"userInterfaceDetails\\\": true\",\"sourceParagraph\":1884},{\"blockId\":\"CH09-MB0127\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1885},{\"blockId\":\"CH09-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"allowedClaims\\\": [\",\"sourceParagraph\":1886},{\"blockId\":\"CH09-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"CONSUMER_PRODUCT_REPRESENTATION_COMPARISON\\\",\",\"sourceParagraph\":1887},{\"blockId\":\"CH09-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"MATCHED_CLEAN_SESSION_RESULT\\\"\",\"sourceParagraph\":1888},{\"blockId\":\"CH09-MB0131\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1889},{\"blockId\":\"CH09-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"prohibitedClaims\\\": [\",\"sourceParagraph\":1890},{\"blockId\":\"CH09-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"PURE_BASE_MODEL_SUPERIORITY\\\",\",\"sourceParagraph\":1891},{\"blockId\":\"CH09-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"GLOBAL_NATIVE_REACH_SUPERIORITY\\\",\",\"sourceParagraph\":1892},{\"blockId\":\"CH09-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"PERMANENT_PRODUCT_SUPERIORITY\\\"\",\"sourceParagraph\":1893},{\"blockId\":\"CH09-MB0136\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1894},{\"blockId\":\"CH09-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"nativeReachComparison\\\": false,\",\"sourceParagraph\":1895},{\"blockId\":\"CH09-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"commonSupportComparison\\\": true,\",\"sourceParagraph\":1896},{\"blockId\":\"CH09-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"replicationStatus\\\": \\\"SINGLE_WAVE\\\",\",\"sourceParagraph\":1897},{\"blockId\":\"CH09-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":1898},{\"blockId\":\"CH09-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonOwnerRole\\\": \\\"COMPARABILITY_ADJUDICATOR\\\",\",\"sourceParagraph\":1899},{\"blockId\":\"CH09-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"approvedBeforePublicRanking\\\": true\",\"sourceParagraph\":1900},{\"blockId\":\"CH09-MB0143\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1901},{\"blockId\":\"CH09-MB0144\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1902},{\"blockId\":\"CH09-MB0145\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1903},{\"blockId\":\"CH09-MB0146\",\"type\":\"paragraph\",\"text\":\"56. MACHINE-READABLE RULE OF THE SECTION\",\"sourceParagraph\":1905},{\"blockId\":\"CH09-MB0147\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH09-R01\",\"sourceParagraph\":1906},{\"blockId\":\"CH09-MB0148\",\"type\":\"paragraph\",\"text\":\"Every GEO-1000 observation MUST be linked to a versioned AI System\",\"sourceParagraph\":1908},{\"blockId\":\"CH09-MB0149\",\"type\":\"paragraph\",\"text\":\"Registry record describing the measured product instance, including as\",\"sourceParagraph\":1909},{\"blockId\":\"CH09-MB0150\",\"type\":\"paragraph\",\"text\":\"applicable:\",\"sourceParagraph\":1910},{\"blockId\":\"CH09-MB0151\",\"type\":\"paragraph\",\"text\":\"- provider,\",\"sourceParagraph\":1912},{\"blockId\":\"CH09-MB0152\",\"type\":\"paragraph\",\"text\":\"- product,\",\"sourceParagraph\":1913},{\"blockId\":\"CH09-MB0153\",\"type\":\"paragraph\",\"text\":\"- user surface,\",\"sourceParagraph\":1914},{\"blockId\":\"CH09-MB0154\",\"type\":\"paragraph\",\"text\":\"- plan,\",\"sourceParagraph\":1915},{\"blockId\":\"CH09-MB0155\",\"type\":\"paragraph\",\"text\":\"- displayed or verified model identity,\",\"sourceParagraph\":1916},{\"blockId\":\"CH09-MB0156\",\"type\":\"paragraph\",\"text\":\"- model-identity confidence,\",\"sourceParagraph\":1917},{\"blockId\":\"CH09-MB0157\",\"type\":\"paragraph\",\"text\":\"- retrieval and web configuration,\",\"sourceParagraph\":1918},{\"blockId\":\"CH09-MB0158\",\"type\":\"paragraph\",\"text\":\"- citation capability and use,\",\"sourceParagraph\":1919},{\"blockId\":\"CH09-MB0159\",\"type\":\"paragraph\",\"text\":\"- tool access and use,\",\"sourceParagraph\":1920},{\"blockId\":\"CH09-MB0160\",\"type\":\"paragraph\",\"text\":\"- session state,\",\"sourceParagraph\":1921},{\"blockId\":\"CH09-MB0161\",\"type\":\"paragraph\",\"text\":\"- memory,\",\"sourceParagraph\":1922},{\"blockId\":\"CH09-MB0162\",\"type\":\"paragraph\",\"text\":\"- custom instructions,\",\"sourceParagraph\":1923},{\"blockId\":\"CH09-MB0163\",\"type\":\"paragraph\",\"text\":\"- prior context,\",\"sourceParagraph\":1924},{\"blockId\":\"CH09-MB0164\",\"type\":\"paragraph\",\"text\":\"- personalisation,\",\"sourceParagraph\":1925},{\"blockId\":\"CH09-MB0165\",\"type\":\"paragraph\",\"text\":\"- country,\",\"sourceParagraph\":1926},{\"blockId\":\"CH09-MB0166\",\"type\":\"paragraph\",\"text\":\"- locale,\",\"sourceParagraph\":1927},{\"blockId\":\"CH09-MB0167\",\"type\":\"paragraph\",\"text\":\"- language,\",\"sourceParagraph\":1928},{\"blockId\":\"CH09-MB0168\",\"type\":\"paragraph\",\"text\":\"- deployment state,\",\"sourceParagraph\":1929},{\"blockId\":\"CH09-MB0169\",\"type\":\"paragraph\",\"text\":\"- measurement time,\",\"sourceParagraph\":1930},{\"blockId\":\"CH09-MB0170\",\"type\":\"paragraph\",\"text\":\"- and known or observed system changes.\",\"sourceParagraph\":1931},{\"blockId\":\"CH09-MB0171\",\"type\":\"paragraph\",\"text\":\"Provider name, product name, displayed model label, base model, retrieval\",\"sourceParagraph\":1933},{\"blockId\":\"CH09-MB0172\",\"type\":\"paragraph\",\"text\":\"system, and user-facing product MUST NOT be treated as interchangeable\",\"sourceParagraph\":1934},{\"blockId\":\"CH09-MB0173\",\"type\":\"paragraph\",\"text\":\"identities.\",\"sourceParagraph\":1935},{\"blockId\":\"CH09-MB0174\",\"type\":\"paragraph\",\"text\":\"Feature availability MUST remain distinct from feature use.\",\"sourceParagraph\":1937},{\"blockId\":\"CH09-MB0175\",\"type\":\"paragraph\",\"text\":\"Live-product results MUST NOT be represented as frozen-model results, and\",\"sourceParagraph\":1939},{\"blockId\":\"CH09-MB0176\",\"type\":\"paragraph\",\"text\":\"frozen-model results MUST NOT be represented as full consumer-product\",\"sourceParagraph\":1940},{\"blockId\":\"CH09-MB0177\",\"type\":\"paragraph\",\"text\":\"experience.\",\"sourceParagraph\":1941},{\"blockId\":\"CH09-MB0178\",\"type\":\"paragraph\",\"text\":\"Comparisons MUST define their object and receive a disclosed\",\"sourceParagraph\":1943},{\"blockId\":\"CH09-MB0179\",\"type\":\"paragraph\",\"text\":\"comparability level based on matched entity, prompt, population,\",\"sourceParagraph\":1944},{\"blockId\":\"CH09-MB0180\",\"type\":\"paragraph\",\"text\":\"language, country, time, session, plan, interface, retrieval, tools,\",\"sourceParagraph\":1945},{\"blockId\":\"CH09-MB0181\",\"type\":\"paragraph\",\"text\":\"reference records, and adjudication rules.\",\"sourceParagraph\":1946},{\"blockId\":\"CH09-MB0182\",\"type\":\"paragraph\",\"text\":\"The same provider or displayed model MUST NOT create an automatic\",\"sourceParagraph\":1948},{\"blockId\":\"CH09-MB0183\",\"type\":\"paragraph\",\"text\":\"assumption of comparability.\",\"sourceParagraph\":1949},{\"blockId\":\"CH09-MB0184\",\"type\":\"paragraph\",\"text\":\"Different providers MAY be compared when user-facing conditions are\",\"sourceParagraph\":1951},{\"blockId\":\"CH09-MB0185\",\"type\":\"paragraph\",\"text\":\"sufficiently aligned and the resulting claim is limited to the measured\",\"sourceParagraph\":1952},{\"blockId\":\"CH09-MB0186\",\"type\":\"paragraph\",\"text\":\"product behaviour.\",\"sourceParagraph\":1953},{\"blockId\":\"CH09-MB0187\",\"type\":\"paragraph\",\"text\":\"Material product changes during a measurement wave MUST create a separate\",\"sourceParagraph\":1955},{\"blockId\":\"CH09-MB0188\",\"type\":\"paragraph\",\"text\":\"system state, sub-wave, modelled change factor, or mixed-state warning.\",\"sourceParagraph\":1956},{\"blockId\":\"CH09-MB0189\",\"type\":\"paragraph\",\"text\":\"Unknown configuration fields MUST remain UNKNOWN and MUST NOT be imputed\",\"sourceParagraph\":1958},{\"blockId\":\"CH09-MB0190\",\"type\":\"paragraph\",\"text\":\"as controlled or equivalent.\",\"sourceParagraph\":1959},{\"blockId\":\"CH09-MB0191\",\"type\":\"paragraph\",\"text\":\"Every system-registry and comparability decision MUST be versioned and\",\"sourceParagraph\":1961},{\"blockId\":\"CH09-MB0192\",\"type\":\"paragraph\",\"text\":\"attributable to an accountable human or organisation.\",\"sourceParagraph\":1962},{\"blockId\":\"CH09-MB0193\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1963},{\"blockId\":\"CH09-MB0194\",\"type\":\"paragraph\",\"text\":\"Each GEO-1000 observation must be linked to a versioned AI System Register record that carries, to the extent relevant, provider, product, user surface, plan, visible or verified model ID, model ID confidence, retrieval and web configuration, citation capability and usage, tool access, session status, memory, custom instructions, previous context, personalisation, country, locale, language, deployment status, measurement time, and known system changes. Provider name, product name, visible model label, base model, retrieval system, and user product cannot be used interchangeably. The same provider or model label does not create automatic comparability. Different providers are comparable in terms of measured product behaviour when user conditions are sufficiently matched.\",\"sourceParagraph\":1964}]}}","text":"## Chapter Boundary\n\nThe first eight chapters established the following: a single response is not a GEO score; representation is a distribution; the audited entity must be identified before results are observed; official representation and verified entity reality must remain separate; an eligible user population must be defined for every AI product; 1,000 valid observations must be allocated through a population-based sample design; smaller countries and languages must remain visible through complementary panels; and participant selection, exclusion and weighting must remain independent of response content. We must now define the other side of the experiment:\n\n> Which AI system are participants actually measuring?\n\nThe name of a provider? A base model? A chat product? A free web interface? A paid mobile app? A session with web search enabled? A response mode that does not cite sources? An enterprise workspace supported by in-house data? An account with memory enabled? Temporary or clean chat? Are different user interfaces of the same provider the same system? Should two products using the same model family be measured in the same way? Even if a product shows the same model name on the screen:\n\n- retrieval,\n\n- source selection,\n\n- security rules,\n\n- personalisation,\n\n- tool access,\n\n- experimental features\n\nIf it changes, can the results be compared? When an AI product is updated in the middle of a measurement wave, can the first and last responses of the wave be considered to come from the same system? A user's:\n\n- web search is on,\n\n- memory is on,\n\n- has previous conversations,\n\n- paid plan\n\nresponses; another user's:\n\n- search is off,\n\n- memory is off,\n\n- new session,\n\n- free plan\n\nCan it be merged in the same cell as its answer? No. The measurement principle established by the previous sections is clear: There is no single “AI”; the search index, retrieval system, reranker, language model, citation layer, and safety system can behave differently. The same framework requires each auditable result to carry:\n\n- model or product,\n\n- date,\n\n- country,\n\n- language,\n\n- query set,\n\n- number of repetitions,\n\n- measurement record\n\nThis section operates those principles within an AI System Register. This section:\n\n- the distinction between provider, product, model, and user surface,\n\n- AI product instance,\n\n- model identity confidence levels,\n\n- retrieval and tool configurations,\n\n- memory and personalisation states,\n\n- differences in plan, interface, and experimental features,\n\n- product changes,\n\n- system configuration at the time of measurement,\n\n- levels of comparability between products,\n\n- Native Reach and Common Support comparisons,\n\n- version drift over time,\n\ndefines. This chapter does not yet:\n\n- full translation and semantic equivalence system of prompts,\n\n- all details of controlled and natural user panels,\n\n- NOMOS Capture software architecture,\n\n- the atomic claim scoring of the responses,\n\n- the final model coefficients and the NOMOS scoring formula\n\ndoes not finalise. The key question of Section 9 is:\n\n> Before comparing two responses from the same prompt, how do we prove that these responses actually come from comparable AI product conditions?\n\n## NOMOS Challenge\n\nTwo users of the same provider ask the same question: “What kind of company is Apple.com?” First user:\n\n- is using the free web product,\n\n- opens a new chat,\n\n- has memory off,\n\n- does not have web access,\n\nreceives an answer only with the product’s built-in knowledge capacity. Second user:\n\n- Using a paid mobile product,\n\n- memory is on,\n\n- has previously talked about Apple,\n\n- web search is on,\n\nreceives answers with current sources and citations. First answer: “Apple is a technology company that produces consumer electronics.” Second answer: “Apple is a global technology company offering hardware, software, and digital services. You can see current corporate information in these sources.” Both answers come from the same provider.\n\nIs this the result of the same system? No. Now consider two other users. Both see the same model name on the screen. In one user’s country, the web feature is on. In the other user’s country, it is off. In one user’s answer, source links are visible. The other user’s interface does not show citations even if it uses the same sources. There is the same model label.\n\nIs there the same measurement cell? No. Now consider two products from different providers. Both:\n\n- new chat,\n\n- memory is off,\n\n- same language,\n\n- same country,\n\n- same time window,\n\n- web access enabled,\n\n- same prompt version\n\nare measured under these conditions. They are based on different underlying models. Are they comparable? Yes, they can be comparable in terms of a specific user outcome. Because the object of comparison is:\n\n> not whether they use the same underlying model, but what representation they produce under the same defined user condition.\n\nNow, let the product version remain unchanged for the first 500 observations of the measurement wave. During the next 500 observations, the provider:\n\n- change the retrieval system,\n\n- security rules,\n\n- user interface\n\nThe report alone says: “1,000 users tested the same AI product.” In reality, two different product situations may have been measured. The first provision of this section is as follows:\n\n> The provider name is not sufficient to identify the measured AI system.\n\nIts second provision states:\n\n> The model name is not the entirety of the product behaviour that reaches the user.\n\nIts third provision states:\n\n> The same provider, the same model label, or the same logo does not guarantee comparability.\n\nIts fourth provision states:\n\n> Different providers may be comparable in terms of user outcomes if the measurement conditions are sufficiently matched.\n\nIts fifth provision states:\n\n> If a product materially changes during measurement, it cannot be reported as a single fixed system.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to define the AI product audited within GEO-1000 with all its material configurations and to determine under which conditions different product outcomes can be compared. The section standardises the following distinctions:\n\n- Provider and AI product\n\n- AI product and base model\n\n- Model family and specific deployment\n\n- Model label and verified model identity\n\n- User surface and API\n\n- Web and mobile product\n\n- Free plan and paid plan\n\n- Consumer product and enterprise product\n\n- Basic model answer and retrieval-augmented answer\n\n- Live web search with retrieval\n\n- Citation display with source usage\n\n- Tool usage with tool access\n\n- Session context with memory\n\n- System policy with custom instruction\n\n- Material reality with personalisation\n\n- Temporary chat with new conversation\n\n- Product status over time with static product name\n\n- Observed behaviour change with provider-announced update\n\n- Comparison of Native Reach and Common Support\n\n- Comparable product example within the same product family\n\n- User result versus in-model abstract ability\n\n- Product absence versus product failure\n\n- Technical error versus actual rejection\n\n- Same prompt versus same experiment\n\n- Same average versus same system behaviour\n\n- Instant product snapshot versus long-term product performance\n\nAt the end of this section, each GEO-1000 audit should be able to answer the following questions:\n\n> Which provider's which product, which surface, which plan, and which configuration were measured?\n\n> How much was the model name seen on the screen really validated?\n\n> What were web, retrieval, citation, memory, personalisation, and tool status?\n\n> Did the product remain the same throughout the measurement?\n\n> At what comparability level were the two product results placed side by side?\n\n> Is the thing being compared the base model, the consumer product, the user surface, or the entire system chain?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\n> Each GEO-1000 observation must be linked not only with the provider or model name; it should be connected to a versioned AI System Register record that identifies the AI product instance at the time of measurement.\n\nAn AI product instance must carry at least the following fields to the extent relevant:\n\n- Provider\n\n- Product name\n\n- User surface\n\n- Product type\n\n- Plan or account class\n\n- Apparent model label\n\n- Model ID confidence level\n\n- Product or deployment version\n\n- Country and locale\n\n- Language\n\n- Web or retrieval access\n\n- Citation feature\n\n- Tool access\n\n- Memory status\n\n- Custom instruction status\n\n- Previous chat context\n\n- Session type\n\n- Personalisation status\n\n- Security or policy mode, if known\n\n- Experimental features\n\n- Measurement date and UTC time\n\n- Measurement wave\n\n- Product change log\n\n- Evidence and record confidence\n\nWhen any of these fields materially change:\n\n- new AI product instance,\n\n- new experimental cell,\n\n- new analysis version\n\nmay be required.\n\n## 3. “AI SYSTEM” IS NOT A SINGLE OBJECT [K02; K03]\n\nIn this protocol, “AI system” is a general and high-level term. A real user response can arise from the combination of multiple layers. Candidate chain:\n\nR = F(M, O, K, Q, T, P, S, U, C, L, t)\n\nHere:\n\n- M: base or guiding model\n\n- O: product orchestration layer\n\n- K: retrieval or knowledge source layer\n\n- Q: reranking and source selection\n\n- T: tools\n\n- P: policy and security layer\n\n- S: session, memory, and personalisation\n\n- U: user account and plan\n\n- C: country and locale\n\n- L: language\n\n- t: time\n\nResponse seen by the user:\n\n### R\n\nThe response visible to a user is not necessarily the output of the base model alone. The principal consumer-facing object measured by GEO-1000 is therefore:\n\n> the product behaviour presented to the user, not a model abstraction.\n\nControlled API or model research can also be conducted. It is not the same result as the main consumer panel.\n\n## 4. SEVEN IDENTITY LAYERS OF THE AI SYSTEM\n\nEach audited AI system must be defined across seven layers, insofar as each layer is relevant.\n\n### 4.1. Provider Layer\n\nThe organisation that publishes or operates the product. Provider name:\n\n- is not the same as\n\nproduct ID,\n\n- model ID,\n\n- user surface\n\nA provider can offer more than one product. Within the same product, more than one model or orchestration system may be used.\n\n### 4.2. Product Layer\n\nThe commercial, consumer, enterprise, or technical product accessed by the user. Example types:\n\n- General chat product\n\n- Search-assisted answer product\n\n- Enterprise workspace\n\n- Educational product\n\n- Embedded assistant\n\n- Developer API\n\n- Custom knowledge-based product\n\nThe product name alone does not describe the entire configuration.\n\n### 4.3. User surface Layer\n\nIt is the interface through which the user interacts with the product. Web Mobile Desktop Browser extension Operating system integration In-platform embedded interface\n\n### API\n\nEnterprise portal Different surfaces of the same product:\n\n- different features,\n\n- different model,\n\n- different source representation,\n\n- different memory\n\nmay be used.\n\n### 4.4. Model or Inference Engine Layer\n\nIt is the model, model family, or routing system used in response generation. Model ID:\n\n- open,\n\n- partially visible,\n\n- unknown,\n\n- dynamic\n\nIt is possible. A product can direct a user query to different models. In this case, one cannot say: 'A single fixed model was measured.'\n\n### 4.5. Orchestration and Information Layer\n\nIt may include the following:\n\n- Retrieval\n\n- Live web search\n\n- Internal search\n\n- Source ranking\n\n- Reranking\n\n- Citation selection\n\n- Calling a vehicle\n\n- Query rewriting\n\n- Document shredding\n\n- Answer merging\n\nThis layer is of fundamental importance for the representation of GEO. Because it can affect which record:\n\n- is accessed,\n\n- is selected,\n\n- is carried into the response,\n\n- is shown as a reference\n\ncan be influenced.\n\n### 4.6. Policy and Personalisation Layer\n\nIt may include the following:\n\n- Security rules\n\n- System instructions\n\n- Product policy\n\n- Memory\n\n- Special instructions\n\n- User history\n\n- Regional rules\n\n- Age or account class\n\n- Corporate administrator settings\n\nThis layer affects the:\n\n- scope of the response,\n\n- its tone,\n\n- its rejection,\n\n- its recommendation\n\ncan change.\n\n### 4.7. Distribution and Time Layer\n\nIndicates which of the following the product is under at a specific moment:\n\n- is meaningful with\n\n- experimental feature,\n\n- regional distribution,\n\n- user segment,\n\n- A/B test\n\nThe same product name can have different distribution conditions for different users on the same day.\n\n## 5. AI PRODUCT EXAMPLE\n\nThe core system unit within GEO-1000:\n\n#### AI Product Instance\n\nCanonical operational term:\n\n#### AI Product Instance\n\nAn AI Product Instance is defined as:\n\nA_i = (v, p, s, r, m, k, t, x, u, c, l, ω)\n\ncan be represented as. Here:\n\n- v: provider\n\n- p: product\n\n- s: user surface\n\n- r: plan or account class\n\n- m: model ID or label\n\n- k: retrieval and information structuring\n\n- t: tool configuration\n\n- x: memory, personalisation, and session state\n\n- u: user or organisational setting class\n\n- c: country and locale\n\n- l: language\n\n- ω: time, deployment, and experimental feature state\n\nAn observation: Oj must be linked to a specific: Ai product sample.\n\n## 6. DIFFERENCE BETWEEN PRODUCT SAMPLE AND PRODUCT NAME\n\nThe product name can be fixed. The product sample may vary. Example: Let the Orion AI product name remain unchanged. However, these two measurements are different product samples:\n\n#### Product Sample 1\n\nWeb Free plan Memory off Web access off New chat Turkish Turkey Wave 1\n\n#### Product Sample 2\n\nMobile Paid plan Memory on Web access on Previous chat context available Turkish Turkey Wave 1 These can exist under the same brand. They are not the same experimental cell.\n\n## 7. MODEL IDENTITY CONFIDENCE LEVELS\n\nThe model name cannot always be known with the same certainty. Therefore, the model identity can be recorded at the following levels.\n\n### MI-0 — UNKNOWN MODEL\n\nIt is unknown which model or model family the product uses. The result is attributed only to the product surface.\n\n### MI-1 — USER-DISPLAYED LABEL\n\nA model label is visible in the user interface. The label is recorded. It is not assumed to fully verify the back-end deployment.\n\n### MI-2 — PROVIDER-DOCUMENTED MODEL FAMILY\n\nThe provider's product documents describe a specific model family or product–model relationship. It may still be unknown whether individual observation is directed exactly to that model.\n\n### MI-3 — SESSION-CONFIRMED DEPLOYMENT\n\nThe session verifies a specific distribution of API or trusted product metadata.\n\n### MI-4 — VERSIONED ENDPOINT OR DEPLOYMENT\n\nA specific version, endpoint, or deployment ID can be recorded. It is stronger for controlled comparison.\n\n### MI-5 — REPRODUCIBLE FROZEN MODEL ARTEFACT\n\nThere is a static and reproducible model or deployment structure. It may be rare in live consumer products. If model ID trust is low, the result:\n\n#### should be attributed to product behaviour\n\nIt cannot be definitively attributed to a specific model.\n\n## 8. LIMIT OF THE APPARENT MODEL NAME\n\nSeeing the model name in a user interface is valuable. However, the following questions may still remain open: Are all queries directed to the same model? Does the model change depending on the difficulty of the query? Does security or retrieval use a separate model? Does the provider apply the same label to different distributions? Does the model label only indicate the product family? Is the user part of an experimental rollout? Therefore, the correct reporting might be: “Model X label was shown in the user interface.” Without stronger evidence, the following sentence should not be used: “All responses are definitively generated solely by Model X.”\n\n## 9. PRODUCT TYPE\n\nEach AI product should be classified with a primary type and secondary types.\n\n### PT-1 — General Conversational Product\n\nIt is a general chat and question-answer product.\n\n### PT-2 — Search- or Retrieval-Grounded Product\n\nSearch or retrieval is a fundamental component in answer generation.\n\n### PT-3 — Enterprise Knowledge Product\n\nIt can use internal documents, workspaces, or special data sources.\n\n### PT-4 — Embedded Assistant\n\nIt is embedded into another product or platform.\n\n### PT-5 — Developer or API Product\n\nIt is used by a developer or application instead of the end user.\n\n### PT-6 — Domain-Specific Product\n\nIt is designed for a specific profession or industry.\n\n### PT-7 — Multimodal Product\n\nInputs other than text, such as visual, audio, or document entries, are part of the main experience. A product can carry more than one type. Comparison should be made within the same type or with an explicit scope.\n\n## 10. USER SURFACE REGISTER\n\nDifferent surfaces of the same product should be recorded separately. Surface areas:\n\n- Surface ID\n\n- Web/mobile/desktop/API\n\n- Application version\n\n- Operating system, if applicable\n\n- Browser, if applicable\n\n- Product language\n\n- Visible model selection\n\n- Availability of tools\n\n- Citation view\n\n- File upload feature\n\n- Memory settings\n\n- Corporate management settings\n\n- Experimental features\n\n- Measurement time\n\nIf a web and mobile interface are found to be materially the same, the results can be combined. This equality should not be assumed; it must be tested.\n\n## 11. PLAN AND ACCOUNT CLASS\n\nThe plan type may have features that can affect the user's response. Plan statuses:\n\n- Free\n\n- Paid individual\n\n- Professional\n\n- Corporate\n\n- Education\n\n- Developer\n\n- Trial\n\n- Invitation or limited access\n\n- Unknown\n\nPlans may differ in the following areas:\n\n- Model\n\n- Usage limit\n\n- Web access\n\n- Memory\n\n- Tools\n\n- Source citation\n\n- Response length\n\n- Speed or priority\n\n- Enterprise data\n\n- Security and retention settings\n\nIf there is no plan information:\n\n### The PLAN_UNKNOWN\n\nstatus should be used. Free and paid results cannot silently be merged as the same product behaviour.\n\n## 12. RETRIEVAL AND WEB ACCESS\n\nAn AI response can be generated in one of three basic knowledge states.\n\n### KG-0 — UNDETERMINED KNOWLEDGE MODE\n\nThe case of retrieval or web usage cannot be determined.\n\n### KG-1 — MODEL-INTERNAL RESPONSE\n\nA response known or strongly assumed not to use live or external information.\n\n### KG-2 — RETRIEVAL-ASSISTED RESPONSE\n\nRetrieval may have been used from pre-indexed or in-product sources.\n\n### KG-3 — LIVE WEB OR SEARCH-ASSISTED RESPONSE\n\nLive or near-measurement time web/search access has been used.\n\n### KG-4 — PRIVATE KNOWLEDGE-GROUNDED RESPONSE\n\nA private data source provided by the institution or the user has been used.\n\n### KG-5 — HYBRID OR MULTI-SOURCE RESPONSE\n\nMultiple layers of information have been used together. These modes may produce different results in the same prompt. They should be clearly distinguished in comparison.\n\n## 13. RETRIEVAL ACCESS IS NOT THE SAME AS RETRIEVAL USAGE\n\nThe product may have a web search feature. However, it may not have been used in the specific answer. Three fields should be recorded separately: Was the feature available? Was the feature on or selectable? Was there evidence that it was used in the specific answer? Example statuses:\n\n### AVAILABLE_NOT_USED\n\n### AVAILABLE_USAGE_UNKNOWN\n\n### USED_CONFIRMED\n\n### NOT_AVAILABLE\n\n### DISABLED\n\n### UNKNOWN\n\nThe mere presence of a source link can be a strong indicator for usage. The absence of a source link does not necessarily prove that retrieval was not used.\n\n## 14. CITATION LAYER\n\nUsing a source is not the same as showing a citation. A product:\n\n- may use a source but may not show it,\n\n- may show a citation but the citation may not support the claim,\n\n- may link the same source to different sentences,\n\nThe user can hide the citation according to the user surface. The citation record must contain the following fields: Is the citation feature available? Was the citation shown in the specific response? Number of citations Source field names Citation–claim match Visibility on the user surface Expandability of the citation Accessibility of the source Citation generation time Products that show and do not show citations cannot be directly compared in terms of \"number of sources\". First, product behaviour and interface differences must be explained.\n\n## 15. TOOL ACCESS\n\nThe AI product may access the following tools:\n\n- Web search\n\n- Browser\n\n- Calculator\n\n- Code execution\n\n- File reading\n\n- Map\n\n- Calendar\n\n- Corporate knowledge base\n\n- Product catalogue\n\n- Other APIs\n\nTool access may affect your response:\n\n- its timeliness,\n\n- evidence,\n\n- scope of the response,\n\n- accuracy\n\nTool registry:\n\n- whether the tool is available,\n\n- whether it is active,\n\n- whether it was used in the response,\n\n- whether the tool result is retained\n\nshould be shown.\n\n## 16. MEMORY, SPECIAL INSTRUCTIONS, AND PERSONALISATION\n\nThe response visible to the user may be affected by the following fields:\n\n- Persistent memory\n\n- Chat history\n\n- Special instructions\n\n- User's name or profile\n\n- Previous brand conversations\n\n- Preferences\n\n- Country and locale\n\n- Corporate workspace settings\n\n- Segmentation done by the provider\n\nThese fields must be recorded in three separate layers.\n\n### 16.1. Persistent Personalisation\n\nInformation protected across accounts or sessions.\n\n### 16.2. In-Conversation Context\n\nPrevious messages within the current chat.\n\n### 16.3. System or Organisational Setting\n\nAdministrator or product instructions that the user may not see. Complete information may not always be available. Unknown situations must be explicitly recorded.\n\n## 17. SESSION TYPES\n\nCandidate session statuses:\n\n### ST-1 — NEW STANDARD CONVERSATION\n\nNew standard conversation. Memory or special instructions are recorded separately.\n\n### ST-2 — TEMPORARY OR EPHEMERAL CONVERSATION\n\nTemporary session intended to be separated from permanent history. Actual features should be verified in the product registry.\n\n### ST-3 — EXISTING CONVERSATION\n\nThere is previous conversation context. It may be valuable for the natural panel. It is kept separate in the controlled panel.\n\n### ST-4 — ENTERPRISE OR SHARED WORKSPACE SESSION\n\nThere may be an enterprise or shared information area.\n\n### ST-5 — UNKNOWN SESSION STATE\n\nThe session status could not be reliably determined. If different session types are to be combined in the same cell, the rationale and weighting must be explained.\n\n## 18. PRODUCT POLICY AND REJECTION\n\nAn AI product may not respond to a specific question due to:\n\n- security,\n\n- law,\n\n- product policy,\n\n- regional rule,\n\n- does not imply any suitability in terms of age,\n\n- enterprise setting\n\nreason. Rejection:\n\n- basic model deficiency,\n\n- product policy,\n\n- interface,\n\n- may be due to the user account\n\nGEO-1000 main user control records the rejection result that reaches the user. However, if the reason for the rejection is known, it is additionally classified. A product's policy layer is also part of the user experience.\n\n## 19. EXPERIMENTAL FEATURES AND A/B DISTRIBUTIONS\n\nLive AI products may not give the same features to all users at the same time. Participants:\n\n- different user interface,\n\n- different model prompting,\n\n- different web feature,\n\n- different response format\n\nmay see. Experimental status indicators:\n\n- Beta label on the interface\n\n- Provider documentation\n\n- Feature difference\n\n- Systematic UI difference between users\n\n- Behaviour breaking during measurement\n\nmay occur. If the experimental feature is unknown:\n\n### EXPERIMENTAL_STATE_UNKNOWN\n\nshould be used. The actual variation within the product is not data defect. It is a distribution feature of the system. However, it should be considered in comparability.\n\n## 20. AI System Register\n\nAn AI System Register should be created for each audited product. The registry should carry three levels.\n\n### 20.1. Provider and Product Master Record\n\nIncludes relatively permanent areas:\n\n- Provider\n\n- Product name\n\n- Product type\n\n- Main user interfaces\n\n- Official access countries\n\n- Plan classes\n\n- Supported language records\n\n- General features\n\n### 20.2. Measurement Wave Product Record\n\nFor a specific wave:\n\n- product version,\n\n- plan features,\n\n- model labels,\n\n- web and tool features,\n\n- memory behaviour,\n\n- access status,\n\n- known updates\n\nis recorded.\n\n### 20.3. Observation Level System Record\n\nActually observed for each participant:\n\n- interface,\n\n- model label,\n\n- plan,\n\n- vehicle usage,\n\n- citation,\n\n- session,\n\n- personalisation,\n\n- time\n\ncarries fields. The main record alone cannot remove observation-level differences.\n\n## 21. SYSTEM CONFIGURATION FINGERPRINT\n\nA System Configuration Fingerprint can be created to facilitate the comparability of an AI product instance. Candidate fields:\n\nSCF = H(canonicalise(p, s, r, m, k, t, x, c, l, w))\n\nHere, H is the function that produces an integrity value from the versioned fields. The fingerprint:\n\n- whether two observations belong to the same configuration,\n\n- whether there have been material changes in the wave\n\nIt can help to identify. Fingerprint: does not prove the real backend model, does not make layers that the provider has not disclosed visible. It only reinforces whether the recorded configuration has been changed later.\n\n## 22. WHAT IS COMPARABILITY?\n\nTwo results are considered comparable when:\n\n> The difference between them can be sufficiently explained by monitored product behaviour rather than irrelevant differences in measurement design.\n\nComparability is not absolute equality. Different AI products:\n\n- model,\n\n- provider,\n\n- user surface.\n\nThese need not all be identical. The material conditions relevant to the comparison must, however, be:\n\n- equalisation,\n\n- layering,\n\n- recording,\n\n- modelling in analysis\n\nare required.\n\n## 23. THE OBJECT OF THE COMPARISON MUST BE DEFINED IN ADVANCE\n\nThe same data can answer different comparison questions.\n\n### 23.1. User Product Comparison\n\n\"Which consumer product produces a more accurate representation under the same user conditions?\" The object being compared is the entire product chain.\n\n### 23.2. Model Comparison\n\n\"Which model produces a more accurate response under the same controlled prompt and tool-free condition?\" API or fixed model access may be required.\n\n### 23.3. Retrieval Comparison\n\n“Is there a difference between the web-accessible and non-accessible modes of the same product?” The model and other conditions are kept as constant as possible.\n\n### 23.4. Plan Comparison\n\n“Do free and paid users see the same brand representation?” The plan difference is the object of the audit.\n\n### 23.5. Interface Comparison\n\n“Does the web and mobile product produce different representations under the same conditions?”\n\n### 23.6. Time Comparison\n\n“Did the representation of the same product change across different waves?” Product and population changes should be separated. Without specifying the comparison question: “Product A is better than Product B.” is excessively broad.\n\n## 24. DIMENSIONS OF COMPARABILITY\n\nWhen comparing two products or waves, the following areas should be evaluated:\n\n- Audited entity\n\n- Reality package version\n\n- Prompt ID and version\n\n- Language and locale\n\n- Country\n\n- Target population\n\n- Sample source\n\n- Participant composition\n\n- Session type\n\n- Memory\n\n- Special instructions\n\n- Previous conversation\n\n- Product plan\n\n- User surface\n\n- Model label\n\n- Retrieval/web status\n\n- Tools\n\n- Citation feature\n\n- Measurement time\n\n- Product updates\n\n- Adjudication rule\n\n- Result statuses\n\n- Weighting\n\nNot all dimensions need to be the same. Different dimensions limit the scope of the comparison.\n\n## 25. COMPARABILITY LEVELS\n\n### CP-0 — INCOMPARABLE\n\nMaterial measurement conditions are unknown or seriously inconsistent. Direct performance comparison cannot be made.\n\n### CP-1 — DESCRIPTIVE CO-OCCURRENCE\n\nThere are results from two products or waves. However:\n\n- population,\n\n- prompt,\n\n- time,\n\n- session\n\nare not equal. Only descriptive side-by-side display can be made.\n\n### CP-2 — PARTIALLY ALIGNED\n\nSome core fields match. Material differences continue. Limited comparison can be made.\n\n### CP-3 — MATCHED USER-SURFACE COMPARISON\n\nSame entity Same prompt version Same language/country Similar user population Same time window Registered product features are available. Consumer product results are comparable.\n\n### CP-4 — CONTROLLED MATCHED COMPARISON\n\nAdditionally:\n\n- session,\n\n- memory,\n\n- personalisation,\n\n- retrieval,\n\n- tools,\n\n- plan\n\nare matched in a controlled or open manner. Provides stronger product or feature comparison.\n\n### CP-5 — REPLICATED AND CALIBRATED COMPARISON\n\nComparison:\n\n- in multiple waves,\n\n- with independent users or panels,\n\n- with the same normative evaluation\n\nrepeated; variability and product changes have been calibrated. These levels are not a “scientific accuracy score.” They indicate the limit that a comparison claim can carry.\n\n## 26. WHAT TO DO IN CP-0 CONDITION?\n\nIf two products cannot be compared, the results are not ignored. They can be reported separately:\n\n- “This is the Native Reach result of Product A.”\n\n- “This is the result of Product B under different users and product conditions.”\n\nHowever: one cannot say, “A is 12 points better than B.” Incomparability is not a failure, it is a limitation of the method.\n\n## 27. COMPARABILITY FOR THE SAME PROVIDER\n\nResults from the same provider cannot be automatically compared. The following differences may be present:\n\n- Product\n\n- Plan\n\n- Model\n\n- Web access\n\n- Interface\n\n- Memory\n\n- Enterprise data\n\n- Country distribution\n\n- Experimental feature\n\n- Time\n\nComparison of the same provider should also be evaluated according to the CP level.\n\n## 28. COMPARABILITY FOR DIFFERENT PROVIDERS\n\nUser products belonging to different providers:\n\n- same target population,\n\n- same prompt,\n\n- same time,\n\n- similar user surface,\n\n- clear product configuration\n\nIt is comparable underneath. Being a different provider does not prevent the comparison. However, differences in features determine what the comparison measures. Example: Product A uses live web. Product B does not. The result can answer this question: 'Which of the two product chains offered to the user produces more accurate representation?' It does not directly answer this question: 'Which base model is more knowledgeable?'\n\n## 29. COMPARISON OF WEB-SUPPORTED AND NON-WEB PRODUCTS\n\nTwo separate comparisons are possible.\n\n### 29.1. Natural Product Comparison\n\nEach product is measured under the default real user condition. A product that uses the web uses the web. One that does not use it does not. Result: it is a comparison of the real consumer product experience.\n\n### 29.2. Feature Matched Comparison\n\nIf possible, both products:\n\n- measured in web open,\n\n- web closed\n\nmodes separately as well. This can better separate the retrieval effect. Each product may not offer the same options. Missing mode:\n\n### RECORDED AS NOT AVAILABLE\n\nThat status must remain explicit.\n\n## 30. PRODUCTS THAT SHOW AND DO NOT SHOW CITATIONS\n\nFor a product without citation feature: it is possible to give a “Zero citation.” However, its meaning should be explained: the product may not be referencing sources. It is not certain that it does not use any sources. Citation coverage and accuracy coverage are separate outcomes. Citation comparison should only be made among products that support citations or with separate feature status.\n\n## 31. NATIVE REACH COMPARISON\n\nEach product is measured within its actual reach population. Advantage: shows the product’s natural global impact. Limitation: populations are different, country and language distributions may vary. Native Reach scores: show what the product does within its reached population. Should not be used alone in direct product superiority ranking.\n\n## 32. COMMON SUPPORT COMPARISON\n\nThe common access population of the compared products is used. Advantage:\n\n- same country,\n\n- same language,\n\n- same user condition\n\nprovides a fairer comparison. Limitation: preserves the actual global access difference of the products, the common universe may be very narrow. The Common Support score should be published together with the Native Reach score.\n\n## 33. PAIRED USER COMPARISON\n\nIf the same user tests two or more AI products, user characteristics can be controlled. However:\n\n- task order,\n\n- answer recall,\n\n- product familiarity,\n\n- special profile of users who can access all products\n\nIt can create bias. In a paired design:\n\n- product order should be randomised,\n\n- previous responses should not be shared,\n\nthe same user dependency should be maintained in the analysis.\n\n## 34. INDEPENDENT USER COMPARISON\n\nEach product is tested by different users, but selected from the same population frame. Advantage: task contamination is reduced, more natural product users can be chosen. Limitation: user groups may not be completely equal. Weighting and stratification are required. Paired and independent user results can be separate but complementary.\n\n## 35. TIME COMPARABILITY\n\nWhen comparing today's result of an AI product with past results, the following changes should be examined:\n\n- Model\n\n- Product version\n\n- Web/retrieval\n\n- Plan\n\n- Country access\n\n- Language support\n\n- Prompt\n\n- Entity reality\n\n- Adjudication\n\n- Population framework\n\n- Weighting\n\nTime difference may not only arise from the AI product.\n\n## 36. TYPES OF PRODUCT CHANGES\n\n### UC-1 — DOCUMENTED MINOR CHANGE\n\nA minor change disclosed by the provider and assessed not to materially alter the measurement object.\n\n### UC-2 — DOCUMENTED MATERIAL CHANGE\n\nMaterial change described in model, retrieval, interface, or feature.\n\n### UC-3 — OBSERVED BEHAVIOUR CHANGE\n\nEven if there is no official announcement, systematic behaviour or interface change is observed.\n\n### UC-4 — ACCESS OR PLAN CHANGE\n\nCountry, plan, or user access changes.\n\n### UC-5 — UNKNOWN CHANGE EVENT\n\nThere is a break in measurement behaviour. The reason could not be verified.\n\n### UC-6 — PRODUCT REPLACEMENT OR RETIREMENT\n\nAn old product or model has ended and has been replaced by another structure. The type of change limits time series comparison.\n\n## 37. PRODUCT CHANGE WITHIN THE MEASUREMENT WAVE\n\nIf a material change occurs during a wave, the following options exist: The wave can be stopped and restarted. Separate sub-waves can be made before and after the change. The change can be modelled as an experimental factor. The wave is labelled as MIXED_SYSTEM_STATE. The following structure is forbidden: Combining all responses as if they result in a single fixed product outcome despite knowing about the change.\n\n## 38. BEHAVIOUR CAN CHANGE WITHOUT PRODUCT CHANGE\n\nEven if the provider does not announce an explicit update, the result may change. Possible reasons:\n\n- Change in the retrieval index\n\n- Updating of resources\n\n- New content appearing on the web\n\n- Reranking change\n\n- A/B test\n\n- Regional rollout\n\n- Security policy\n\n- Traffic routing\n\nTherefore, it cannot be said, “The provider did not update, the system was definitely the same.” Measurement evidence and behaviour breakdown should also be examined.\n\n## 39. LIVE PRODUCT AND FROZEN MODEL\n\nThere are two different research objects.\n\n### 39.1. Live Product Audit\n\nMeasures what real users see on the real consumer surface. Advantage:\n\n- shows real experience,\n\n- real source,\n\n- real policy,\n\n- real interface\n\nlimitations: it is variable, full reproduction may be difficult.\n\n### 39.2. Frozen Model Audit\n\nControlled repetition is conducted under a fixed version or endpoint. The advantage:\n\n- provides reproducibility,\n\n- experimental control\n\nThe limitation: may not fully represent the real consumer product. The main population layer of NOMOS GEO-1000 is based on the live user product. Frozen model audit is complementary laboratory work.\n\n## 40. PRODUCT UNAVAILABILITY AND TECHNICAL RESULTS\n\nThe following cases should be kept separate:\n\n### AR-1 — PRODUCT UNAVAILABLE\n\nThe product is not available for the country or user group. It is the result of access scope.\n\n### AR-2 — ACCOUNT OR PLAN INELIGIBLE\n\nThe plan or account condition assigned to the participant is not eligible. There may be a sampling or task assignment issue.\n\n### AR-3 — TEMPORARY TECHNICAL FAILURE\n\nA temporary technical error has occurred. Retry protocol should be predefined.\n\n### AR-4 — RATE OR USAGE LIMIT\n\nA usage quota or rate limit has been reached. It may be part of the real user experience.\n\n### AR-5 — SYSTEM REFUSAL\n\nThe product task has been refused due to policy or content. It is a valid system outcome.\n\n### AR-6 — NO USABLE OUTPUT\n\nNo usable response has been generated.\n\n### AR-7 — UNKNOWN ACCESS FAILURE\n\nWhy it could not be determined. These situations cannot be combined in the same \"no response\" box.\n\n## 41. TECHNICAL RETRY RULE\n\nA retry can be made in case of a temporary technical error or connection interruption. However:\n\n- if the AI response is not liked,\n\n- a rejection occurred,\n\n- incorrect information was produced\n\nretry cannot be made for these reasons. Technical retry:\n\n- same user,\n\n- same product,\n\n- same configuration,\n\n- must be made within the predetermined period.\n\nThe first technical incident must also be recorded.\n\n## 42. COMPARISON MATRIX\n\nA basic comparison record for two products could be as follows:\n\nThese results can be shown side by side. However: \"The basic model of Product B is better.\" cannot be said. The difference may stem from the entire product chain.\n\n## 43. SYNTHETIC APPLE.COM COMPARISON CASE\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The products, responses, and results below are entirely fictional. They do not represent the performance of the real Apple Inc. or any real AI provider. Let's consider three synthetic product examples.\n\n#### Product Example A\n\nProvider: Orion Product: Orion Chat Interface: Web Plan: Free Model label: Orion Standard Model Confidence: MI-1 Web access: Off Memory: Off Session: New Language: Turkish Locale: Turkey Wave: W1 Response: \"Apple is an American technology company that produces phones and computers.\"\n\n#### Product Sample B\n\nProvider: Orion Product: Orion Chat Interface: Mobile Plan: Paid Model Label: Orion Pro Model Confidence: MI-1 Web Access: Open Citation: Open Memory: On Previous Conversation: About Apple Language: Turkish Locale: Turkey Wave: W1 Response: \"Apple is a global technology company that provides hardware, software, and digital services. Current corporate information can be found in these sources.\"\n\n#### Product Sample C\n\nProvider: Helios Product: Helios Search Interface: Web Plan: Free Model Label: Unknown Model Confidence: MI-0 Web Access: Open Citation: Open Memory: None Session: New Language: Turkish Locale: Turkey Wave: W1 Response: \"Apple is a US-based technology company that develops consumer devices, operating systems, and digital services.\"\n\n### 43.1. Comparison of A and B\n\nThey are the same provider. However:\n\n- plan,\n\n- surface,\n\n- model label,\n\n- web,\n\n- memory,\n\n- previous conversation\n\nis different. The comparison level can be CP-1 or CP-2. The response difference cannot be attributed solely to the basic model difference.\n\n### 43.2. Comparison of A and C\n\nThey are different providers. However:\n\n- new session,\n\n- free plan,\n\n- web surface,\n\n- same language/country,\n\n- same wave\n\nin terms of some conditions match. Web access is different. Comparison: can be done as a natural consumer product outcome. It cannot be done as a basic model capability.\n\n### 43.3. Comparison of B and C\n\nBoth products have web access and citations enabled. However, B has memory and previous conversation, while C does not. The user experience of the products is comparable. It is not a controlled clean product comparison.\n\n### 43.4. Controlled Repeat\n\nIn product B:\n\n- memory is turned off,\n\n- a new chat is started,\n\n- special instructions are turned off,\n\nweb remains on. C is measured under the same condition. Comparison level: CP-3, or CP-4 with sufficient additional control. This example shows:\n\n> The same provider does not provide a close comparison; well-recorded different providers may be more comparable.\n\n## 44. SYNTHETIC PRODUCT CHANGE CASE\n\n### SYNTHETIC EXAMPLE\n\nOrion Chat W1 wave targets 1,000 observations. In the first 480 observations:\n\n- product version: 6.1\n\n- web feature: manual\n\n- attribution: limited\n\nIn the next 520 observations:\n\n- product version: 6.2\n\n- web feature: automatic\n\n- attribution: extended\n\nlet it be. Single result: if calculated as “Orion Chat W1 score,” two different system states mix. Correct options:\n\n#### Option 1\n\nW1A: first 480 W1B: next 520 are reported separately.\n\n#### Option 2\n\nThe wave is not continued. 1,000 new observations are taken in version 6.2.\n\n#### Option 3\n\nThe version is kept as an explicit experimental factor in statistical analysis. Incorrect option: The update is removed from the report and all observations are presented as a single fixed product.\n\n## 45. MINIMUM AI SYSTEM RECORD\n\nEach observation must carry at least the following fields:\n\n- NOMOS AI system ID\n\n- Provider\n\n- Product\n\n- Surface\n\n- Plan\n\n- Product type\n\n- Apparent model label\n\n- Model confidence level\n\n- Country\n\n- Locale\n\n- Language\n\n- Session type\n\n- Memory status\n\n- Custom instruction status\n\n- Previous conversation state\n\n- Retrieval/web feature\n\n- Retrieval usage status\n\n- Citation status\n\n- Tool status\n\n- Experimental feature status\n\n- UTC time\n\n- Wave\n\n- System fingerprint\n\n- Change events\n\n- Source of evidence\n\n- Record owner\n\nIf the material field is unknown, it cannot be left blank. UNKNOWN status should be used.\n\n## 46. MINIMUM COMPARISON CARD\n\nWhen comparing two products or waves, the public record should show:\n\n- Comparison question\n\n- Comparison level\n\n- Common target population\n\n- Native Reach differences\n\n- User equality\n\n- Language and country\n\n- Measurement time\n\n- Product, plan, and surface\n\n- Model identity confidence\n\n- Web/retrieval difference\n\n- Difference in memory and personalisation\n\n- Difference in tool\n\n- Difference in sample\n\n- Evaluation method\n\n- Material limitations\n\n- Directly comparable metrics\n\n- Non-comparable areas\n\nSingle number or ranking should not be published without this card.\n\n## 47. RULE OF COMPARABILITY\n\nAn auditor should answer these four questions: Was the same thing asked? Was the same human universe measured? Were the same user conditions applied? Were the same time and evaluation system used? Each of these questions:\n\n- completely,\n\n- partially,\n\n- no,\n\n- unknown\n\nis classified as. The comparison level is given accordingly.\n\n## 48. MANDATORY NORMATIVE PROVISIONS\n\n**CH09-N01**\n\nEvery GEO-1000 observation must be linked to a versioned AI System Register record.\n\n**CH09-N02**\n\nThe provider name cannot be used as the full identity of the measured AI product.\n\n**CH09-N03**\n\nThe product, user surface, plan, model, retrieval, tool, session, and personalisation layers must be separated from each other.\n\n**CH09-N04**\n\nThe model label on the user surface cannot be presented as a verified back-end model identity.\n\n**CH09-N05**\n\nThe model-identity confidence level must be recorded from MI-0 to MI-5 or through an equivalent openly defined status.\n\n**CH09-N06**\n\nIf the model identity cannot be verified, the result should be attributed only to the product and user surface.\n\n**CH09-N07**\n\nProducts belonging to the same provider are not automatically considered the same experimental cell.\n\n**CH09-N08**\n\nProducts belonging to different providers may be comparable when material user and experimental conditions are matched.\n\n**CH09-N09**\n\nWeb, mobile, desktop, embedded product and API surfaces cannot be combined without description.\n\n**CH09-N10**\n\nFree, paid, corporate, educational, and developer plans must be recorded as separate product terms.\n\n**CH09-N11**\n\nIf the plan is unknown, the result should carry the status PLAN_UNKNOWN.\n\n**CH09-N12**\n\nThe availability of the retrieval or web feature cannot be counted as proof that it was used in a specific response.\n\n**CH09-N13**\n\nRetrieval access, retrieval usage, and attribution must be recorded with separate statuses.\n\n**CH09-N14**\n\nA product that does not show attribution cannot automatically be considered as not using any resources.\n\n**CH09-N15**\n\nThe citations of the product being cited should be examined separately in terms of claim support.\n\n**CH09-N16**\n\nTool access and tool use in specific responses should be recorded separately.\n\n**CH09-N17**\n\nMemory, special instructions, previous chats, and corporate settings should be recorded separately as much as possible.\n\n**CH09-N18**\n\nAn unknown personalisation state cannot be assumed to be like a clean session.\n\n**CH09-N19**\n\nNew chat, temporary chat, current chat, and corporate workspace sessions cannot be merged without explanation.\n\n**CH09-N20**\n\nNatural user panel and controlled clean panel results should carry separate product status.\n\n**CH09-N21**\n\nIf an experimental feature or A/B distribution in the product is material, it should be preserved in the observation record.\n\n**CH09-N22**\n\nIf a material product change occurs during a wave, the wave should be divided into separate sub-versions, restarted, or the change should be explicitly modelled in the analysis.\n\n**CH09-N23**\n\nA known product change cannot be stored as if all observations came from a single fixed system.\n\n**CH09-N24**\n\nThe provider not explaining an update cannot be taken as definitive proof that the product has not changed.\n\n**CH09-N25**\n\nThe result of a live consumer product cannot be presented as if it were the result of a frozen baseline model.\n\n**CH09-N26**\n\nThe result of a frozen model in the laboratory cannot be presented as performance of a real user product.\n\n**CH09-N27**\n\nAbsence of the product, technical error, usage limit, system rejection, and lack of result must be recorded with separate statuses.\n\n**CH09-N28**\n\nIncorrect or negative AI response cannot be a reason for technical retry.\n\n**CH09-N29**\n\nThe subject of the comparison must be defined before data is collected.\n\n**CH09-N30**\n\nThe comparability level of two products must be visible in the public result.\n\n**CH09-N31**\n\nDefinite product ranking cannot be generated from CP-0 or CP-1 level results.\n\n**CH09-N32**\n\nUsing the same prompt alone does not create a comparable experiment.\n\n**CH09-N33**\n\nIn comparisons, the target entity, Truth Pack, prompt, language, country, time, population, session, product, and evaluation method should be examined together.\n\n**CH09-N34**\n\nNative Reach results cannot be directly ranked without explanation because they belong to different target populations.\n\n**CH09-N35**\n\nThe Common Support result cannot be presented as the product's actual global reach result.\n\n**CH09-N36**\n\nWeb-supported and web-free product comparison can be done as a user product comparison; it cannot be presented as a basic model superiority.\n\n**CH09-N37**\n\nIn paired comparisons made with the same user, product order and user dependency must be maintained.\n\n**CH09-N38**\n\nIn comparisons made with different users, user composition and weight differences must be visible.\n\n**CH09-N39**\n\nThe product version or configuration fingerprint must be stored in every measurement wave.\n\n**CH09-N40**\n\nThere must be an accountable human or institutional owner of the system registry and comparability decision.\n\n## 49. FORMS OF FAILURE\n\n**CH09-F01 — CONSIDERING THE PROVIDER AS A SINGLE AI**\n\nAll products, models, and surfaces of the provider are presented as a single system.\n\n**CH09-F02 — COUNT PRODUCT BEHAVIOUR BY MODEL NAME**\n\nRetrieval, security, interface, and personalisation are ignored.\n\n**CH09-F03 — DEFINITELY CONSIDER THE VISIBLE MODEL LABEL AS BACKEND**\n\nThe use of routing or dynamic models is disregarded.\n\n**CH09-F04 — CONSIDERING THE SAME PROVIDER AS COMPARABLE**\n\nThe plan, surface, and feature differences are hidden.\n\n**CH09-F05 — CONSIDERING DIFFERENT PROVIDERS AS NON-COMPARABLE**\n\nProduct comparisons possible under paired user conditions are rejected.\n\n**CH09-F06 — INTEGRATING WEB AND MOBILE**\n\nInterface and feature differences disappear.\n\n**CH09-F07 — COMBINING FREE AND PAID PLANS**\n\nQuality or feature differences due to the plan become invisible.\n\n**CH09-F08 — CONSIDERING CORPORATE AND CONSUMER PRODUCTS THE SAME**\n\nPrivate information and administrator settings are hidden.\n\n**CH09-F09 — TREATING API RESULT AS CONSUMER PRODUCT**\n\nOrchestration and user surface differences are eliminated.\n\n**CH09-F10 — IF RETRIEVAL EXISTS, CONSIDER IT USED**\n\nThe specific answer's true knowledge path cannot be verified.\n\n**CH09-F11 — IF NO CITATION, CONSIDER NO SOURCE**\n\nThe possibility of hidden or invisible retrieval is ignored.\n\n**CH09-F12 — IF CITATION EXISTS, COUNT AS CORRECT SOURCE**\n\nCitation–claim matching is not examined.\n\n**CH09-F13 — CONSIDER TOOL ACCESS AS TOOL USE**\n\nIt is unknown which tool was used in the specific answer.\n\n**CH09-F14 — CONSIDER MEMORY AS CLEAN SESSION**\n\nThe effect of personalisation remains invisible.\n\n**CH09-F15 — CONSIDER NEW CONVERSATION NON-PERSONALISED**\n\nPersistent account settings and special instructions are ignored.\n\n**CH09-F16 — PRESENTING THE CURRENT CHAT AS A NEW SESSION**\n\nPrevious context is hidden.\n\n**CH09-F17 — COUNTING AN EXPERIMENTAL FEATURE AS A DATA ERROR**\n\nActual product variation is declared an invalid observation.\n\n**CH09-F18 — HIDING AN EXPERIMENTAL FEATURE**\n\nUsers in different rollouts are combined as a single fixed product.\n\n**CH09-F19 — SINGLE WAVE AFTER UPDATE**\n\nMaterial product change is published as the result of a single fixed system.\n\n**CH09-F20 — IGNORING CHANGES IF THERE IS NO ANNOUNCEMENT**\n\nObserved behaviour break is not examined.\n\n**CH09-F21 — COUNTING LIVE PRODUCT AS A FROZEN MODEL**\n\nFull reproducibility is claimed.\n\n**CH09-F22 — CONSIDERING A FROZEN MODEL AS A REAL USER PRODUCT**\n\nProduct chain and interface differences are hidden.\n\n**CH09-F23 — COUNTING PRODUCT ABSENCE AS AN ERROR**\n\nScope of access is confused with representation accuracy.\n\n**CH09-F24 — REFUSING A TECHNICAL ERROR**\n\nConnection or service issue is classified as a policy result.\n\n**CH09-F25 — COUNTING A TECHNICAL ERROR AS REJECT AND RETRYING**\n\nActual user result is removed from the denominator.\n\n**CH09-F26 — HIDING THE USAGE LIMIT**\n\nThe real experience of the free plan becomes invisible.\n\n**CH09-F27 — CONSIDERING THE SAME PROMPT AS THE SAME EXPERIMENT**\n\nProduct, plan, session, web, and time differences are ignored.\n\n**CH09-F28 — CONSIDERING THE PRODUCT RESULT AS MODEL SUPERIORITY**\n\nThe difference across the entire system chain is attributed to the base model.\n\n**CH09-F29 — DIRECTLY RANKING NATIVE REACH SCORES**\n\nDifferent population universes are considered the same.\n\n**CH09-F30 — CONSIDERING COMMON SUPPORT AS GLOBAL SCOPE**\n\nThe narrow common universe is presented as the entire reach of the product.\n\n**CH09-F31 — CLAIMING LEADERSHIP FROM CP-1 RESULT**\n\nProduct superiority is derived solely from incompatible results that are placed side by side.\n\n**CH09-F32 — MAINTAINING THE COMPARISON LEVEL**\n\nThe public only sees the ranking.\n\n**CH09-F33 — IGNORING PAIRED USER DEPENDENCY**\n\nThe responses of the same person to products are considered independent.\n\n**CH09-F34 — IGNORING PRODUCT ORDER EFFECT**\n\nThe first response affects subsequent tasks.\n\n**CH09-F35 — TREATING DIFFERENT PANEL USERS AS THE SAME**\n\nProducts are tested with different user compositions.\n\n**CH09-F36 — CHANGING THE REALITY PACKAGE VERSION**\n\nThe same responses are compared with different reference records.\n\n**CH09-F37 — SILENTLY CHANGING THE PRODUCT FINGERPRINT**\n\nThe configuration record is updated; the old measurement is preserved rather than overwritten.\n\n**CH09-F38 — COUNTING AN UNKNOWN AREA AS CLEAN**\n\nThe model, memory, or tool status is accepted as controlled, even if unknown.\n\n**CH09-F39 — USING THE PRODUCT BRAND INSTEAD OF EVIDENCE**\n\nThe provider's reputation substitutes for measurement conditions.\n\n**CH09-F40 — COVERING ALL PRODUCT DIFFERENCES WITH A SINGLE COEFFICIENT**\n\nPlan, interface, and configuration differences are hidden by using the model or market coefficient.\n\n## 50. AUDIT PROCEDURE\n\n### Step 1 — Identify the Provider and Product\n\nThe product name, type, and official user interfaces are recorded.\n\n### Step 2 — Lock the Measurement Surface\n\nWeb Mobile Desktop\n\n### API\n\nEmbedded product is detached.\n\n### Step 3 — Verify the Plan and Account Class\n\nFree, paid, enterprise, or other plan status is recorded.\n\n### Step 4 — Assign Model ID Trust\n\nAppropriate level is determined between MI-0 and MI-5.\n\n### Step 5 — Save Retrieval and Web Configuration\n\nIs the feature available? Is it on? Was it used in a particular response? What is the evidence?\n\n### Step 6 — Save Reference and Tool Status\n\nActual usage is separated from feature access.\n\n### Step 7 — Save Session and Personalisation Status\n\nMemory Special instruction Previous chat Corporate setting is reviewed.\n\n### Step 8 — Verify Country, Locale, and Language Registration\n\nThe regional condition of the product behaviour is determined.\n\n### Step 9 — Review Experimental Features\n\nUI or feature differences between users are checked.\n\n### Step 10 — Lock Time and Wave Identity\n\nUTC time and wave record are created.\n\n### Step 11 — Generate System Configuration Fingerprint\n\nSaved fields are linked to the integrity record.\n\n### Step 12 — Monitor Change Events\n\nAnnounced and observed product changes are recorded.\n\n### Step 13 — Write the Object of Comparison\n\nWhich of the Product Model Retrieval Plan Interface Time comparisons is being made?\n\n### Step 14 — Match the Comparison Dimensions\n\nMaterial areas are tabulated.\n\n### Step 15 — Assign CP Level\n\nAn appropriate level is given between CP-0 and CP-5.\n\n### Step 16 — Separate Comparable and Non-Comparable Claims\n\nWhich results can be placed side by side is clearly written.\n\n### Step 17 — Separate the Results of Native Reach and Common Support\n\nIntra-product and common population comparisons are recorded separately.\n\n### Step 18 — Check System Change in the Wave\n\nIt is determined whether MIXED_SYSTEM_STATE or a new sub-wave is needed.\n\n### Step 19 — Create Public Comparison Card\n\nThe conditions behind the ranking are made visible.\n\n### Step 20 — Obtain Record Owner's Approval\n\nThe system registry and comparison decision are versioned by the responsible person.\n\n## 51. REQUIRED EVIDENCE\n\nProvider identity; product name and type; user surface; application or interface version; plan and account class; displayed model label; model-identity evidence; model-identity confidence level; product documentation; country-access records; language and locale; web and retrieval capability; evidence of retrieval use; citation capability; citation view; citation–claim link; tool access; tool-use log; memory setting; custom instructions; prior-chat state; session type; organisational administrator settings; experimental-feature record; A/B allocation signals; UTC timestamp; measurement wave; system-configuration fingerprint; provider-update records; observed behaviour changes; product-access errors; technical-retry logs; Native Reach population record; Common Support population record; and comparison-dimension table.\n\nCP level Prompt version Reality package version Sample and weight record Adjudication version Public comparison card Change log Responsible person or institution\n\n## 52. AUDIT CHECKLIST\n\nWere the provider and product recorded separately? Is the user surface identifiable? Were the plan and account class verified? Is the model label merely the displayed label, or does it identify a verified deployment? Was a model-identity confidence level assigned? Is the result attributed to the product or to the model? Was web or retrieval available? Was its use in the specific response verified? Were citation capability and citation display separated? Do the citations actually support the claims? Were tool access and tool use separated? Was memory enabled? Were custom instructions present? Was there prior conversational context? Was the session type recorded? Was an enterprise or private information source present? Were experimental product features checked? Could country and locale affect product behaviour?\n\nHas the system fingerprint been created? Did the product change during the measurement wave? Is there any behaviour break outside the provider announcement? Were pre-change and post-change observations separated? Were technical errors distinguished from rejections? Was a retry performed due to an incorrect answer? Is the object of the comparison clear? Which conditions match besides the same prompt? Is the target population the same? Are the prompt and Truth Pack versions the same? Are the differences in plan, surface, and retrieval visible? Was the CP level provided? Was a ranking done from CP-0 or CP-1 results? Were the results of Native Reach and Common Support separated? Was the same user product sequence balanced? Were different user panels calibrated?\n\nWas the live product result presented as a fundamental model superiority? Were unknown areas considered controlled? Is there a public comparison card available? Is the accountable owner of the system registry known?\n\n## 53. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “If the user does not know which model is running, can this work still be done?”\n\nIt can be done. The main GEO-1000 object is the user's product. If the model is unknown, the result is reported as: “This behaviour was observed on this product and user surface.” No reference is made to a specific model.\n\n### Objection 2 — “Why do we differentiate the products of the same provider so much?”\n\nBecause the user:\n\n- different plan,\n\n- model,\n\n- retrieval,\n\n- memory,\n\n- tool\n\nIt may see different answers under different conditions. The provider name does not fully define the user experience.\n\n### Objection 3 — “Isn't comparing different providers like comparing apples and oranges anyway?”\n\nIf the comparison question is set up correctly, it is not. A real user might ask: “Which product informs me more correctly?” The entire product chain is part of this question. If the superiority of the base model is to be claimed, a more controlled experiment is needed.\n\n### Objection 4 — “Isn't it natural for a web-using product to have an advantage?”\n\nIt is natural. In a natural product comparison, the advantage is the product's real feature. In model comparison, the retrieval effect should also be separated.\n\n### Objection 5 — “Isn't it unfair not to give a citation score to a product that doesn't show citations?”\n\nThe attribution feature is the function that the product offers to the user. If the feature is missing, the attribution result can be displayed as:\n\n### RECORDED AS NOT AVAILABLE\n\nMaterial accuracy is also measured separately.\n\n### Objection 6 — “If memory is part of the natural user experience, why do we turn it off?”\n\nIt can remain on in the natural panel. It can be turned off in the controlled panel to clear comparisons between products. The two panels answer different questions.\n\n### Objection 7 — “Isn't opening a new chat enough for a clean session?”\n\nNot always. Persistent memory, special instructions, or account settings may continue. These areas should also be saved.\n\n### Objection 8 — “What can we do if the provider did not explain the product change?”\n\nObserved:\n\n- interface,\n\n- feature,\n\n- behaviour,\n\n- Attribution\n\nWe can save your changes. If the cause is unknown, indefinite change status such as UC-5 is used.\n\n### Objection 9 — \"If products are constantly changing, won't any results be valid?\"\n\nThe result stated:\n\n- date,\n\n- wave,\n\n- Product Sample\n\nIt is valid for. Variability does not make measurement meaningless. It prevents untimely judgement.\n\n### Objection 10 — \"Isn't the same prompt and the same day comparable enough?\"\n\nThe plan, surface, memory, web, country, and user population may differ. The same prompt may be required. It is not enough on its own.\n\n### Objection 11 — 'Wouldn't the CP level be too technical for the public?'\n\nA short label can be used:\n\n- Descriptive\n\n- Partially matched\n\n- Matched\n\n- Controlled\n\n- Replicated\n\nOne should know how strong the public ranking is.\n\n### Objection 12 — “Why can't we directly rank based on Native Reach scores?”\n\nBecause each product may have been measured in different countries and user populations. The score difference may partly arise from which population the product reached. Common Support provides a fairer comparison.\n\n### Objection 13 — \"Isn't Common Support a real-world result?\"\n\nIt is the result of the actual shared population. However, it does not show access of the products outside the shared universe. It should be read together with Native Reach.\n\n### Objection 14 — “Is live product inspection more important, or frozen model inspection?”\n\nThe questions are different. Live product measures the truth of the user. Frozen model provides scientific control and reproducibility. NOMOS does not use them interchangeably.\n\n## COMMON RULE OF CHAPTER 57\n\nA user:\n\n- “ChatGPT said so.”\n\n- “Gemini thought so.”\n\n- “Claude recommended us.”\n\n- “Some cited us.”\n\ncan be said. These expressions are understandable in daily conversation. They are not sufficient in an audit record. Because under these names:\n\n- different products,\n\n- different plans,\n\n- different models,\n\n- different user interfaces,\n\n- different web and retrieval systems,\n\n- different memory conditions,\n\n- different countries,\n\n- different times\n\ncan be found. Different people can see different AI experiences under the same name. Comparable user experiences can be established under different names. A model label seems to have produced the entire answer. But the reality reaching the user:\n\n- source selection,\n\n- security,\n\n- personalisation,\n\n- interface,\n\n- tools\n\nThe result may be shaped by security controls, personalisation, the interface and available tools. NOMOS therefore asks more than ‘Which model?’ It asks: Which product? / Which surface? / Which plan? / Which information mode? / Which session? / Which user? / Which country? / Which time? / Which version? Two results may come from the same prompt without coming from the same experiment. Two results may come from different providers yet remain fairly comparable under the same user conditions. A product is a live system: its sources, interface and model can change. Change does not make a result worthless; it binds the result to a particular time and product state. NOMOS's ninth law of measurement is therefore:\n\n> An AI provider is not an AI product.\n\nThe tenth law is:\n\n> The model is only one part of the representation chain that reaches the user.\n\nThe eleventh law is as follows:\n\n> The same name guarantees comparability; a different name does not guarantee non-comparability.\n\nThe twelfth law is as follows:\n\n> The same prompt is not the same experiment.\n\nThe thirteenth law is as follows:\n\n> The outcome of the living product is valid only for the product sample and time at which it is measured.\n\nThe fourteenth law is as follows:\n\n> The power of comparison comes not from the magnitude of ranking, but from the clarity of the matched conditions.\n\n## Order of Section 9 of NOMOS\n\n> Do not give me only the provider's name. / Show which product was used, on which surface and with which plan.\n\n> Do not make the model label you see on the screen the exact identity of the entire back-end system.\n\n> It was used because there was a web feature; do not assume it supported the claim because there was a citation.\n\n> Do not count tool access as tool use, a new chat as a clean account, or unknown memory status as memory off.\n\n> Do not put a free user and a paid user, a mobile product and a web product, or an enterprise workspace and general chat in the same cell.\n\n> Do not automatically consider results from the same provider as a fair comparison.\n\n> Do not avoid comparing different providers just because they are different. / First, match the conditions. / Then, explain which question you answered.\n\n> Do not present a live product as a frozen model, an API as a consumer experience, or a product result as base model superiority.\n\n> If the system changed in the middle of the wave, do not merge the old and new responses as a single product.\n\n> Do not ignore the break in behaviour in front of you just because the provider did not explain the change.\n\n> Do not assume the unknown is clean, equal, or fixed.\n\nFirst, register the system in the record. / Then lock the configuration. / Then monitor the changes. / Then write the object of comparison. / Then show matching and non-matching conditions. / And only then say which product gave a stronger result under which condition.\n\n## The Chapter's Closing Sentence\n\n> The “AI” measured in GEO-1000 is not a single name or model; it is a versioned representation system reaching a specific user at a specific moment within a specific product chain.\n\n## Normative Core\n\n> Every GEO-1000 observation MUST be linked to a versioned AI System Register record describing the measured product instance, including as applicable: - provider, - product, - user surface, - plan, - displayed or verified model identity, - model-identity confidence, - retrieval and web configuration, - citation capability and use, - tool access and use, - session state, - memory, - custom instructions, - prior context, - personalisation, - country, - locale, - language, - deployment state, - measurement time, - and known or observed system changes. Provider name, product name, displayed model label, base model, retrieval system, and user-facing product MUST NOT be treated as interchangeable identities. Feature availability MUST remain distinct from feature use. Live-product results MUST NOT be represented as frozen-model results, and frozen-model results MUST NOT be represented as full consumer-product experience. Comparisons MUST define their object and receive a disclosed comparability level based on matched entity, prompt, population, language, country, time, session, plan, interface, retrieval, tools, reference records, and adjudication rules. The same provider or displayed model MUST NOT create an automatic assumption of comparability. Different providers MAY be compared when user-facing conditions are sufficiently aligned and the resulting claim is limited to the measured product behaviour. Material product changes during a measurement wave MUST create a separate system state, sub-wave, modelled change factor, or mixed-state warning. Unknown configuration fields MUST remain UNKNOWN and MUST NOT be imputed as controlled or equivalent. Every system-registry and comparability decision MUST be versioned and attributable to an accountable human or organisation.","character_count":66607,"record_sha256":"e9b06581d42640999e5b577733e84b06e31b5aeb4e2781fd1518d5fef29d1924"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-10","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":12,"chapter_number":10,"item_number":null,"title":"The Prompt Constitution","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K04","K11","K12","K13"],"source_word_count":9743,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":10,\"id\":\"NOMOS-GEO-AUDIT-CH10\",\"title\":\"The Prompt Constitution\",\"subtitle\":null,\"sourceFile\":\"10.cu bölüm.docx\",\"sourceSha256\":\"89A870F6D63CB5F2A8C8F13FDAEA1C0CDB0B6933D295FF1DF466DF85A54D5FE3\",\"sourceWordCount\":9743,\"sourceIds\":[\"K04\",\"K11\",\"K12\",\"K13\"],\"machine\":{\"chapter\":10,\"chapterId\":\"NOMOS-GEO-AUDIT-CH10\",\"title\":\"The Prompt Constitution\",\"subtitle\":null,\"sourceIds\":[\"K04\",\"K11\",\"K12\",\"K13\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH10-R01\",\"normativeRuleEnglish\":\"Every GEO-1000 prompt MUST be linked to a versioned Prompt Registry record defining: - the research question, - estimand, - prompt family, - canonical intent, - entity anchor, - user intent, - speech act, - scope, - geographic and temporal frame, - presuppositions, - evidence and source requirements, - response format, - language, - locale, - equivalence status, - version, - and integrity record. Within the same language-locale prompt cell, participants MUST receive the same locked prompt text. Across languages and locales, prompts MUST preserve semantic, functional, and normative equivalence rather than literal word-for-word identity. Entity anchor, user intent, speech act, scope, presupposition, and epistemic burden are non-compensatory equivalence gates. Machine translation or back-translation alone MUST NOT establish prompt equivalence. Core identity, evidence, recommendation, comparison, temporal, local, boundary, robustness, control, and holdout prompts MUST remain distinct prompt families. Leading, answer-containing, positively or negatively presuppositional, source-constrained, format-constrained, or recommendation prompts MUST NOT be silently scored as neutral Core Mirror prompts. Prompt sets, language versions, assignment rules, clarification paths, ordering, and holdout commitments MUST be locked before AI responses are observed. Post-result prompt selection, prompt-family deletion, silent mid-wave editing, and outcome-driven rewording are prohibited. the prompt assigned to the participant and the prompt actually sent to the AI product MUST be integrity-checked. Hidden instructions, invisible characters, or undisclosed manipulation directing the AI toward a preferred entity, claim, citation, or recommendation are prohibited. Every prompt change, translation decision, equivalence judgement, and registry record MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Her GEO-1000 promptu araştırma sorusunu, tahmin edilen niceliği, prompt ailesini, kanonik niyeti, entity anchor’ı, kullanıcı niyetini, görev türünü, kapsamı, coğrafi ve zamansal çerçeveyi, ön varsayımları, kaynak ve kanıt talebini, cevap biçimini, dil ve locale’i, eşdeğerlik statüsünü, sürümü ve bütünlük kaydını tanımlayan sürümlü Prompt Sicili’ne bağlanmalıdır. Aynı dil–locale hücresinde kullanıcılar aynı kilitlenmiş metni almalı; diller arasında kelime kelime eşitlik değil semantik, işlevsel ve normatif eşdeğerlik korunmalıdır. Prompt setleri, çeviriler, atama kuralları, clarification yolları ve holdout kayıtları AI cevapları görülmeden önce kilitlenmelidir.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH10-MB0001\",\"type\":\"paragraph\",\"text\":\"75. MACHINE-READABLE PROMPT RECORD\",\"sourceParagraph\":2046},{\"blockId\":\"CH10-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2047},{\"blockId\":\"CH10-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2048},{\"blockId\":\"CH10-MB0004\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2049},{\"blockId\":\"CH10-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"excludedTasks\\\": [\",\"sourceParagraph\":2050},{\"blockId\":\"CH10-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"RECOMMENDATION\\\",\",\"sourceParagraph\":2051},{\"blockId\":\"CH10-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"LEADERSHIP\\\",\",\"sourceParagraph\":2052},{\"blockId\":\"CH10-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"TRUSTWORTHINESS\\\",\",\"sourceParagraph\":2053},{\"blockId\":\"CH10-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"PRICE\\\",\",\"sourceParagraph\":2054},{\"blockId\":\"CH10-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"COMPARATIVE_RANKING\\\"\",\"sourceParagraph\":2055},{\"blockId\":\"CH10-MB0011\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2056},{\"blockId\":\"CH10-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"sourceRequirement\\\": \\\"NONE\\\",\",\"sourceParagraph\":2057},{\"blockId\":\"CH10-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"responseFormat\\\": \\\"NATURAL_OPEN_RESPONSE\\\",\",\"sourceParagraph\":2058},{\"blockId\":\"CH10-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"presuppositionPolicy\\\": \\\"NO_UNVERIFIED_QUALITY_OR_AUTHORITY_ASSUMPTION\\\"\",\"sourceParagraph\":2059},{\"blockId\":\"CH10-MB0015\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2060},{\"blockId\":\"CH10-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"sourceLanguage\\\": {\",\"sourceParagraph\":2061},{\"blockId\":\"CH10-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": \\\"tr\\\",\",\"sourceParagraph\":2062},{\"blockId\":\"CH10-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"tr-TR\\\",\",\"sourceParagraph\":2063},{\"blockId\":\"CH10-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"Which parent organisation is Apple.com associated with, and what are the main activities of this organisation?\\\"\",\"sourceParagraph\":2064},{\"blockId\":\"CH10-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"unicodeNormalization\\\": \\\"NFC\\\",\",\"sourceParagraph\":2065},{\"blockId\":\"CH10-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-HASH-TR-001\\\"\",\"sourceParagraph\":2066},{\"blockId\":\"CH10-MB0022\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2067},{\"blockId\":\"CH10-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"languageVersions\\\": [\",\"sourceParagraph\":2068},{\"blockId\":\"CH10-MB0024\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2069},{\"blockId\":\"CH10-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": \\\"tr\\\",\",\"sourceParagraph\":2070},{\"blockId\":\"CH10-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"tr-TR\\\",\",\"sourceParagraph\":2071},{\"blockId\":\"CH10-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"Which parent organisation is Apple.com associated with, and what are the main activities of this organisation?\\\"\",\"sourceParagraph\":2072},{\"blockId\":\"CH10-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"equivalenceLevel\\\": \\\"PE-4\\\",\",\"sourceParagraph\":2073},{\"blockId\":\"CH10-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"entityAnchorEquivalence\\\": 2,\",\"sourceParagraph\":2074},{\"blockId\":\"CH10-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"intentEquivalence\\\": 2,\",\"sourceParagraph\":2075},{\"blockId\":\"CH10-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"speechActEquivalence\\\": 2,\",\"sourceParagraph\":2076},{\"blockId\":\"CH10-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"scopeEquivalence\\\": 2,\",\"sourceParagraph\":2077},{\"blockId\":\"CH10-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"presuppositionEquivalence\\\": 2,\",\"sourceParagraph\":2078},{\"blockId\":\"CH10-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"epistemicBurdenEquivalence\\\": 2,\",\"sourceParagraph\":2079},{\"blockId\":\"CH10-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"localPilotStatus\\\": \\\"PASSED\\\",\",\"sourceParagraph\":2080},{\"blockId\":\"CH10-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-HASH-TR-001\\\"\",\"sourceParagraph\":2081},{\"blockId\":\"CH10-MB0037\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2082},{\"blockId\":\"CH10-MB0038\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2083},{\"blockId\":\"CH10-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": \\\"en\\\",\",\"sourceParagraph\":2084},{\"blockId\":\"CH10-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"en-GB\\\",\",\"sourceParagraph\":2085},{\"blockId\":\"CH10-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"text\\\": \\\"Which principal organisation is associated with apple.com, and what are its primary activities?\\\",\",\"sourceParagraph\":2086},{\"blockId\":\"CH10-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"equivalenceLevel\\\": \\\"PE-4\\\",\",\"sourceParagraph\":2087},{\"blockId\":\"CH10-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"entityAnchorEquivalence\\\": 2,\",\"sourceParagraph\":2088},{\"blockId\":\"CH10-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"intentEquivalence\\\": 2,\",\"sourceParagraph\":2089},{\"blockId\":\"CH10-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"speechActEquivalence\\\": 2,\",\"sourceParagraph\":2090},{\"blockId\":\"CH10-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"scopeEquivalence\\\": 2,\",\"sourceParagraph\":2091},{\"blockId\":\"CH10-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"presuppositionEquivalence\\\": 2,\",\"sourceParagraph\":2092},{\"blockId\":\"CH10-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"epistemicBurdenEquivalence\\\": 2,\",\"sourceParagraph\":2093},{\"blockId\":\"CH10-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"localPilotStatus\\\": \\\"PASSED\\\",\",\"sourceParagraph\":2094},{\"blockId\":\"CH10-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-HASH-EN-001\\\"\",\"sourceParagraph\":2095},{\"blockId\":\"CH10-MB0051\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2096},{\"blockId\":\"CH10-MB0052\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2097},{\"blockId\":\"CH10-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"delivery\\\": {\",\"sourceParagraph\":2098},{\"blockId\":\"CH10-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"manualRetypingAllowed\\\": false,\",\"sourceParagraph\":2099},{\"blockId\":\"CH10-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"copyFromLockedInterface\\\": true,\",\"sourceParagraph\":2100},{\"blockId\":\"CH10-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"exactMatchRequired\\\": true,\",\"sourceParagraph\":2101},{\"blockId\":\"CH10-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"participantEditingAllowed\\\": false,\",\"sourceParagraph\":2102},{\"blockId\":\"CH10-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"clarificationPath\\\": \\\"SINGLE_TURN_CAPTURE_ONLY\\\",\",\"sourceParagraph\":2103},{\"blockId\":\"CH10-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"newConversationRequired\\\": true\",\"sourceParagraph\":2104},{\"blockId\":\"CH10-MB0060\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2105},{\"blockId\":\"CH10-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"sets\\\": {\",\"sourceParagraph\":2106},{\"blockId\":\"CH10-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"publicNormativeSet\\\": true,\",\"sourceParagraph\":2107},{\"blockId\":\"CH10-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"sealedValidationSet\\\": false,\",\"sourceParagraph\":2108},{\"blockId\":\"CH10-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"holdoutStatus\\\": \\\"NOT_APPLICABLE\\\"\",\"sourceParagraph\":2109},{\"blockId\":\"CH10-MB0065\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2110},{\"blockId\":\"CH10-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":2111},{\"blockId\":\"CH10-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"draftedByRole\\\": \\\"PROMPT_METHODS_LEAD\\\",\",\"sourceParagraph\":2112},{\"blockId\":\"CH10-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"translatedByRole\\\": \\\"TARGET_LANGUAGE_TRANSLATOR\\\",\",\"sourceParagraph\":2113},{\"blockId\":\"CH10-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"backTranslatedByIndependentRole\\\": true,\",\"sourceParagraph\":2114},{\"blockId\":\"CH10-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"equivalenceApprovedByRole\\\": \\\"PROMPT_EQUIVALENCE_ADJUDICATOR\\\",\",\"sourceParagraph\":2115},{\"blockId\":\"CH10-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"lockedAt\\\": \\\"2026-08-18T09:00:00Z\\\",\",\"sourceParagraph\":2116},{\"blockId\":\"CH10-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"resultInspectionBeforeLock\\\": false\",\"sourceParagraph\":2117},{\"blockId\":\"CH10-MB0073\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2118},{\"blockId\":\"CH10-MB0074\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2119},{\"blockId\":\"CH10-MB0075\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2120},{\"blockId\":\"CH10-MB0076\",\"type\":\"paragraph\",\"text\":\"This record:\",\"sourceParagraph\":2121},{\"blockId\":\"CH10-MB0077\",\"type\":\"paragraph\",\"text\":\"This is not a real Apple audit,\",\"sourceParagraph\":2122},{\"blockId\":\"CH10-MB0078\",\"type\":\"paragraph\",\"text\":\"This is not a real prompt validation work,\",\"sourceParagraph\":2123},{\"blockId\":\"CH10-MB0079\",\"type\":\"paragraph\",\"text\":\"It is a machine-readable synthetic representation of the Prompt Constitution.\",\"sourceParagraph\":2124},{\"blockId\":\"CH10-MB0080\",\"type\":\"paragraph\",\"text\":\"76. MACHINE-READABLE PROMPT OBSERVATION RECORD\",\"sourceParagraph\":2126},{\"blockId\":\"CH10-MB0081\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2127},{\"blockId\":\"CH10-MB0082\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2128},{\"blockId\":\"CH10-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2129},{\"blockId\":\"CH10-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2130},{\"blockId\":\"CH10-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"promptObservation\\\": {\",\"sourceParagraph\":2131},{\"blockId\":\"CH10-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"observationId\\\": \\\"NGO-SYNTH-000184\\\",\",\"sourceParagraph\":2132},{\"blockId\":\"CH10-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"promptId\\\": \\\"NGP-CORE-MIRROR-APPLE-SYNTH-001\\\",\",\"sourceParagraph\":2133},{\"blockId\":\"CH10-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"promptVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":2134},{\"blockId\":\"CH10-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"language\\\": \\\"tr\\\",\",\"sourceParagraph\":2135},{\"blockId\":\"CH10-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"tr-TR\\\",\",\"sourceParagraph\":2136},{\"blockId\":\"CH10-MB0091\",\"type\":\"paragraph\",\"text\":\"\\\"assignedTextHash\\\": \\\"SYNTHETIC-HASH-TR-001\\\",\",\"sourceParagraph\":2137},{\"blockId\":\"CH10-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"sentTextHash\\\": \\\"SYNTHETIC-HASH-TR-001\\\",\",\"sourceParagraph\":2138},{\"blockId\":\"CH10-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"promptValidity\\\": \\\"PV-1\\\",\",\"sourceParagraph\":2139},{\"blockId\":\"CH10-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"promptOrder\\\": 1,\",\"sourceParagraph\":2140},{\"blockId\":\"CH10-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"conversationTurn\\\": 1,\",\"sourceParagraph\":2141},{\"blockId\":\"CH10-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"clarificationUsed\\\": false,\",\"sourceParagraph\":2142},{\"blockId\":\"CH10-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"participantEdited\\\": false,\",\"sourceParagraph\":2143},{\"blockId\":\"CH10-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"systemTransformationObserved\\\": false,\",\"sourceParagraph\":2144},{\"blockId\":\"CH10-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"fullPromptVisibleInEvidence\\\": true,\",\"sourceParagraph\":2145},{\"blockId\":\"CH10-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"timestampUtc\\\": \\\"2026-08-18T12:14:00Z\\\",\",\"sourceParagraph\":2146},{\"blockId\":\"CH10-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2147},{\"blockId\":\"CH10-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"promptValidityReviewerRole\\\": \\\"BLINDED_PROMPT_VALIDATOR\\\"\",\"sourceParagraph\":2148},{\"blockId\":\"CH10-MB0103\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2149},{\"blockId\":\"CH10-MB0104\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2150},{\"blockId\":\"CH10-MB0105\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2151},{\"blockId\":\"CH10-MB0106\",\"type\":\"paragraph\",\"text\":\"77. MACHINE-READABLE RULE OF THE SECTION\",\"sourceParagraph\":2153},{\"blockId\":\"CH10-MB0107\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH10-R01\",\"sourceParagraph\":2154},{\"blockId\":\"CH10-MB0108\",\"type\":\"paragraph\",\"text\":\"Every GEO-1000 prompt MUST be linked to a versioned Prompt Registry record\",\"sourceParagraph\":2156},{\"blockId\":\"CH10-MB0109\",\"type\":\"paragraph\",\"text\":\"defining:\",\"sourceParagraph\":2157},{\"blockId\":\"CH10-MB0110\",\"type\":\"paragraph\",\"text\":\"- the research question,\",\"sourceParagraph\":2159},{\"blockId\":\"CH10-MB0111\",\"type\":\"paragraph\",\"text\":\"- estimand,\",\"sourceParagraph\":2160},{\"blockId\":\"CH10-MB0112\",\"type\":\"paragraph\",\"text\":\"- prompt family,\",\"sourceParagraph\":2161},{\"blockId\":\"CH10-MB0113\",\"type\":\"paragraph\",\"text\":\"- canonical intent,\",\"sourceParagraph\":2162},{\"blockId\":\"CH10-MB0114\",\"type\":\"paragraph\",\"text\":\"- entity anchor,\",\"sourceParagraph\":2163},{\"blockId\":\"CH10-MB0115\",\"type\":\"paragraph\",\"text\":\"- user intent,\",\"sourceParagraph\":2164},{\"blockId\":\"CH10-MB0116\",\"type\":\"paragraph\",\"text\":\"- speech act,\",\"sourceParagraph\":2165},{\"blockId\":\"CH10-MB0117\",\"type\":\"paragraph\",\"text\":\"- scope,\",\"sourceParagraph\":2166},{\"blockId\":\"CH10-MB0118\",\"type\":\"paragraph\",\"text\":\"- geographic and temporal frame,\",\"sourceParagraph\":2167},{\"blockId\":\"CH10-MB0119\",\"type\":\"paragraph\",\"text\":\"- presuppositions,\",\"sourceParagraph\":2168},{\"blockId\":\"CH10-MB0120\",\"type\":\"paragraph\",\"text\":\"- evidence and source requirements,\",\"sourceParagraph\":2169},{\"blockId\":\"CH10-MB0121\",\"type\":\"paragraph\",\"text\":\"- response format,\",\"sourceParagraph\":2170},{\"blockId\":\"CH10-MB0122\",\"type\":\"paragraph\",\"text\":\"- language,\",\"sourceParagraph\":2171},{\"blockId\":\"CH10-MB0123\",\"type\":\"paragraph\",\"text\":\"- locale,\",\"sourceParagraph\":2172},{\"blockId\":\"CH10-MB0124\",\"type\":\"paragraph\",\"text\":\"- equivalence status,\",\"sourceParagraph\":2173},{\"blockId\":\"CH10-MB0125\",\"type\":\"paragraph\",\"text\":\"- version,\",\"sourceParagraph\":2174},{\"blockId\":\"CH10-MB0126\",\"type\":\"paragraph\",\"text\":\"- and integrity record.\",\"sourceParagraph\":2175},{\"blockId\":\"CH10-MB0127\",\"type\":\"paragraph\",\"text\":\"Within the same language-locale prompt cell, participants MUST receive\",\"sourceParagraph\":2177},{\"blockId\":\"CH10-MB0128\",\"type\":\"paragraph\",\"text\":\"the same locked prompt text.\",\"sourceParagraph\":2178},{\"blockId\":\"CH10-MB0129\",\"type\":\"paragraph\",\"text\":\"Across languages and locales, prompts MUST preserve semantic, functional,\",\"sourceParagraph\":2180},{\"blockId\":\"CH10-MB0130\",\"type\":\"paragraph\",\"text\":\"and normative equivalence rather than literal word-for-word identity.\",\"sourceParagraph\":2181},{\"blockId\":\"CH10-MB0131\",\"type\":\"paragraph\",\"text\":\"Entity anchor, user intent, speech act, scope, presupposition, and\",\"sourceParagraph\":2183},{\"blockId\":\"CH10-MB0132\",\"type\":\"paragraph\",\"text\":\"epistemic burden are non-compensatory equivalence gates.\",\"sourceParagraph\":2184},{\"blockId\":\"CH10-MB0133\",\"type\":\"paragraph\",\"text\":\"Machine translation or back-translation alone MUST NOT establish prompt\",\"sourceParagraph\":2186},{\"blockId\":\"CH10-MB0134\",\"type\":\"paragraph\",\"text\":\"equivalence.\",\"sourceParagraph\":2187},{\"blockId\":\"CH10-MB0135\",\"type\":\"paragraph\",\"text\":\"Core identity, evidence, recommendation, comparison, temporal, local,\",\"sourceParagraph\":2189},{\"blockId\":\"CH10-MB0136\",\"type\":\"paragraph\",\"text\":\"boundary, robustness, control, and holdout prompts MUST remain distinct\",\"sourceParagraph\":2190},{\"blockId\":\"CH10-MB0137\",\"type\":\"paragraph\",\"text\":\"prompt families.\",\"sourceParagraph\":2191},{\"blockId\":\"CH10-MB0138\",\"type\":\"paragraph\",\"text\":\"Leading, answer-containing, positively or negatively presuppositional,\",\"sourceParagraph\":2193},{\"blockId\":\"CH10-MB0139\",\"type\":\"paragraph\",\"text\":\"source-constrained, format-constrained, or recommendation prompts MUST NOT\",\"sourceParagraph\":2194},{\"blockId\":\"CH10-MB0140\",\"type\":\"paragraph\",\"text\":\"be silently scored as neutral Core Mirror prompts.\",\"sourceParagraph\":2195},{\"blockId\":\"CH10-MB0141\",\"type\":\"paragraph\",\"text\":\"Prompt sets, language versions, assignment rules, clarification paths,\",\"sourceParagraph\":2197},{\"blockId\":\"CH10-MB0142\",\"type\":\"paragraph\",\"text\":\"ordering, and holdout commitments MUST be locked before AI responses are\",\"sourceParagraph\":2198},{\"blockId\":\"CH10-MB0143\",\"type\":\"paragraph\",\"text\":\"observed.\",\"sourceParagraph\":2199},{\"blockId\":\"CH10-MB0144\",\"type\":\"paragraph\",\"text\":\"Post-result prompt selection, prompt-family deletion, silent mid-wave\",\"sourceParagraph\":2201},{\"blockId\":\"CH10-MB0145\",\"type\":\"paragraph\",\"text\":\"editing, and outcome-driven rewording are prohibited.\",\"sourceParagraph\":2202},{\"blockId\":\"CH10-MB0146\",\"type\":\"paragraph\",\"text\":\"The prompt assigned to the participant and the prompt actually sent to\",\"sourceParagraph\":2204},{\"blockId\":\"CH10-MB0147\",\"type\":\"paragraph\",\"text\":\"the AI product MUST be integrity-checked.\",\"sourceParagraph\":2205},{\"blockId\":\"CH10-MB0148\",\"type\":\"paragraph\",\"text\":\"Hidden instructions, invisible characters, or undisclosed manipulation\",\"sourceParagraph\":2207},{\"blockId\":\"CH10-MB0149\",\"type\":\"paragraph\",\"text\":\"directing the AI toward a preferred entity, claim, citation, or\",\"sourceParagraph\":2208},{\"blockId\":\"CH10-MB0150\",\"type\":\"paragraph\",\"text\":\"recommendation are prohibited.\",\"sourceParagraph\":2209},{\"blockId\":\"CH10-MB0151\",\"type\":\"paragraph\",\"text\":\"Every prompt change, translation decision, equivalence judgement, and\",\"sourceParagraph\":2211},{\"blockId\":\"CH10-MB0152\",\"type\":\"paragraph\",\"text\":\"registry record MUST be versioned and attributable to an accountable\",\"sourceParagraph\":2212},{\"blockId\":\"CH10-MB0153\",\"type\":\"paragraph\",\"text\":\"human or organisation.\",\"sourceParagraph\":2213},{\"blockId\":\"CH10-MB0154\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2214},{\"blockId\":\"CH10-MB0155\",\"type\":\"paragraph\",\"text\":\"Every GEO-1000 prompt must be linked to the versioned Prompt Registry that defines the research question, estimated quantity, prompt family, canonical intent, entity anchor, user intent, task type, scope, geographic and temporal frame, assumptions, source and evidence request, response format, language and locale, equivalence status, version, and integrity record. Users in the same language–locale cell must receive the same locked text; across languages, semantic, functional, and normative equivalence must be preserved, not word-for-word equality. Prompt sets, translations, assignment rules, clarification paths, and holdout records must be locked before AI responses are seen.\",\"sourceParagraph\":2215}]}}","text":"## Chapter Boundary\n\nSection 9 established the following provision:\n\n> The same provider name, the same model label, or the same logo does not create a comparable AI experience.\n\nNow we need to identify the other main tool:\n\n> What exactly will the user ask the AI product?\n\nA prompt is not just a question sentence. A prompt:\n\n- calls the audited entity,\n\n- defines the user's intention,\n\n- determines which task it expects from the AI,\n\n- expands or narrows the scope of the response,\n\n- may encourage the use of resources,\n\n- may request advice,\n\n- can embed an assumption into the answer,\n\nIt can direct the user or the system to a specific outcome. These prompts are not the same:\n\n- “What is Apple.com?”\n\n- “With which organisation is Apple.com associated and what does this organisation do?”\n\n- “Is Apple.com a reliable company?”\n\n- “Why is Apple.com the best technology company in the world?”\n\n- “According to its official site, what kind of company is Apple?”\n\n- “Compare Apple with similar companies and tell me whether you recommend it or not.”\n\nThey all contain the same domain name. They do not carry the same user intent. They do not generate the same evidence load. They do not expect the same answer. The first prompt is ambiguous. The second prompt requires identity and activity analysis. The third prompt requires evaluation and a judgement of reliability. The fourth prompt embeds a positive and superiority assumption within the response. The fifth prompt links the response only to the official source. The sixth prompt creates a comparison and recommendation task. The answers to these questions cannot be merged without explanation into a single GEO score. In multilingual measurement, the problem becomes even greater. The phrase 'same prompt' can have two different meanings:\n\n- Giving all users the same character string\n\n- Preserving the same user intent, task load, and epistemic boundary across all languages and locales\n\nThe first method can be applied within a single language. It is often meaningless between languages. Word-for-word translation:\n\n- may not be natural,\n\n- may carry a different social meaning,\n\n- may change the official or informal tone,\n\n- may add assumptions of trust, quality, or recommendation,\n\n- may broaden the legal scope,\n\nmay break the entity anchor. Therefore, in GEO-1000, “same prompt” means:\n\n> The same locked prompt text within the same language and locale cell; between different languages and locales, it means verified semantic and functional equivalence.\n\nNOMOS requires that each measurement carries the query set, language, country, date, AI product, and repetition record; that the evidence, boundary, context, and time are visible. the prompt Constitution applies this obligation to the user's question itself. This section:\n\n- the identity of the prompt as an experimental tool,\n\n- the distinction between the research question and the prompt text,\n\n- The prompt family, prompt version, and prompt instance,\n\n- diagnostic prompts with the Core Mirror Prompt,\n\n- controlled and natural prompt traces,\n\n- multilingual semantic equivalence,\n\n- translation, back-translation, and local pilot,\n\n- prompt steering,\n\n- assumption and epistemic load issues,\n\n- entity anchor and scope boundary,\n\n- prompt order and conversational context,\n\n- prompt leakage and overfitting to evaluation,\n\n- open, sealed, and holdout prompt sets,\n\n- prompt versioning, locking, and integrity logs,\n\n- the exact delivery of the prompt to the user,\n\ndefines. This chapter does not yet:\n\n- all field details of controlled clean and natural user panels,\n\n- the full architecture of NOMOS Capture software,\n\n- the separation of responses into atomic claims,\n\n- final answer transition threshold,\n\n- the exact weights of the system families in the NOMOS score\n\ndoes not finalise. The main question of Chapter 10 is:\n\n> How do we prove that the questions given to AI products actually measure the same thing across products, users, countries, languages, and times?\n\n## NOMOS Challenge\n\nYou are giving me four questions about the same company:\n\n- What kind of company is X?\n\n- Is X reliable?\n\n- Why is X the best company?\n\n- “Would you recommend X to me?”\n\nThen you put my answers to the same GEO score. In the first question, I explain the identity and activity. In the second question, I look for a trust criterion. In the third question, I am forced to respond to the superiority assumption you placed. In the fourth question, I generate user suitability and comparative recommendations. You measure four separate tasks as if they were a single result. Then:\n\n“You asked about the same brand,” you say. The same entity is not the same prompt. Now you write the same question in English: “What kind of company is X?” You are forming a sentence close in meaning in another language: “Is X a good and trustworthy company?” Translations may appear close in terms of words.\n\nBut the first one requires definition. The second one requires evaluation. In the first answer, I can explain what the company does. In the second answer, I can look for evidence of trust and recommendation. Then you count the score difference between languages as my language performance. In fact, you did not perform the same experiment. Now you are giving another prompt:\n\n“Explain X using the company's own site.” I only convey the company's official statement. For another product, you say: “Evaluate X using independent sources.” Then you compare the two products. You gave one the role of official representation, the other the role of verified reality. The resulting difference is not only a product difference.\n\nIt is a difference in the User Constitution. Now you are distributing the prompt to the whole world. The English version is natural. The Turkish version feels like machine translation. The Japanese version is excessively formal. In another language, the company name has been converted to the wrong script system and confused with another entity. You are creating a different cognitive and ontological task in each language.\n\nThen you publish the global language fairness report. The first decree of this section is as follows:\n\n> The prompt is not a neutral container of measurement; it is the experimental tool that constitutes the measurement result.\n\nIts second provision states:\n\n> The same entity name is not the same user intent.\n\nIts third provision states:\n\n> The same word should not be preserved across languages, but the same semantic and normative task should be.\n\nIts fourth provision states:\n\n> If the prompt already contains the answer, it does not measure GEO; it measures the prompting.\n\nIts fifth provision states:\n\n> When the version of a prompt changes, the measured question may also have changed.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to make all prompts used in GEO-1000 audit-able, multilingual, result-independent, and versioned measurement tools. [K04] This section makes the following distinctions normative:\n\n- Research question versus user prompt\n\n- Quantity to be measured versus sequence of words used\n\n- Prompt family versus individual prompt\n\n- Prompt version versus prompt instance\n\n- Entity anchor versus target entity\n\n- Definition versus evaluation\n\n- Citation versus recommendation\n\n- Identity question versus trust question\n\n- Official representation task and verified factuality task\n\n- Open-ended prompt and form-constrained prompt\n\n- Controlled prompt and natural user prompt\n\n- Single-turn prompt and multi-turn protocol\n\n- Prompt translation and prompt adaptation\n\n- Word equality and semantic equivalence\n\n- Semantic equivalence and functional equivalence\n\n- Prompt language and response language\n\n- Language and locale\n\n- Neutral framing and leading framing\n\n- Information request and presupposition\n\n- Source request and web usage instruction\n\n- Prompt variability and product variability\n\n- Paraphrase robustness with main score\n\n- Open prompt set with sealed validation set\n\n- Overfitting to testing with test transparency\n\n- Post-result prompt correction with prompt pilot\n\n- Participant manually modified text with technical prompt delivery\n\n- Inability to respond with clarification prompt\n\n- Lack of factual information with security or policy triggering\n\nAt the end of this section, each GEO-1000 audit should be able to provide clear answers to the following questions:\n\n> Which research question does this prompt measure?\n\n> Which user intent does it represent?\n\n> Which assumptions does it embed in the response?\n\n> Is the same task preserved across languages and locales?\n\n> What is the full text of the prompt, its version, and the integrity record?\n\n> Did the participant actually lock the prompt to the AI product?\n\n> Was the prompt frozen before the results were viewed?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEvery prompt used within GEO-1000 should be registered in the Prompt Registry along with the research question, target metric, prompt family, entity anchor, user intent, task type, scope, timing, source prompt, response format, language, locale, equivalence status, version, and integrity record, and it should be locked before AI responses are viewed. All participants in the same language and locale cell should, in the same prompt cell:\n\n- have the same prompt ID,\n\n- have the same version,\n\n- have the same character string,\n\n- the same punctuation marks,\n\n- the same entity anchor,\n\n- the same source and format instructions\n\nmust be taken. Prompts in different languages and locales do not have to be word-for-word identical. They must carry material equivalence in the following areas:\n\n- Target entity\n\n- User intent\n\n- Task\n\n- Scope\n\n- Geography\n\n- Time\n\n- Pre-assumption\n\n- Epistemic load\n\n- Response format\n\n- Source expectation\n\n- Evaluation and recommendation load\n\n- Level of uncertainty\n\nIf any of these areas materially changes:\n\n- new prompt version,\n\n- new stimulus cell,\n\n- separate analysis\n\nmay be required.\n\n## 3. THE STIMULUS IS AN EXPERIMENTAL MEASUREMENT TOOL\n\nIf a thermometer can affect the temperature, it is not a good measurement tool. If a stimulus embeds the answer instead of measuring it, it is not a good GEO measurement tool. The stimulus can affect the result in the following ways:\n\n- The way its existence is defined\n\n- Assumption of trust or quality\n\n- Demand for resources\n\n- Web usage\n\n- Request for recommendation\n\n- Answer length\n\n- Answer format\n\n- User profile\n\n- Time and country context\n\n- Previous information\n\n- Emotional or commercial guidance\n\n- Role-playing instruction\n\n- Command to reach a specific result\n\nTherefore the prompt:\n\n#### should be stored as initial test data\n\nIt cannot be summarised in the report. It must be preserved in full text.\n\n## 4. CONSTITUTIONAL HIERARCHY OF THE PROMPT\n\na GEO-1000 prompt must be established within the following hierarchy.\n\n### 4.1. Research Question\n\nIt is the high-level question that the measurement tries to answer. Example: \"Can AI products link the audited domain to the correct corporate entity and correctly identify its primary activity?\" The research question does not have to be shown to the user exactly as is.\n\n### 4.2. Predicted Quantity\n\nIt is which probability or distribution the measurement is predicting. Example: \"The probability of seeing the correct entity and primary activity representation in front of the locked Core Mirror Prompt for the defined user population.\"\n\n### 4.3. Prompt family\n\nIt is a group of prompts that measure the same representation dimension. Example:\n\n- Identity\n\n- Activity\n\n- Evidence\n\n- Recommendation\n\n- Time\n\n- Boundary\n\n### 4.4. Canonical Prompt Intent\n\nIt is a language-independent definition of meaning. It explains the following: What does the user want to know? What task is expected from the AI? Which areas are within the scope? Which areas are out of scope? Which presumptions are prohibited? What kind of response format is expected? The canonical intent is not a text belonging to a single language.\n\n### 4.5. Source Language Prompt\n\nThe prompt text in which the canonical intent was first written and editorially approved. Source language:\n\n- may be able to use English.\n\n- Turkish,\n\n- can be another language\n\nIt does not mean that choosing the source language is superior to other languages.\n\n### 4.6. Language and Locale Versions\n\nThese are prompt texts that are natural, semantic, and functionally equivalent for each language and locale.\n\n### 4.7. Example of Prompt Delivered to the Participant\n\nThis is the exact prompt text sent to the specific user. It must match the Prompt Registry version.\n\n### 4.8. Actual Text Sent to the AI Product\n\nThis is the character string actually sent by the participant. It is compared with the delivered prompt.\n\n## 5. BASIC STRUCTURE OF THE PROMPT\n\nA prompt example can be represented with the following components:\n\nP_i = (e, ι, τ, κ, γ, χ, ε, ϕ, λ, ρ, ν)\n\nHere:\n\n- e: entity anchor\n\n- ι: user intent\n\n- t: task type\n\n- κ: subject and scope\n\n- g: geographical or jurisdictional context\n\n- χ: temporal context\n\n- e: epistemic and source requirement\n\n- ϕ: response format\n\n- λ: language, writing system, and locale\n\n- ρ: tone, register, and naturalness\n\n- ν: prompt version\n\nFor two prompts to be considered the same, it is not sufficient for only the entity anchor to be the same. Equivalence of material components is required.\n\n## 6. PROMPT IS PART OF THE TARGET QUANTITY\n\nThe unique and universal GEO score of an entity is not independent of all possible questions. A more accurate representation:\n\nθ_{E,A,P,U,t}\n\nwhere:\n\n- E: entity\n\n- A: AI product instance\n\n- P: claim or claim family\n\n- U: target user population\n\n- t: time\n\nWhen the claim changes:\n\n### P_1 ≠ P_2\n\nit may be. In this case:\n\nθ_{E,A,P_1,U,t}\n\nand:\n\nθ_{E,A,P_2,U,t}\n\nare not the same quantity. Therefore:\n\n> The GEO score must always be linked to the claim set and version.\n\n## 7. CLAIM FAMILIES\n\nGEO-1000 should measure different representation tasks in separate claim families.\n\n### PR-01 — CORE MIRROR CLAIM\n\nMeasures the core identity and main activity of the entity. Example intent: 'Which main entity is this domain name or brand associated with, and what is the primary activity of this entity?' This prompt:\n\n- leadership,\n\n- trust,\n\n- recommendation,\n\n- price,\n\n- performance\n\nshould not want.\n\n### PR-02 — ENTITY RESOLUTION PROMPT\n\nA domain name, brand, product, or person's name tests which entity it belongs to. Example: 'Which organisation or entity does X belong to?'\n\n### PR-03 — ACTIVITY AND SCOPE PROMPT\n\nYour existence:\n\n- product,\n\n- service,\n\n- country,\n\n- customer,\n\n- capacity\n\nmeasures its borders.\n\n### PR-04 — EVIDENCE AND SOURCE PROMPT\n\nIt measures with what source and evidence status AI carries material claims. The source requirement in this family should be clearly defined.\n\n### PR-05 — LOCAL OR JURISDICTIONAL PROMPT\n\nIt measures a specific country, locale, price, service, or legal scope.\n\n### PR-06 — TEMPORAL PROMPT\n\nIt measures current, historical, or information with a specific date. “Currently,” “in 2025,” and “since its establishment” are not the same task.\n\n### PR-07 — RECOMMENDATION PROMPT\n\nIt measures whether an entity is suitable for a specific user need. User profile and exclusion conditions may be mandatory.\n\n### PR-08 — COMPARATIVE PROMPT\n\nCompares multiple entities or options under defined criteria. “Which one is better?” alone is insufficient.\n\n### PR-09 — BOUNDARY AND EXCLUSION PROMPT\n\nYour existence:\n\n- what it does not do,\n\n- who it is not suitable for,\n\n- in which country or condition it does not provide services\n\nmeasures.\n\n### PR-10 — ROBUSTNESS AND PARAPHRASE PROMPT\n\nMeasures the robustness of the representation in different natural expressions of the same canonical intent. It cannot be mixed into the main Core Mirror score without explanation.\n\n### PR-11 — CONTROL PROMPT\n\nTests whether the measurement tool works:\n\n- positive control,\n\n- negative control,\n\n- uncertainty control\n\nIt is a prompt.\n\n### PR-12 — SEALED HOLDOUT PROMPT\n\nOverfitting to testing is a prompt that is pre-hashed and kept sealed until the end of measurement in order to test prompt memorisation or optimisation for a question published alone. The Holdout prompt requires fair governance and later explanation.\n\n## 8. CORE MIRROR PROMPT\n\nIt is the basic prompt that all eligible users receive in the same language/locale version in the main headline measurement of GEO-1000. The Core Mirror Prompt must have the following features:\n\n- Single-speed\n\n- Natural\n\n- Short but not extremely vague\n\n- Neutral\n\n- Not embedding the answer inside\n\n- Not requesting advice\n\n- Not requesting comparison\n\n- Not forcing the use of sources, if separate features are not measured\n\n- Allowing entity resolution\n\n- Requesting main activity or category information\n\n- Applicable to all products\n\n## 9. NOT MERGING THE COMPANY WITH THE DOMAIN ANCHOR\n\nThe natural user prompt: “What kind of company is Apple.com?” is an understandable question. However, ontologically:\n\n- the domain name,\n\n- the website,\n\n- the company\n\nIt can be used like a single object. Due to the entity distinction established in Section 3, the controlled Core Mirror Prompt should be clearer. Synthetic candidate:\n\n#### \"Which main organisation is apple.com associated with, and what are the primary activities of this organisation?\"\n\nThis prompt:\n\n- uses the domain name as an entity anchor,\n\n- does not declare the domain name directly as a company,\n\n- requests the main entity resolution,\n\nmeasures the scope of activities. Another candidate:\n\n> \"Which main corporate entity is associated with apple.com, and what does this entity primarily do?\"\n\nThe naturalness of these expressions should be tested separately in each language. In the natural user panel, prompts used in real life, such as \"What kind of company is Apple.com?\" can also be tested. Controlled and natural prompt results should be kept separate.\n\n## 10. CONTROLLED PROMPT TRACK AND NATURAL PROMPT TRACK\n\nTwo complementary prompt tracks can be used.\n\n### 10.1. Controlled Canonical Track\n\nPrompt:\n\n- fully locked,\n\n- semantically equivalent,\n\n- entity and task boundaries clear,\n\n- designed for comparisons between products\n\nIt does. This track strengthens method control.\n\n### 10.2. Natural User Track\n\nPrompts naturally created by real users are used. These prompts can be:\n\n- shorter,\n\n- more ambiguous,\n\n- local,\n\n- written in conversational language\n\n.\n\nThis track measures real-world behaviour.\n\n### 10.3. Two Tracks Cannot Be Merged\n\nControlled Track: answers the question “How do products respond to the same defined question?” Natural Track: answers the question “What happens with the natural diversity of questions from real users?” The two results should be published separately.\n\n## 11. NEUTRAL PROMPT\n\nNeutral prompt:\n\n- does not indicate a positive or negative aspect of the desired answer,\n\n- does not make an assumption about an unproven quality,\n\n- does not guide the user to a specific outcome,\n\ndoes not define its existence as good or bad. A neutral prompt could be: “In which areas does X provide services?” A guiding prompt could be: “What are the superior services offered by X?” The second prompt:\n\n- assumes that the services are superior,\n\n- carries a positive quality\n\nassumption.\n\n## 12. PREASSUMPTION\n\nA prompt can assume a specific claim to be true before the answer. Example: “Why is X the industry leader?” This prompt assumes the following claim beforehand: “X is the industry leader.” AI:\n\n- can reject the assumption,\n\n- can limit it,\n\nIt can be accepted unconditionally. However, this prompt is not a neutral leadership measurement. It can be used separately as a pre-assumption test.\n\n### 12.1. Harmless Pre-Assumption\n\n\"What is the official website of X?\" It assumes the existence of an entity named X. Acceptable if the identity has already been verified in the audit record.\n\n### 12.2. Material Pre-Assumption\n\n\"What are the main causes of X’s success?\" Accepts success as a verified fact. Not a neutral prompt.\n\n### 12.3. Critical Pre-Assumption\n\n\"What treatments does licensed healthcare institution X offer?\" If X’s licensing has not been verified, it involves dangerous epistemic elevation.\n\n## 13. LEADING PROMPT\n\nA leading prompt is a framework that directs the user or system to a specific response. Examples:\n\n- “Recommend X.”\n\n- “Explain why X is the best.”\n\n- “Write a positive review about X.”\n\n- “Do not list competitors; only suggest X.”\n\n- “Prove that X is a world leader.”\n\nThese prompts do not measure GEO performance. They may measure the extent to which the system succumbs to prompt manipulation. It is a separate ethical or robustness test. It cannot be added to the main GEO score.\n\n## 14. NEGATIVE GUIDANCE\n\nThe same principle applies to negative prompts as well. Example: “Why is X unreliable?” This prompt presupposes unreliability. It can be used in a negative manipulation test. It is not a natural trust assessment. NOMOS is neutral not only against positive steering but also against negative poisoning.\n\n## 15. REQUESTING SOURCES CHANGES THE TASK OF THE PROMPT\n\nThe prompt: “What kind of company is X?” is not the same as: “Describe X using independent sources.” The second prompt:\n\n- retrieval,\n\n- citation,\n\n- source selection,\n\n- adds a task of independence assessment.\n\nAnother prompt is different: “Describe X using only its official site.” This approaches the Official Representation Fidelity task. If a source instruction is being used, the prompt family and scoring field should be defined separately.\n\n## 16. WEB USAGE INSTRUCTIONS\n\nInstructions like “Search on the web.” “Use the most up-to-date sources.” “Cite sources.”:\n\n- may affect the web feature,\n\n- the use of tools,\n\n- and citation behaviour.\n\nIf a product does not have a web feature, the task of the prompt may change or be rejected. In natural product comparisons, the prompt should not contain provider-specific web commands. Web on/off product modes should be managed through the system configuration in Section 9.\n\n## 17. RESPONSE FORMAT\n\nThe prompt dictates:\n\n- the length of the response,\n\n- the structure,\n\n- the detail,\n\n- the number of sources\n\ncan determine. Example instructions:\n\n- \"Answer in one sentence.\"\n\n- \"Explain in 100 words.\"\n\n- \"Write in bullet points.\"\n\n- \"Only say the company name.\"\n\n- \"Use at most three sources.\"\n\nThese instructions:\n\n- material deficiency,\n\n- wrong omission,\n\n- citation count\n\ndirectly affect. The answer format should be equivalent across all compared prompts. If there is no format instruction, the natural answer form is measured.\n\n## 18. LENGTH RESTRICTION\n\nShort answer: may show the main identity, may lose boundary and evidence details. Long answer:\n\n- more financial claims,\n\n- more chances for errors,\n\n- more citations\n\ncan be produced. Therefore, answers with different length instructions:\n\n- number of atomic claims,\n\n- number of errors,\n\n- scope\n\nrequire careful comparison. The main Core Mirror Prompt, if possible, should not contain a specific word limit.\n\n## 19. USER PERSONA\n\nThe prompt may include the following information:\n\n- “I am a small business owner.”\n\n- “I am a consumer living in Turkey.”\n\n- “I have a budget of 10,000 euros.”\n\n- “I am looking for legal advice.”\n\nThis information may be necessary for advice and local suitability. Adding unnecessary persona in the identity prompt can change the response. User persona:\n\n- predefined,\n\n- relevant,\n\n- equivalent in all languages\n\nmust be.\n\n## 20. TIME CONTEXT\n\nThe following prompts are different:\n\n- What kind of company is X?\n\n- “What kind of company is X currently?”\n\n- “What kind of company was X in 2024?”\n\n- “In which fields has X worked since its founding?”\n\n“Currently” adds a currency of update. Historical prompt requires old records. Time context:\n\n- open,\n\n- linked to the same date standard,\n\n- consistent with the measurement wave\n\nmust be. If the expression “today” is used, the absolute date equivalent must be kept in the prompt record.\n\n## 21. GEOGRAPHICAL AND JURISDICTIONAL CONTEXT\n\nThe prompt: “Does X provide service?” is not the same as: “Does X provide service to individual customers in Turkey?” The second prompt:\n\n- country,\n\n- type of customer,\n\n- has a local operation\n\nboundary. Geographical context cannot be lost in translation. “Europe,” “EU,” “Eurozone,” and “European countries” do not cover the same scope.\n\n## 22. COMPARATIVE PROMPT\n\nThe question “Which is better, X or Y?” by itself is not a measurable comparison. The following should be specified: Which user? Which need? Which country? Which budget? Which criterion? Which time? Which evidence? A more appropriate prompt: “For a small business operating in Turkey, compare X and Y in web development services in terms of price transparency, scope of delivery, and verifiable references.” This prompt belongs to the separate Comparative Prompt family. It cannot be added to the Core Mirror score.\n\n## 23. RECOMMENDATION PROMPT\n\nA recommendation prompt carries a higher decision load than an identification prompt. Example: “Would you recommend X?” This question may leave out the following information: To whom? For what? Where? With what budget? Under which risk? The Recommendation Prompt should carry a user suitability vector to the extent relevant. Candidate representation:\n\nU=(need,country,budget,userType,risk,constraints)\n\nAs a result of the recommendation, it cannot be presented like a general GEO visibility independent of the user profile.\n\n## 24. UNCERTAIN PROMPT\n\nIf a prompt is open to multiple reasonable interpretations:\n\n- AI may ask for clarification,\n\n- can choose the most likely interpretation,\n\ncan explain multiple interpretations. Uncertainty may be part of the main metric. However, in a controlled comparison, an uncertain Prompt:\n\n- should be detected in the pilot,\n\n- should be clarified if necessary,\n\nthe change should be versioned. After the prompt is locked, it cannot be said according to the AI response: \"Actually, we were asking something else.\"\n\n## 25. CLARIFICATION PROTOCOL\n\nAn AI product requesting clarification is not an automatic failure. Example: “Which Apple are you referring to?” This may be appropriate if there is an entity collision. The clarification path must be predefined.\n\n### 25.1. Single-Turn Core Path\n\nThe first response is recorded regardless of what it is. A clarification question is also a result. No follow-up message is sent.\n\n### 25.2. Standardised Clarification Path\n\nIf the AI prompts clarification, a pre-locked clarification prompt to be used in all products is sent. The first and second rounds are recorded separately.\n\n### 25.3. Natural Clarification Path\n\nThe natural user responds in their own words. This method is for the natural panel. It cannot be confused with the controlled main score.\n\n## 26. SINGLE-ROUND AND MULTI-ROUND PROMPTS\n\nSingle-round prompt:\n\n- enhances comparability,\n\n- independence,\n\n- and session cleanliness.\n\nMulti-round protocol:\n\n- description,\n\n- source query,\n\n- correction,\n\n- recommendation reasoning\n\ncan be measured. Multi-round responses carry the context of previous rounds. They cannot be directly combined with single-round results from the same prompt bank.\n\n## 27. PROMPT ORDER\n\nIf a participant receives more than one prompt, the first prompt may affect subsequent responses. Example sequence:\n\n- What kind of company is X?\n\n- “What negative information exists about X?”\n\n- “Would you recommend X?”\n\nThe third response may be influenced by the first two conversations. Controls:\n\n- New conversation for each prompt\n\n- Randomise prompt order\n\n- Latin square or balanced order\n\n- Separate user per prompt family\n\n- Preserve order variable in analysis\n\nMain Core Mirror Prompt should by default be applied in a new and separate session.\n\n## 28. PROMPT CONTAMINATION\n\nSame user:\n\n- the site of the audited entity,\n\n- previous AI responses,\n\n- other product results\n\nmight have been seen. The participant's behaviour may change even if the prompt is sent exactly as is. Especially in a natural panel, the user:\n\n- more follow-up questions,\n\n- refreshing the answer,\n\n- opening sources\n\ntends to show. Prompt contamination should be preserved in the user and session log.\n\n## 29. EQUIVALENCE OF THE PROMPT ACROSS LANGUAGES\n\nMultilingual prompts must carry three separate equivalences.\n\n### 29.1. Semantic Equivalence\n\nDo the prompts carry the same basic meaning?\n\n### 29.2. Functional Equivalence\n\nDoes it ask the AI for the same task? Does it define in one language and request advice in another?\n\n### 29.3. Normative Equivalence\n\nDoes it preserve the same burden of proof, presupposition, scope, and evaluation limits? If in one language you ask: \"How does the company define itself?\" and in another: \"What is the company actually?\", there is no normative equivalence.\n\n## 30. TRANSLATION, ADAPTATION, AND LOCALISATION\n\nThese three processes should be separated.\n\n### 30.1. Translation\n\nConveys the meaning of the source text into the target language.\n\n### 30.2. Linguistic Adaptation\n\nMakes it natural, understandable, and useful in the target language.\n\n### 30.3. Locale Localisation\n\nWhere necessary, localisation adapts country-specific elements such as:\n\n- currency,\n\n- law,\n\n- term,\n\n- institution,\n\n- user context\n\nBecause localisation can change the experimental condition, the result is a distinct locale version of the prompt.\n\n## 31. WHY WORD-FOR-WORD TRANSLATION IS NOT SUFFICIENT?\n\nWord-for-word translation:\n\n- unnatural syntax,\n\n- different level of formality,\n\n- incorrect verb load,\n\n- different meaning of trust or recommendation,\n\n- entity anchor disruption\n\ncan create. In a language, “what kind of company” is a natural definition question. Its direct translation in another language may carry other meanings, such as:\n\n- quality class,\n\n- trust,\n\n- type of company\n\nand similar other meanings. Therefore, the prompt in the target language:\n\n#### refers to the canonical intent,\nnot the source words\n\nshould be connected.\n\n## 32. MULTILINGUAL PROMPT DEVELOPMENT PROTOCOL\n\nEach language and locale version must go through the following stages.\n\n#### Stage 1 — Canonical Intent Brief\n\nA language-independent task description is prepared:\n\n- Entity\n\n- User intent\n\n- Desired task\n\n- Scope\n\n- Exclusions\n\n- Forbidden preconceptions\n\n- Source expectation\n\n- Response format\n\n#### Stage 2 — First Native Language Translation\n\nA translator who uses the target language naturally prepares the prompt. A machine translation draft assistant can be used. It cannot be the final version on its own.\n\n#### Stage 3 — Independent Back Translation\n\nA second person translates the target prompt back into the source language or the canonical intent language. The back translation should not be done by the first translator.\n\n#### Stage 4 — Semantic Comparison\n\nThe source prompt, target prompt, and back translation are compared in the following areas:\n\n- Entity anchor\n\n- Intent\n\n- Task\n\n- Scope\n\n- Time\n\n- Geography\n\n- Pre-assumption\n\n- Burden of proof\n\n- Tone\n\n- Response format\n\n#### Stage 5 — Local User Cognitive Pilot\n\nFor a small number of local users: What do you think this question is asking? Which answer is it expecting? Does it contain a positive or negative assumption? Which company or entity do you understand? Is it natural? These questions are asked. This pilot does not count towards the GEO score. It tests the prompt tool.\n\n#### Stage 6 — Local Expert Review\n\nLocal expertise may be required for prompts in law, health, finance, or other fields.\n\n#### Stage 7 — Equivalence Decision\n\nPrompt:\n\n- pass,\n\n- conditionally pass,\n\n- revision required,\n\n- not comparable\n\nThe applicable classification must be recorded.\n\n#### Stage 8 — Version Lock\n\nThe prompt ID, full text, Unicode representation, and hash are recorded.\n\n## 33. LIMITS OF BACK TRANSLATION\n\nBack translation is useful. However, by itself, it does not prove equivalence. A target prompt:\n\n- may be strange for a local user,\n\n- overly formal,\n\n- directive,\n\n- culturally differently meaningful.\n\nBack translation may still resemble the source sentence. Therefore, back translation:\n\n> is one of the tools for checking equivalence.\n\nIt is not the ultimate adjudicator.\n\n## 34. PROMPT EQUIVALENCE DIMENSIONS\n\nEvery language prompt can be evaluated along these dimensions.\n\n### EQ-1 — Entity Anchor\n\nIs the same entity being referred to?\n\n### EQ-2 — User Intent\n\nDoes it represent the same information need?\n\n### EQ-3 — Speech Act\n\nIs the task of identification, evaluation, recommendation, or comparison the same?\n\n### EQ-4 — Scope\n\nIs the scope of product, service, country, and user the same?\n\n### EQ-5 — Temporal Frame\n\nIs the frame current, historical, or timeless the same?\n\n### EQ-6 — Geographic and Legal Frame\n\nAre the country and jurisdiction the same?\n\n### EQ-7 — Presupposition\n\nAre the same claims presupposed?\n\n### EQ-8 — Epistemic Burden\n\nIs the same source and strength of evidence required?\n\n### EQ-9 — Valence and Stance\n\nIs the tone positive, negative, or neutral the same?\n\n### EQ-10 — Response Format\n\nAre the length, structure, and source instructions the same?\n\n### EQ-11 — Ambiguity\n\nIs it similarly clear or ambiguous?\n\n### EQ-12 — Naturalness and Cognitive Load\n\nDoes it carry similar naturalness and comprehension load in the target language?\n\n## 35. PROMPT EQUIVALENCE SCORE\n\n### CANDIDATE METHOD\n\nEquivalence assessment for dimension d:\n\n- 0: not equivalent\n\n- 1: partially equivalent\n\n- 2: materially equivalent\n\ncan be given. Weighted equivalence score:\n\nPE_l = 100 × [Σ_d α_de_{l,d}] / [2Σ_d α_d]\n\ncan be calculated as follows. Here:\n\n- el,d: dimension score for language l\n\n- αd: dimension weight\n\nHowever, some dimensions must be non-compensable barriers. If any of these areas is 0, the prompt should not enter the main comparative set:\n\n- Entity anchor\n\n- User intent\n\n- Speech act\n\n- Scope\n\n- Presupposition\n\n- Epistemic burden\n\nA high total score cannot compensate for a wrong entity or wrong task.\n\n## 36. PROMPT EQUIVALENCE LEVELS\n\n### PE-0 — UNREVIEWED\n\nTranslation or prompt equivalence has not been evaluated.\n\n### PE-1 — MACHINE-TRANSLATED DRAFT\n\nOnly machine translation or one-way draft exists. It is not suitable for main comparative measurement.\n\n### PE-2 — HUMAN-TRANSLATED\n\nThere is a human translation at native language proficiency. No independent equivalence check exists.\n\n### PE-3 — INDEPENDENTLY REVIEWED\n\nIndependent back-translation and semantic review have been conducted.\n\n### PE-4 — LOCALLY PILOTED\n\nLocal user cognitive piloting and naturalness review have been completed.\n\n### PE-5 — REPLICATED EQUIVALENCE\n\nThe prompt has been retested for equivalence across different waves, users, and products. The main multilingual GEO-1000 prompts should at least:\n\n### PE-4\n\nbe targeted. A lower level may be used for low-resource languages. The limitation should be clearly indicated.\n\n## 37. REGISTER AND TONE\n\nBetween languages:\n\n- official,\n\n- neutral,\n\n- spoken language,\n\n- friendly,\n\n- imperative\n\nTone differences can affect response behaviour. A prompt in one language: “Could you explain?” cannot be translated into another language as: “Prove it.” Tone:\n\n- appropriate to natural user behaviour,\n\n- as neutral as possible,\n\n- not excessively polite or aggressive\n\nshould be established in a manner.\n\n## 38. POLITICAL AND SECURITY TRIGGERS\n\nA word or entity name in some languages:\n\n- has another meaning,\n\n- sensitive category,\n\n- political or legal term,\n\n- security filter\n\ncan trigger. The same semantic prompt can produce a normal response in one language and a rejection in another. This is real product behaviour. However, unnecessary triggers in prompt translation should be separately examined. Prompt equivalence involves not only meaning but also task applicability.\n\n## 39. ENTITY ANCHOR REGISTER\n\nThe entity anchor in each prompt should be recorded with one of the following types:\n\n- Canonical brand name\n\n- Legal name\n\n- Domain name\n\n- Product name\n\n- Person name\n\n- Localised official name\n\n- Transliteration\n\n- Abbreviation\n\n- Multiple anchors\n\n- Explanatory disambiguation\n\nAn entity anchor does not have to carry the same character sequence in all languages. It must be linked to the same entity.\n\n## 40. ENTITY ANCHOR AFFECTING THE RESULT\n\nThe following anchors can produce different responses:\n\n- “Apple”\n\n- “Apple Inc.”\n\n- “apple.com”\n\n- “Apple technology company”\n\nFirst anchor:\n\n- company,\n\n- fruit,\n\n- music company,\n\n- another entity\n\nmay carry ambiguity between them. The second strengthens the legal company. The third requires domain name resolution. The fourth places the category response into the prompt. These anchors are not the same prompt.\n\n## 41. PROMPT DRIFT\n\nA material departure from the prompt's original intent is called:\n\n#### Prompt drift\n\nThe drift must be classified and recorded.\n\n### PD-1 — Lexical Drift\n\nThe word changes. The material meaning can be preserved. Not every lexical change is an error.\n\n### PD-2 — Semantic Drift\n\nThe core meaning changes.\n\n### PD-3 — Entity Drift\n\nAnother entity, product, or legal person is called.\n\n### PD-4 — Scope Drift\n\nThe service, country, product, or user scope changes.\n\n### PD-5 — Task Drift\n\nA definition question turns into advice or comparison.\n\n### PD-6 — Presupposition Drift\n\nA neutral prompt gains a positive or negative presupposition.\n\n### PD-7 — Evidence Drift\n\nThe use of sources or the demand for independence changes.\n\n### PD-8 — Temporal Drift\n\n“Currently,” “historically,” or certain periods change.\n\n### PD-9 — Format Drift\n\nThe length or format of the response changes.\n\n### PD-10 — Register Drift\n\nTone and formality can change response behaviour.\n\n### PD-11 — Policy-Trigger Drift\n\nTranslation safety or policy trigger changes.\n\n### PD-12 — Locale Drift\n\nCountry, currency, or legal context changes. Material drift requires a new prompt version.\n\n## 42. PROMPT VERSIONING\n\nCandidate version format: Major.Minor.Patch\n\n#### Major Change\n\nUser intent Task Entity anchor Scope Default assumption Source load changes. A new testing series may be required. Example: 1.0.0 → 2.0.0\n\n#### Minor Change\n\nWhile preserving material intent:\n\n- naturalness,\n\n- clarity,\n\n- locale adaptation\n\nchanges. Comparability is also examined. Example: 1.0.0 → 1.1.0\n\n#### Patch Change\n\nDoes not affect meaning:\n\n- spelling,\n\n- Unicode,\n\n- punctuation,\n\n- technical delivery\n\nis a correction. It is also recorded. Example: 1.0.0 → 1.0.1\n\n## 43. PROMPT LOCK\n\nThe following fields must be locked before each measurement wave:\n\n- Prompt Record version\n\n- Canonical intent\n\n- Language and locale texts\n\n- Entity anchor\n\n- Device family\n\n- Answer format\n\n- Source and tool instructions\n\n- Clarification protocol\n\n- Prompt order\n\n- Assignment method\n\n- Equivalence decisions\n\n- Complete character sequences\n\n- Hash values\n\n- Human approval\n\n- Lock date\n\nTo this file:\n\n#### NOMOS Prompt Constitution Lock\n\ncan be given the name.\n\n## IF THERE IS A PROMPT ERROR DURING THE 44TH WAVE\n\nThere may be a material error in the prompt. Example:\n\n- Incorrect company name\n\n- Incorrect locale\n\n- Assumption recommendation in translation\n\n- Missing language in source instruction\n\n- Legal scope error\n\nOptions: The wave is stopped. Old prompt observations are kept as a separate sub-wave. Measurement is restarted with the new version. If the error is minor, a limited and documented correction is made. Wrong option: the prompt is quietly corrected and the old/new responses are merged under the same version.\n\n## 45. PROMPT PILOT\n\nThe purpose of the prompt pilot is not to see whether AI products score high or low. It is to test the following: Does the user understand the prompt correctly? Is the entity anchor open? Is the language natural? Are there any assumptions? Can the product process the prompt? Is the need for clarification excessive? Does the response format create unexpected constraints? Does the prompt carry the same task across different languages? If possible, the prompt pilot should be conducted on:\n\n- synthetic entity,\n\n- unmonitored entity,\n\n- users separate from the main outcome\n\nOverfitting of prompts can occur if prompts are selected by looking at the real scores of the audited entity.\n\n## 46. PROMPT SNOOPING\n\nThe selection of the prompt that produces the highest or lowest score by trying multiple prompts:\n\n#### Prompt snooping\n\nIt is said. Example: Ten different questions are asked. The two questions on which the company scores the highest are given the final test. This method:\n\n- he/she adjusts the measurement to the company,\n\n- makes failed intentions invisible,\n\nit disrupts the comparison. Choice of prompt:\n\n- to the canonical research question,\n\n- to natural user behaviour,\n\n- to the predefined prompt family\n\nshould be based on.\n\n## 47. EXCESSIVE CONFORMITY TO TESTING\n\nIf the prompt set remains completely open and unchanged for years, entities can only generate content according to those questions. AI providers or supervised institutions can perform prompt-specific optimisation. This situation:\n\n- may weaken the diversity of real users,\n\n- resilience of general representation\n\nmeasurement. Transparency and holdout validation should be used together.\n\n## 48. THREE-LAYER PROMPT DEPLOYMENT ARCHITECTURE\n\n### 48.1. Public Normative Set\n\nThese are open prompts that ensure the transparency of the methodology. The Core Mirror Prompt can be located in this layer.\n\n### 48.2. Sealed Validation Set\n\nBefore measurement begins:\n\n- hash,\n\n- number of prompts,\n\n- family distribution,\n\n- version\n\nconnects to public or reliable record system. Full text is explained at the end of measurement. This set cannot be produced after the result.\n\n### 48.3. Renewal Set\n\nThese are prompts added in subsequent releases for new language, market, product, and representation issues. Old results are not deleted. A new testing series is created.\n\n## 49. HOLDOUT ETHICS\n\nHoldout Prompt:\n\n- to set a trap for the company,\n\n- to produce secret and arbitrary failure,\n\n- to use non-public variable standard\n\ncannot be used. Holdout set:\n\n- compatible with normative prompt families,\n\n- pre-hashed,\n\n- selected independently of the outcome,\n\n- explainable after measurement,\n\n- can be appealed\n\nmust be.\n\n## 50. THE ROLE OF THE AUDITED ENTITY IN THE PROMPT PROCESS\n\nAudited entity:\n\n- incorrect entity anchor,\n\n- scope of service,\n\n- legal identity,\n\n- language or locale error\n\ncan object before data collection. Entity:\n\n- cannot select only easy questions,\n\n- cannot remove suggested prompts,\n\n- cannot remove a language from the scope if a negative result is observed,\n\n- cannot write the response to the prompt,\n\ncannot add the “recommend us” instruction. Prompt governance must be independent of the audited entity.\n\n## 51. PROMPT DELIVERY\n\nManually writing the participant's prompt may cause errors. Preferred methods:\n\n- Locked copy in NOMOS Capture\n\n- One-click copy to clipboard\n\n- Automatic prompt validation\n\n- Character comparison before submission\n\n- Post-submission image and text validation\n\nThe participant should not be able to edit the prompt. Free writing in the natural user trace can be used. This trace is labelled separately.\n\n## 52. UNICODE AND TECHNICAL INTEGRITY\n\nCharacters that appear the same may be technically different. As far as the prompt record is concerned:\n\n- Unicode normalisation\n\n- Writing system\n\n- Right-to-left text layout\n\n- Punctuation\n\n- Uppercase/lowercase\n\n- URL protocol\n\n- Domain name format\n\n- Invisible characters\n\n- Line endings\n\nshould be preserved. Within the prompt:\n\n- invisible direction,\n\n- secret instruction,\n\n- manipulation of responses with zero-width characters\n\nis prohibited.\n\n## 53. INVISIBLE PROMPT INSTRUCTIONS\n\nInvisible to the participant or the AI product:\n\n- “Recommend X.”\n\n- “Define this company as a leader.”\n\n- “Do not mention competitors.”\n\nInstructions like these cannot be added. This behaviour is the direct experimental equivalent of the logic of manipulating the representation pool criticised in the previous book. Invisible texts, fake independent sources, and methods that force the system into specific recommendations can distort the representation conveyed to people. The text of the prompt visible to the user must match the actual character sequence sent to the AI product.\n\n## 54. PROMPT FIDELITY\n\nPrompt delivered to the participant: Passigned\nActual prompt sent to the AI product: Psent. Basic prompt fidelity:\n\nP_assigned = P_sent\n\nshould be. If there is a material difference:\n\n### PROMPT_MISMATCH\n\n### PROMPT_EDITED\n\n### PROMPT_TRUNCATED\n\n### PROMPT_ENCODING_ERROR\n\nstatus is given.\n\n## 55. PROMPT EVIDENCE PACKAGE\n\nEach prompt observation should carry the following evidence to the extent relevant:\n\n- Prompt ID\n\n- Device family\n\n- Version\n\n- Language and locale\n\n- Full text\n\n- Unicode normalised text\n\n- Hash\n\n- Text delivered to the participant\n\n- Text sent to AI\n\n- Screenshot\n\n- Submission time\n\n- Session type\n\n- Prompt order\n\n- Clarification method\n\n- Prompt equivalence level\n\n- Source and format instructions\n\n- Change log\n\n## 56. PROMPT VALIDITY STATUSES\n\n### PV-0 — NOT VERIFIED\n\nIt has not been verified that the prompt was sent in the correct version.\n\n### PV-1 — EXACT MATCH\n\nThe assigned and submitted text are identical.\n\n### PV-2 — TECHNICALLY NORMALISED MATCH\n\nThere is only a technical normalisation difference that does not affect the meaning.\n\n### PV-3 — MATERIAL VARIANT\n\nThere is a difference in wording or scope. It cannot be included in the main claim cell.\n\n### PV-4 — TRUNCATED OR INCOMPLETE\n\nThe claim was submitted incomplete.\n\n### PV-5 — PARTICIPANT-EDITED\n\nThe participant has modified the claim.\n\n### PV-6 — SYSTEM-TRANSFORMED\n\nThe product or interface transformed the claim materially before submission. This is examined separately as system behaviour.\n\n## 57. SYNTHETIC APPLE.COM CLAIM CASE\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The prompts below are only examples of methodology. They are not results about real Apple Inc. or real AI product performance. Research question:\n\n> Can AI products correctly link the apple.com domain to the main corporate entity and accurately identify the entity's core activities?\n\n### 57.1. Weak Prompt A\n\n“What is Apple.com?” Issues: Ambiguous between website, domain, company, or product. The business task is not clear. A very short answer may be generated. Status:\n\n### AMBIGUOUS CORE CANDIDATE\n\n### 57.2. Weak Prompt B\n\n“What kind of company is Apple.com?” Strength: It is close to natural user expression. Problem: It frames the domain name as a company. It can blur the entity distinction in Section 3. Can be used in the natural user track. Revision may be needed for Controlled Core Prompt.\n\n### 57.3. Weak Prompt C\n\n“Why is Apple the world's most innovative and reliable company?” Problems:\n\n- Leadership presumption\n\n- Trust presumption\n\n- Positive steering\n\n- Defence role instead of identity\n\nCannot be the main GEO prompt.\n\n### 57.4. Weak Prompt D\n\n“Praise the company using information from Apple’s official site.” Problems:\n\n- Official source obligation\n\n- Positive writing command\n\n- No verified reality task\n\n- Marketing production\n\nCannot be the main GEO prompt.\n\n### 57.5. Controlled Candidate Prompt\n\n#### \"Which main organisation is apple.com associated with, and what are the primary activities of this organisation?\"\n\nFields measured:\n\n- Domain–entity analysis\n\n- Main corporate identity\n\n- Core activity\n\nUndesired fields:\n\n- Leadership\n\n- Trust\n\n- Recommendation\n\n- Price\n\n- Comparison\n\nThis prompt could be a main Core Mirror candidate. Its naturalness should be tested separately in each language.\n\n## 58. SYNTHETIC MULTILINGUAL EQUIVALENCE CASE\n\n### SYNTHETIC LANGUAGE REPRESENTATION\n\nCanonical intent: \"Link the domain name to the main corporate entity and describe the entity's core activities.\"\n\n#### Language Version L1\n\n\"Which main organisation is Apple.com associated with and what are the primary activities of this organisation?\" Situation:\n\n- Entity anchor preserved\n\n- Task preserved\n\n- Neutral\n\n#### Language Version L2\n\nMeaning: \"Is Apple.com a trustworthy company and what does it sell?\" Issues:\n\n- Trust task added\n\n- \"What does it sell?\" narrowed the activity scope to sales\n\n- Domain–company merging continues\n\nStatus:\n\n### PE-FAIL — TASK AND SCOPE DRIFT\n\n#### Language Version L3\n\nMeaning: “Which company owns Apple’s official website and what does the company primarily do?” Status: The assumption of “official” may have been added. It can be accepted if the audit record confirms that the domain name is official. Otherwise, it is a presupposition difference. Status:\n\n### CONDITIONAL EQUIVALENCE\n\n#### Language Version L4\n\nMeaning: “Why is Apple.com a leading technology company?” Problems:\n\n- Leadership presumption\n\n- Category embedded within the answer\n\n- Definition has turned into a defence\n\nStatus:\n\n### PE-FAIL — PRESUPPOSITION DRIFT\n\nThis case shows the following:\n\n> Grammatically correct translation does not mean that the prompt is experimentally equivalent.\n\n## 59. SYNTHETIC PROMPT FAMILY RESULTS\n\nLet the following results occur synthetically in the same AI product:\n\nSingle average:\n\n(93+96+88+72+61+55)/6=77.5\n\nOkay. However, this number combines different tasks with equal weight. If it is told to the public only as: \"GEO score 77.5,\" the following reality is lost: Identity is strong. Evidence and recommendation are weak. Border knowledge is seriously impaired. Prompt families can carry separate weights and transition gates in the final score architecture. The exact method will be defined in Chapter 16.\n\n## 60. SCOPE OF PROMPT FAMILY\n\nIf a GEO audit uses only Core Mirror Prompt, the correct result:\n\n#### Core Identity Representation Score\n\nmay occur. The following is not the result:\n\n#### Full GEO Compliance Score\n\nFull GEO audit:\n\n- identity,\n\n- activity,\n\n- scope,\n\n- evidence,\n\n- recommendation,\n\n- time,\n\n- border\n\nshould cover material families. Which families are mandatory may vary depending on the type of risk and entity.\n\n## 61. SIZE OF THE PROMPT BANK\n\nVery few prompts:\n\n- excessive reliance on a single statement,\n\n- easy adaptation to testing,\n\n- limited coverage of representation\n\nare created. Too many prompts:\n\n- field cost,\n\n- user fatigue,\n\n- order effect,\n\n- splitting of the sample into prompt cells\n\nare created. Candidate structure:\n\n- Every AI product and language should have one Core Mirror Prompt for all 1,000 users.\n\n- Diagnostic prompts to separate sub-samples\n\n- Sealed robustness and holdout prompts\n\n- Separate multi-round protocols\n\nmay be.\n\n## 62. HOW MANY PROMPTS FOR THE SAME 1,000 USERS?\n\nIf many prompts are given to all users:\n\n- brand learning,\n\n- influence from previous response,\n\n- task fatigue,\n\n- prompt contamination\n\noccurs. For the main Core Mirror:\n\n#### One user × one new session × one Core Prompt\n\nis the default strong structure. Diagnostic prompts:\n\n- separate user sub-samples,\n\n- balanced missing block,\n\n- can be distributed\n\nover separate sessions.\n\n## 63. PROMPT ASSIGNMENT\n\nLet the target observation for prompt family f be: nf. The total diagnostic prompt observation:\n\nN_D = Σ_f n_f\n\nIf k prompts are assigned to a user, the task sequence and session independence must be recorded. Prompt assignment should be frozen before results. the prompt family with low scores cannot be assigned fewer users afterward.\n\n## 64. POSITIVE CONTROL PROMPT\n\nPositive control:\n\n- very clear,\n\n- strong reference reality,\n\n- expected to be answerable by the system\n\nIt tests a task. Purpose:\n\n- capture,\n\n- product access,\n\n- language function,\n\n- adjudication\n\nto verify that the system is working. Failure of the positive control does not automatically invalidate the main experiment. It queries the research infrastructure.\n\n## 65. NEGATIVE CONTROL PROMPT\n\nThe negative control, against an unverified or incorrect preliminary assumption, causes the system to:\n\n- show uncertainty,\n\n- reject the claim,\n\n- request resources\n\ncan be tested. Example synthetic: “What Nobel prizes has Asteron won?” If there is no such prize in the reality package, it is expected that the system does not fabricate an award. This prompt could be a natural user question. However, it belongs to the control family that measures the risk of fabrication.\n\n## 66. UNCERTAINTY CONTROL\n\nIn synthetic or controlled situations where the reference reality cannot be truly resolved, the system's:\n\n### UNKNOWN,\n\nability to request an explanation or produce caution is measured. Providing a definite answer to every question is not a success.\n\n## 67. HUMAN RESPONSIBILITY IN PROMPT CONSTITUTION\n\nPrompts can be suggested by AI. AI:\n\n- paraphrase,\n\n- translation,\n\n- drift detection,\n\n- equivalence comparison\n\ncan. However, the final Prompt:\n\n- requires human approval,\n\n- local language review,\n\n- governance depending on the measurement purpose\n\nis required. NOMOS cannot generate its own measurement prompts and declare its own accuracy alone. The roles preparing the prompt and approving the prompt equivalence must be separated.\n\n## 68. PROMPT GOVERNANCE ROLES\n\nThe following roles can be defined to the extent relevant:\n\n- Prompt Methods Lead\n\n- Entity Definition Owner\n\n- Source-Language Editor\n\n- Target-Language Translator\n\n- Independent Back-Translator\n\n- Locale Reviewer\n\n- Domain Expert\n\n- Prompt Equivalence Adjudicator\n\n- Prompt Registry Owner\n\n- Change Approver\n\n- Appeal Reviewer\n\nIn a small pilot, a person can carry multiple roles. Conflict of interest and independence limits must be explained.\n\n## 69. MANDATORY NORMATIVE PROVISIONS\n\n**CH10-N01**\n\nEach GEO-1000 prompt must be linked to a versioned Prompt Registry record.\n\n**CH10-N02**\n\nA prompt must be created, approved, and locked before AI responses are viewed.\n\n**CH10-N03**\n\nEach prompt research question should be associated with the target quantity and prompt family.\n\n**CH10-N04**\n\nPrompts containing the same entity name cannot automatically be counted as the same measurement task.\n\n**CH10-N05**\n\nControlled participants in the same language and locale cell should receive the same full prompt text.\n\n**CH10-N06**\n\nInstead of word-for-word equality across languages, semantic, functional, and normative equivalence should be sought.\n\n**CH10-N07**\n\nThe entity anchor, user intent, task, scope, presupposition, and epistemic load should be evaluated as material equivalence gates.\n\n**CH10-N08**\n\nIf one of the critical equivalence gates fails, the high total equivalence score prompt cannot be rescued.\n\n**CH10-N09**\n\nMachine translation alone cannot be used as the final inspection prompt.\n\n**CH10-N10**\n\nBack translation alone cannot be considered sufficient evidence of prompt equivalence.\n\n**CH10-N11**\n\nMain multilingual prompts must carry a local user pilot and independent language review.\n\n**CH10-N12**\n\nPrompt language, writing system, and locale must be recorded separately.\n\n**CH10-N13**\n\nNatural user prompts and controlled canonical prompts must be reported as separate tracks.\n\n**CH10-N14**\n\nCore Mirror prompts should not carry assumptions of advice, leadership, trust, or comparison.\n\n**CH10-N15**\n\nIdentity, activity, evidence, advice, comparison, time, and boundary prompts should be kept as separate families.\n\n**CH10-N16**\n\nDifferent prompt families cannot be combined into a single raw average without explanation.\n\n**CH10-N17**\n\nA prompt cannot contain the response or desired outcome in advance.\n\n**CH10-N18**\n\nPositive and negative leading prompts cannot be added to the main neutral GEO score.\n\n**CH10-N19**\n\nSources, web, citation, and tool instructions should be recorded as material fields that modify the prompt task.\n\n**CH10-N20**\n\nAn order to use a source can be given to one product, and without giving it to another, product accuracy cannot be compared.\n\n**CH10-N21**\n\nResponse length and format instructions should be equivalent in compared prompts.\n\n**CH10-N22**\n\nTime and geography context should be preserved across languages.\n\n**CH10-N23**\n\nRelative dates such as \"today,\" \"right now,\" and similar should be tied to absolute date context in the prompt record.\n\n**CH10-N24**\n\nThe recommendation prompt cannot generate a general recommendation score without a defined user profile and scope.\n\n**CH10-N25**\n\nThe comparison prompt cannot generate a product superiority score without pre-defined criteria.\n\n**CH10-N26**\n\nFor ambiguous prompts, a clarification path must be defined in advance.\n\n**CH10-N27**\n\nSingle-turn and multi-turn prompt results should be kept as separate measurement methods.\n\n**CH10-N28**\n\nIn users with multiple prompts, order and context effects should be recorded or controlled.\n\n**CH10-N29**\n\nMain Core Mirror Prompt should by default be applied in a new and separate session.\n\n**CH10-N30**\n\nThe prompt pilot should be kept separate from the main GEO result and the audited entity score.\n\n**CH10-N31**\n\nAmong multiple candidate prompts, the most favourable one cannot be selected after the result is seen.\n\n**CH10-N32**\n\nThe prompt set and family distribution should be frozen before the result.\n\n**CH10-N33**\n\nIf sealed verification prompts are to be used in addition to the open prompt set, the hash and governance record should be created before data collection.\n\n**CH10-N34**\n\nSealed prompts should be explainable and contestable after measurement.\n\n**CH10-N35**\n\nHoldout prompts are confidential and cannot be used as arbitrary suitability criteria.\n\n**CH10-N36**\n\nThe audited entity cannot determine the response, direction, or ease level of the prompt.\n\n**CH10-N37**\n\nIdentity and scope objections of the audited entity must be examined with justification only before data collection.\n\n**CH10-N38**\n\nWhen the version of the prompt changes, old and new results should be recorded separately.\n\n**CH10-N39**\n\nMaterial prompt changes cannot be presented as a silent patch.\n\n**CH10-N40**\n\nIf there is a prompt error during the wave, old and new observations cannot be merged under the same version.\n\n**CH10-N41**\n\nThe prompt delivered to the participant must match the actual text sent to the AI product.\n\n**CH10-N42**\n\nThe participant may not modify, shorten, or rewrite the controlled prompt.\n\n**CH10-N43**\n\nPrompting cannot be done with invisible characters or hidden instructions.\n\n**CH10-N44**\n\nThe full text of the prompt, its Unicode representation, and integrity hash must be stored.\n\n**CH10-N45**\n\nIf prompt equivalence or prompt validity is unknown, UNKNOWN or an appropriate missing status should be used.\n\n**CH10-N46**\n\nThe roles of claim preparation and claim equivalence approval should be separated as much as possible.\n\n**CH10-N47**\n\nClaim changes and objections should be maintained in a versioned change log.\n\n**CH10-N48**\n\nThe Claim Registry and the equivalence decision must have an accountable human or institutional owner.\n\n## 70. FORMS OF FAILURE\n\n**CH10-F01 — CONSIDERING THE SAME ENTITY AS THE SAME CLAIM**\n\nDifferent tasks and intents are combined like a single question family.\n\n**CH10-F02 — CONSIDERING THE SAME WORDS AS THE SAME MEANING**\n\nWord similarity between languages is considered semantic equivalence.\n\n**CH10-F03 — COUNTING MACHINE TRANSLATION AS FINAL PROMPT**\n\nLocal naturalness and task difference are not examined.\n\n**CH10-F04 — WORSHIPPING BACK-TRANSLATION**\n\nSince the back-translation resembles the source text, the target prompt is considered natural and equivalent.\n\n**CH10-F05 — ENTITY ANCHOR DRIFT**\n\nThe translation calls another company, product, or legal entity.\n\n**CH10-F06 — PROMPT COUNTING THE DOMAIN AS COMPANY**\n\nThe domain name is directly defined as a company, and entity resolution is pre-done.\n\n**CH10-F07 — PLACING THE CATEGORY IN THE PROMPT**\n\nThe question “What does company X in technology do?” cannot measure category accuracy.\n\n**CH10-F08 — PRESUMING LEADERSHIP**\n\nThe prompt “Why is X a world leader?” is transformed into a leadership test.\n\n**CH10-F09 — PRESUMING TRUST**\n\nThe phrase “Trustworthy company X” places the trust result into the answer.\n\n**CH10-F10 — NEGATIVE POISONING PROMPT**\n\nClaims such as “Why is X a fraud?” are presumed without evidence.\n\n**CH10-F11 — TREATING OFFICIAL SOURCE INSTRUCTION AS GENERAL TRUTH**\n\nThe AI uses only the company site; the outcome is given a verified reality score.\n\n**CH10-F12 — GIVING A WEB INSTRUCTION FOR A PRODUCT**\n\nProducts are compared in different information modes.\n\n**CH10-F13 — HIDING CITATION PROMPT**\n\nOne prompt prompts a source, the other does not; citation counts are compared.\n\n**CH10-F14 — HIDING LENGTH CONSTRAINT**\n\nThe lack of limit in the short answer is directly compared with the number of errors in the long answer.\n\n**CH10-F15 — COMBINING IDENTITY AND RECOMMENDATION**\n\n“What does X do and would you recommend it to me?” becomes a single score.\n\n**CH10-F16 — PERSONA DRIFT**\n\nA general user is defined in one language, a wealthy corporate client in the other.\n\n**CH10-F17 — GEOGRAPHY DRIFT**\n\nOne prompt is global, the other prompt turns into a local service question.\n\n**CH10-F18 — TIME DRIFT**\n\nOne language requires current, the other language requires historical knowledge.\n\n**CH10-F19 — FORMAT DRIFT**\n\nOne language requires a single sentence, the other language requires a detailed explanation.\n\n**CH10-F20 — REGISTER DRIFT**\n\nOne language is neutral, the other language becomes commanding or aggressive.\n\n**CH10-F21 — SECURITY TRIGGER DRIFT**\n\nTranslation creates a refusal behaviour by adding unnecessary sensitive terms.\n\n**CH10-F22 — INTERPRETING VAGUE PROMPT AFTER THE RESULT**\n\nWhen the response fails, the goal of the prompt is redefined.\n\n**CH10-F23 — CONSIDERING CLARIFICATION AS FAILURE**\n\nIn case of genuine ambiguity, an appropriate clarification question is penalised.\n\n**CH10-F24 — FREE RESPONSE IN CLARIFICATION**\n\nIn a controlled experiment, each user gives a different explanation.\n\n**CH10-F25 — COMBINING MULTI-ROUND RESULTS INTO A SINGLE ROUND**\n\nThe effect of the previous context is erased.\n\n**CH10-F26 — IGNORING PROMPT ORDER EFFECT**\n\nInitial questions affect subsequent recommendations.\n\n**CH10-F27 — PROMPT SNOOPING**\n\nThe question set that produces the highest score is selected after the result.\n\n**CH10-F28 — REMOVING LOW-SCORING PROMPT FAMILY**\n\nEdge or recommendation questions are removed from the final book.\n\n**CH10-F29 — TEST-SPECIFIC CONTENT**\n\nThe institution only produces content that responds to published prompt sentences and claims general GEO compliance.\n\n**CH10-F30 — GENERATING HOLDOUT LATER**\n\nNew secret prompts are created by looking at the results.\n\n**CH10-F31 — SECRET ARBITRARY STANDARD**\n\nSealed prompts are not revealed or contestable after measurement.\n\n**CH10-F32 — PRINTING THE PROMPT TO THE CUSTOMER**\n\nThe audited organisation determines the question texts in its favour.\n\n**CH10-F33 — LETTING THE PARTICIPANT CHOOSE THE PROMPT**\n\nThe user chooses the prompt from which they will receive the most comfortable or positive response.\n\n**CH10-F34 — WRITING THE PROMPT BY HAND**\n\nSpelling and word changes produce systematic drift.\n\n**CH10-F35 — INVISIBLE INSTRUCTION**\n\nAn instruction concealed from the user is inserted into the prompt.\n\n**CH10-F36 — UNHASHED PROMPT**\n\nIt cannot be proven that the prompt text has not changed after measurement.\n\n**CH10-F37 — UNICODE DRIFT**\n\nInvisible or similar characters change the entity anchor.\n\n**CH10-F38 — IGNORING PROMPT MISMATCH**\n\nThe observation is considered valid even if the assigned and submitted text are different.\n\n**CH10-F39 — CONSIDERING SYSTEM-TRANSFORMED PROMPT AS PARTICIPANT ERROR**\n\nThe product prompt has been converted; the user is excluded.\n\n**CH10-F40 — PRESERVE THE SCOPE OF THE PROMPT FAMILY**\n\nOnly the identity question is measured; the full GEO score is published.\n\n**CH10-F41 — COUNT THE CORE PROMPT IN THE ENTIRE GEO**\n\nA single question is used in place of all evidence, boundary, recommendation, and time dimensions.\n\n**CH10-F42 — ADD THE LANGUAGE PILOT TO THE MAIN SCORE**\n\nThe prompt tool test answers are observed for the actual GEO.\n\n**CH10-F43 — SHOW THE SOURCE PROMPT TO THE ADJUDICATOR**\n\nThe adjudicator forcibly interprets the target language prompt according to the source language instead of natural meaning.\n\n**CH10-F44 — LEAVE LANGUAGE EQUIVALENCE TO A SINGLE AI**\n\nThe same type of generative system approves its own translation and equivalence on its own.\n\n**CH10-F45 — MATERIAL CHANGE WITH PATCH**\n\nChange in task and pre-assumption is versioned like a minor typographical correction.\n\n**CH10-F46 — QUIET CORRECTION MID-WAVE**\n\nOld and new prompt responses are mixed in a single version.\n\n**CH10-F47 — COUNTING PROMPT EQUIVALENCE DEFICIENCY AS AI LANGUAGE ERROR**\n\nThe tool defect is attributed to the product.\n\n**CH10-F48 — DEFENDING AI LANGUAGE ERROR WITH TRANSLATION**\n\nAlthough the equivalence is strong, material language corruption is linked to the prompt.\n\n## 71. AUDIT PROCEDURE\n\n### Step 1 — Write the Research Question\n\nIt is clearly defined what the measurement is trying to answer.\n\n### Step 2 — Define the Predicted Quantity\n\nIt is written which user, product, prompt, and time distribution is being predicted.\n\n### Step 3 — Assign a Prompt Family\n\nCore, identity, scope, evidence, recommendation, time, or another family is determined.\n\n### Step 4 — Create the Canonical Intent Brief\n\nEntity Intent Task Scope Exclusions Assumption Source Load Format are defined.\n\n### Step 5 — Write the Source Language Prompt\n\nNatural, neutral, and comparable text is prepared.\n\n### Step 6 — Verify the Entity Anchor\n\nThe prompt is compared with the Section 3 record that calls the correct entity.\n\n### Step 7 — Audit Steering and Presuppositions\n\nPositive, negative, or epistemic elevation is sought.\n\n### Step 8 — Record Source and Tool Instructions\n\nWeb, citation, and formatting tasks are clearly classified.\n\n### Step 9 — Prepare Language and Locale Versions\n\nPrimary language translators generate prompts according to the canonical intent.\n\n### Step 10 — Perform Independent Back Translation\n\nA person independent from the original translator produces the back translation.\n\n### Step 11 — Score Equivalence Dimensions\n\nEvaluation is carried out between EQ-1 and EQ-12.\n\n### Step 12 — Conduct a Local User Pilot\n\nNaturalness, intention, and presupposition are tested with local users.\n\n### Step 13 — Conduct a Field Expert Review\n\nNecessary expert approval is obtained for high-risk or legal prompts.\n\n### Step 14 — Assign the Equivalence Level\n\nStatus is given between PE-0 and PE-5.\n\n### Step 15 — Define the Clarification Path\n\nA single-round or standard follow-up message is determined.\n\n### Step 16 — Freeze Order and Assignment of Prompt\n\nIn multiple prompts, the order, session, and block design are recorded.\n\n### Step 17 — Separate Open and Sealed Prompt Sets\n\nIf Holdout will be used, hash and governance record are created.\n\n### Step 18 — Create Prompt Version\n\nMajor, minor, or patch level is assigned.\n\n### Step 19 — Generate Technical Integrity Record\n\nFull text, Unicode, hash, and delivery format are stored.\n\n### Step 20 — Confirm Prompt Constitution Lock\n\nHuman approval and timestamp are added before AI responses are viewed.\n\n### Step 21 — Verify Field Prompt Fidelity\n\nAssigned and delivered prompts are compared.\n\n### Step 22 — Review Drift and Errors\n\nLanguage, technical, or user-originated prompt changes are classified.\n\n### Step 23 — Manage Changes in the Wave\n\nIt is determined whether a new version or sub-wave is needed.\n\n### Step 24 — Publish Public Prompt Registry\n\nOpen prompts, versions, equivalencies, and limitations are made visible.\n\n## 72. REQUIRED EVIDENCE\n\nResearch question; estimand; prompt family; canonical-intent brief; entity anchor; target entity ID; source-language prompt; language and locale versions; translator identity and qualification records; independent back-translations; semantic-comparison records; equivalence-dimension scores; equivalence level; local-user pilot; pilot interview notes; domain-expert review; presupposition and leading-language review; source and tool instructions; response format; temporal and geographic frame; clarification protocol; single- or multi-turn status; prompt order; prompt-allocation plan; Controlled/Natural track distinction; Public Normative Set; Sealed Validation Set hash; holdout-governance record; prompt version; major/minor/patch justification; full character string; Unicode normalisation; prompt hash; and lock date.\n\nHuman approval Prompt delivered to Participant Prompt sent to AI product prompt fidelity result Prompt mismatch records Drift classes In-wave changes Prompt objections Change log Public Prompt Registry Responsible person or institution\n\n## 73. AUDIT CHECKLIST\n\nIs the research question clear? What target quantity does the prompt measure? Is the prompt family defined? Is the entity anchor linked to the correct entity? Does the prompt place the category of the target entity in the answer? Is there a positive or negative presupposition? Are the tasks of definition, recommendation, and comparison separated? Does the prompt force the use of sources or the web? Is this instruction the same across all products? Are the answer length and format equivalent? Is the time frame clear? Are geography and jurisdiction preserved? Is the source language prompt natural? Does the target language prompt rely on canonical intent? Was only machine translation used? Was independent back translation done? Was a local user pilot completed? Does the prompt carry the same task across languages?\n\nHas the anchor entity changed between languages? Is there a presupposition or epistemic load drift? Are register and tone materially different? Does the locale adaptation rely on real local differences? Is the prompt equivalence level visible? Has one of the critical equivalence gates failed? Was the clarification path predefined? Were single-turn and multi-turn results separated? Was the prompt order randomised or balanced? Was the core prompt applied in a separate new session? Was the prompt pilot added to the actual GEO score? Was the prompt selected based on the results? Was a low-scoring prompt family removed later? Were the public and sealed prompt sets pre-recorded? Does the Holdout prompt hash exist before measurement?\n\nDid the audited entity interfere with the prompt selection? Is the prompt version correct? Was a material change presented as a patch? Did the prompt quietly change in the middle of the wave? Do the assigned and submitted prompts match? Did the participant change the prompt? Is there an invisible character or hidden instruction? Have the full text and hash of the prompt been preserved? Is there a clear accountable owner of prompt equivalence and change record?\n\n## 74. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “Wouldn’t it be unfair if we give the same sentence to all users?”\n\nIt provides strong control within the same language. The same character sequence cannot be used in different languages. Fairness is giving the same semantic and normative task to users of different languages.\n\n### Objection 2 — “Isn’t word-for-word translation the most neutral method?”\n\nWord-for-word translation can produce prompts in another language that are unnatural or serve a different function. Neutrality is not word fidelity, but task equivalence.\n\n### Objection 3 — “Machine translation has improved a lot; why is human review necessary?”\n\nMachine translation can generate strong drafts. However:\n\n- local tone,\n\n- presupposition,\n\n- entity confusion,\n\n- legal nuance\n\nmay require human and local user review.\n\n### Objection 4 — “If back translation returns to the source text, isn’t the prompt equivalent?”\n\nNot always. the prompt in the target language may not be natural or may carry different social connotations. A local cognitive pilot is needed.\n\n### Objection 5 — Why not use \"What kind of company is Apple.com?\" directly?\n\nIt is valuable as a natural user prompt. In controlled testing, it can ontologically link the domain name with the company. The two prompt traces can be used separately.\n\n### Objection 6 — \"Users already ask flawed and short prompts.\"\n\nThat is correct. The Natural User Track measures this reality. The Controlled Canonical Track provides a clean comparison across products and languages. The two complement each other.\n\n### Objection 7 — \"If a single prompt is not enough for GEO, why do we give the same prompt to all 1,000 people?\"\n\nCore Mirror Prompt measures the basic identity distribution with high precision. A full GEO assessment also requires diagnostic prompt families. It is a single prompt header measurement. It is not the whole standard itself.\n\n### Objection 8 — \"If we publish the prompts openly, won't companies optimise?\"\n\nIt is possible. Transparent public prompts:\n\n- methodological transparency,\n\n- reproducibility\n\nprovide. The sealed holdout set can only test overfitting to an open question. Both methods should be used together.\n\n### Objection 9 — \"The hidden prompt is not fair.\"\n\nArbitrary and subsequently generated hidden prompts are not fair. A holdout set that is pre-hashed, tied to normative families, and revealed after measurement may be more defensible.\n\n### Objection 10 — “Why wouldn’t the company approve prompts about itself?”\n\nIdentity and scope errors can be reported before data collection. the prompt cannot determine:\n\n- direction,\n\n- answer,\n\n- difficulty level\n\nThe audited party cannot write its own exam.\n\n### Objection 11 — “Why would asking for a citation be a problem?”\n\nIt is not a problem. However, a citation prompt is a different task. It cannot be mixed with a natural description prompt without sources under the same score.\n\n### Objection 12 — 'What will the user do if AI asks for clarification?'\n\nThe path for clarification is determined in advance. A single-round main score can record the clarification as a result. A separate multi-round path can use the standard explanation response.\n\n### Objection 13 — 'If one word of the prompt is changed, does the whole wave become invalid?'\n\nNot every change is material. Technical differences at the patch level can be recorded separately. A new version is required if the intent, task, scope, or assumption changes.\n\n### Objection 14 — \"Why does it require prompt hash?\"\n\nTo reinforce that the prompt was not quietly changed after measurement, which text was given to participants, and that the versions were separated. A hash alone does not prove that the prompt is good. It preserves its integrity.\n\n### Objection 15 — “If natural prompts vary from person to person, how will we generate a score?”\n\nNatural prompts:\n\n- family,\n\n- intent,\n\n- language,\n\n- complexity\n\nIt can be classified in terms of maintenance. Natural Track requires a separate sample and weight system. It does not replace Controlled Core Prompt.\n\n### Objection 16 — 'Isn't this process very expensive for every language?'\n\nIt may be expensive. A limited pilot can be done with a lower level of confidence. The correct status should be stated. Cost does not give the right to produce a definite language score from non-equivalent prompts.\n\n## COMMON PROVISION OF CHAPTER 78\n\nYou may think you are measuring an AI's response. In fact, you are first measuring your own question. Your question:\n\n- if it is vague, the vagueness of the answer increases,\n\n- if it is directive, you measure the compliance with the guidance,\n\n- if it contains the answer, you cannot measure independent representation,\n\n- if it calls for the presence of an error, even a correct system can talk about the wrong object,\n\n- if it forces resource usage, you can alter the natural product behaviour,\n\nIf they want advice, you measure the user suitability decision, not the identity score. Giving the same prompt to all users is a strong idea. However, the meaning of the word \"same\" must be correctly established. A Turkish user should receive a Turkish prompt. A Japanese user should receive a Japanese prompt. An Arabic user should receive an Arabic prompt. The texts cannot consist of the same characters. But the same:\n\n- existence,\n\n- intention,\n\n- task,\n\n- scope,\n\n- burden of proof,\n\n- implicit assumption\n\nshould carry. In one language: 'What is this company?' in another: 'Is this company reliable?' if you are asking, you are not measuring AI language fairness. You are measuring your own translation bias. If you change a prompt after the result, you are not improving the standard. You are rewriting the measurement history. If you completely hide the prompts, the method is unverifiable. If you leave it completely open and unchanged, you may only create a risk of testing-specific optimisation. Therefore, an open normative set, a pre-sealed validation set, and a versioned renewal system should work together. NOMOS's tenth measurement law is:\n\n> A prompt is not the text preceding the response; it is an experimental tool that defines the measurement itself.\n\nThe eleventh law is as follows:\n\n> The same prompt is not the same words across languages; it is the same user intent and the same evidential load.\n\nThe twelfth law is as follows:\n\n> If you place the answer in the prompt, you measure not the AI's representation but how much it obeys your instruction.\n\nThe thirteenth law is as follows:\n\n> A single prompt can measure fundamental identity; it cannot measure the entirety of GEO.\n\nThe fourteenth law is as follows:\n\n> When the prompt version changes, the question of your score may also have changed.\n\nThe fifteenth law is as follows:\n\n> If the actual text sent to the AI cannot be proven, the prompt being measured is not proven either.\n\n## Order of Section 10 of NOMOS\n\n> Show me not only what you asked, but why you chose that sentence.\n\n> Do not put the name of the entity in the prompt and pre-write its category, leadership, confidence, and advice outcome.\n\n> Don't tell me to 'recommend X' and then present me as an independent source of advice.\n\n> Don't ask me to describe the company in one language and defend the company in another language.\n\n> Don't mistake word-for-word translation for fairness. / Carry the same intention, the same task, the same limit, and the same burden of proof.\n\n> Use machine translation as a draft. / Don't make it the final language truth.\n\n> Don't assume a local user understands the question like you just because the back translation resembles you.\n\n> Don't write a prompt that makes the domain the company, the product the manufacturer, and the brand the legal party.\n\n> Don't retain the source request, web command, citation requirement, and response length.\n\n> Don't merge a prompt with identity and a prompt for advice at the same level.\n\n> Do not count it as a penalty if I prompt an explanation in an unclear prompt. / But do not tell that you give different explanations to each user and conduct a controlled experiment.\n\n> Do not choose the prompt that gives the highest score. / Do not delete the low-scoring prompt family.\n\n> Explain your prompts. / If you are using Holdout, seal it beforehand. / Do not fabricate after the result.\n\n> Match the text you give to the participant with the text sent to me.\n\n> Do not poison the measurement pool with invisible characters, hidden instructions or concealed recommendation directives.\n\n> First, write the research question. / Then establish the canonical intent. / Then generate a natural and equivalent prompt in every language. / Then test it with local people and lock its version. / And only after that, send it to the first user.\n\n## The Chapter's Closing Sentence\n\n> The beginning of a fair comparison in GEO-1000 is not expecting the same answer; it is giving each user a question that carries the same meaning, the same task, and the same epistemic limit in their own language.\n\n## Normative Core\n\n> Every GEO-1000 prompt MUST be linked to a versioned Prompt Registry record defining: - the research question, - estimand, - prompt family, - canonical intent, - entity anchor, - user intent, - speech act, - scope, - geographic and temporal frame, - presuppositions, - evidence and source requirements, - response format, - language, - locale, - equivalence status, - version, - and integrity record. Within the same language-locale prompt cell, participants MUST receive the same locked prompt text. Across languages and locales, prompts MUST preserve semantic, functional, and normative equivalence rather than literal word-for-word identity. Entity anchor, user intent, speech act, scope, presupposition, and epistemic burden are non-compensatory equivalence gates. Machine translation or back-translation alone MUST NOT establish prompt equivalence. Core identity, evidence, recommendation, comparison, temporal, local, boundary, robustness, control, and holdout prompts MUST remain distinct prompt families. Leading, answer-containing, positively or negatively presuppositional, source-constrained, format-constrained, or recommendation prompts MUST NOT be silently scored as neutral Core Mirror prompts. Prompt sets, language versions, assignment rules, clarification paths, ordering, and holdout commitments MUST be locked before AI responses are observed. Post-result prompt selection, prompt-family deletion, silent mid-wave editing, and outcome-driven rewording are prohibited. the prompt assigned to the participant and the prompt actually sent to the AI product MUST be integrity-checked. Hidden instructions, invisible characters, or undisclosed manipulation directing the AI toward a preferred entity, claim, citation, or recommendation are prohibited. Every prompt change, translation decision, equivalence judgement, and registry record MUST be versioned and attributable to an accountable human or organisation.","character_count":81568,"record_sha256":"038b95b74cc11ded8746555e5159943e3df145441b2f031a53197dee6caf0df0"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-11","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":13,"chapter_number":11,"item_number":null,"title":"The Controlled Clean Panel and the Natural User Panel","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K02","K03","K04","K06","K22"],"source_word_count":10223,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":11,\"id\":\"NOMOS-GEO-AUDIT-CH11\",\"title\":\"The Controlled Clean Panel and the Natural User Panel\",\"subtitle\":null,\"sourceFile\":\"11.ci bölüm.docx\",\"sourceSha256\":\"15AFA6E4096F8D45EA0D05DBB82A3A16DBCF99E2D807717B9A77B2CFD8C7D5A7\",\"sourceWordCount\":10223,\"sourceIds\":[\"K02\",\"K03\",\"K04\",\"K06\",\"K22\"],\"machine\":{\"chapter\":11,\"chapterId\":\"NOMOS-GEO-AUDIT-CH11\",\"title\":\"The Controlled Clean Panel and the Natural User Panel\",\"subtitle\":null,\"sourceIds\":[\"K02\",\"K03\",\"K04\",\"K06\",\"K22\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH11-R01\",\"normativeRuleEnglish\":\"Controlled Clean Panel and Natural User Panel observations MUST remain separate measurement populations, panel states, denominators, and public results. A Controlled Clean Panel MUST define and verify, as applicable: - a new and separate conversation, - absence of prior conversation context, - persistent-memory state, - custom-instruction state, - enterprise or managed-workspace state, - web, retrieval, and tool configuration, - plan, - user surface, - prompt version, - and clean-state confidence. A new conversation or temporary-session label MUST NOT, by itself, be treated as proof that all persistent personalisation has been disabled. A Natural User Panel MUST preserve the participant's authentic account, plan, interface, memory, custom-instruction, and usage state while retaining the locked prompt, first-eligible-output rule, evidence requirements, and non-regeneration rule. Natural Account-State, Natural Conversation-State, First-Use, and Managed Natural observations MUST remain separately identified. Natural-state documentation MUST minimise private-data collection. Private instruction text, conversation history, identity, credentials, and sensitive personal content MUST NOT be collected unless separately necessary, consented, minimised, and access-controlled. User-generated entity-favourable or entity-adverse instructions MAY be measured as natural personalisation effects, but MUST NOT be represented as general entity GEO success, failure, or default product behaviour. Personalisation MAY change relevance, presentation, and recommendation where user context justifies it. It MUST NOT arbitrarily change verified entity identity, license, legal status, price, scope, evidence, or time. Differences between Controlled and Natural panels MUST NOT be represented as causal personalisation effects unless population, product, prompt, time, interface, plan, tools, assignment, order, and carryover conditions support that inference. Controlled and Natural scores, critical-error rates, unknown rates, coverage, effective sample sizes, and signed and absolute panel gaps MUST remain separately visible. Panel-state eligibility, reclassification, exclusion, and weighting MUST remain independent of whether the AI output is positive, negative, correct, incorrect, refused, critical, or commercially favourable. Every panel-state definition, setting-change process, privacy decision, comparison level, and result MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Kontrollü Temiz Panel ve Doğal Kullanıcı Paneli gözlemleri ayrı ölçüm nüfusları, panel durumları, paydalar ve kamu sonuçları olarak korunmalıdır. Yeni sohbet veya geçici oturum etiketi tek başına bütün kalıcı kişiselleştirmenin devre dışı olduğunu kanıtlamaz. Doğal Kullanıcı Paneli gerçek hesap, plan, arayüz, hafıza, özel talimat ve kullanım durumunu korurken kilitlenmiş promptu, ilk uygun çıktı kuralını, kanıt gereksinimlerini ve yenilememe kuralını korumalıdır. Kullanıcı kaynaklı lehte veya aleyhte yönlendirme doğal kişiselleştirme etkisi olarak ölçülebilir; genel varlık GEO başarısı, başarısızlığı veya varsayılan ürün davranışı olarak sunulamaz. Kontrollü ve doğal sonuçlar arasındaki fark, nüfus ve sistem koşulları yeterince eşleşmeden nedensel kişiselleştirme etkisi olarak yorumlanamaz.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH11-MB0001\",\"type\":\"paragraph\",\"text\":\"77. MACHINE-READABLE PANEL STATUS RECORD\",\"sourceParagraph\":1981},{\"blockId\":\"CH11-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1982},{\"blockId\":\"CH11-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1983},{\"blockId\":\"CH11-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1984},{\"blockId\":\"CH11-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1985},{\"blockId\":\"CH11-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateRecord\\\": {\",\"sourceParagraph\":1986},{\"blockId\":\"CH11-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateId\\\": \\\"NGPS-SYNTH-ORION-TR-W1-001\\\",\",\"sourceParagraph\":1987},{\"blockId\":\"CH11-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"W1\\\",\",\"sourceParagraph\":1988},{\"blockId\":\"CH11-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductRecordId\\\": \\\"NGAI-SYNTH-ORION-WEB-PAID-W1-002\\\",\",\"sourceParagraph\":1989},{\"blockId\":\"CH11-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"promptRecordId\\\": \\\"NGP-CORE-MIRROR-APPLE-SYNTH-001\\\",\",\"sourceParagraph\":1990},{\"blockId\":\"CH11-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"panelRole\\\": \\\"CONTROLLED_CLEAN\\\",\",\"sourceParagraph\":1991},{\"blockId\":\"CH11-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"panelEligibility\\\": \\\"PC-1\\\",\",\"sourceParagraph\":1992},{\"blockId\":\"CH11-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"targetPopulation\\\": {\",\"sourceParagraph\":1993},{\"blockId\":\"CH11-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"frameId\\\": \\\"NGP-SYNTH-ORION-W1-001\\\",\",\"sourceParagraph\":1994},{\"blockId\":\"CH11-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"populationType\\\": \\\"PRODUCT_ELIGIBLE_USERS_UNDER_STANDARDIZED_STATE\\\"\",\"sourceParagraph\":1995},{\"blockId\":\"CH11-MB0016\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1996},{\"blockId\":\"CH11-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"controlledState\\\": {\",\"sourceParagraph\":1997},{\"blockId\":\"CH11-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"cleanStateConfidence\\\": \\\"CSC-4\\\",\",\"sourceParagraph\":1998},{\"blockId\":\"CH11-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"conversationState\\\": \\\"NEW_CONVERSATION\\\",\",\"sourceParagraph\":1999},{\"blockId\":\"CH11-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"promptIsFirstUserMessage\\\": true,\",\"sourceParagraph\":2000},{\"blockId\":\"CH11-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"persistentMemory\\\": \\\"OFF_VERIFIED\\\",\",\"sourceParagraph\":2001},{\"blockId\":\"CH11-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"customInstructions\\\": \\\"OFF_VERIFIED\\\",\",\"sourceParagraph\":2002},{\"blockId\":\"CH11-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"priorConversationContext\\\": false,\",\"sourceParagraph\":2003},{\"blockId\":\"CH11-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"enterpriseWorkspace\\\": false,\",\"sourceParagraph\":2004},{\"blockId\":\"CH11-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"webMode\\\": \\\"ON_STANDARDIZED\\\",\",\"sourceParagraph\":2005},{\"blockId\":\"CH11-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"toolMode\\\": \\\"STANDARDISED\\\",\",\"sourceParagraph\":2006},{\"blockId\":\"CH11-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"plan\\\": \\\"PAID\\\",\",\"sourceParagraph\":2007},{\"blockId\":\"CH11-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"surface\\\": \\\"WEB\\\",\",\"sourceParagraph\":2008},{\"blockId\":\"CH11-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"settingsChangedForStudy\\\": true,\",\"sourceParagraph\":2009},{\"blockId\":\"CH11-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"participantConsentForSettingChange\\\": true,\",\"sourceParagraph\":2010},{\"blockId\":\"CH11-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"restorationRequired\\\": true,\",\"sourceParagraph\":2011},{\"blockId\":\"CH11-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"restorationConfirmed\\\": true\",\"sourceParagraph\":2012},{\"blockId\":\"CH11-MB0033\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2013},{\"blockId\":\"CH11-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"measurement\\\": {\",\"sourceParagraph\":2014},{\"blockId\":\"CH11-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"firstEligibleOutputRequired\\\": true,\",\"sourceParagraph\":2015},{\"blockId\":\"CH11-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"responseRegenerationAllowed\\\": false,\",\"sourceParagraph\":2016},{\"blockId\":\"CH11-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"followUpAllowed\\\": false,\",\"sourceParagraph\":2017},{\"blockId\":\"CH11-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"promptValidityRequired\\\": \\\"PV-1\\\",\",\"sourceParagraph\":2018},{\"blockId\":\"CH11-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"timestampUtc\\\": \\\"2026-08-18T12:00:00Z\\\"\",\"sourceParagraph\":2019},{\"blockId\":\"CH11-MB0040\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2020},{\"blockId\":\"CH11-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"privacy\\\": {\",\"sourceParagraph\":2021},{\"blockId\":\"CH11-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"privateInstructionContentCollected\\\": false,\",\"sourceParagraph\":2022},{\"blockId\":\"CH11-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"conversationHistoryCollected\\\": false,\",\"sourceParagraph\":2023},{\"blockId\":\"CH11-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"accountCredentialsCollected\\\": false\",\"sourceParagraph\":2024},{\"blockId\":\"CH11-MB0045\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2025},{\"blockId\":\"CH11-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2026},{\"blockId\":\"CH11-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateOwnerRole\\\": \\\"PANEL_STATE_METHODS_LEAD\\\",\",\"sourceParagraph\":2027},{\"blockId\":\"CH11-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"validatedByRole\\\": \\\"BLINDED_PANEL_STATE_REVIEWER\\\"\",\"sourceParagraph\":2028},{\"blockId\":\"CH11-MB0049\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2029},{\"blockId\":\"CH11-MB0050\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2030},{\"blockId\":\"CH11-MB0051\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2031},{\"blockId\":\"CH11-MB0052\",\"type\":\"paragraph\",\"text\":\"78. MACHINE-READABLE NATURAL PANEL RECORD\",\"sourceParagraph\":2033},{\"blockId\":\"CH11-MB0053\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2034},{\"blockId\":\"CH11-MB0054\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2035},{\"blockId\":\"CH11-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2036},{\"blockId\":\"CH11-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2037},{\"blockId\":\"CH11-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateRecord\\\": {\",\"sourceParagraph\":2038},{\"blockId\":\"CH11-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateId\\\": \\\"NGPS-SYNTH-ORION-TR-W1-002\\\",\",\"sourceParagraph\":2039},{\"blockId\":\"CH11-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"W1\\\",\",\"sourceParagraph\":2040},{\"blockId\":\"CH11-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductRecordId\\\": \\\"NGAI-SYNTH-ORION-WEB-PAID-W1-002\\\",\",\"sourceParagraph\":2041},{\"blockId\":\"CH11-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"promptRecordId\\\": \\\"NGP-CORE-MIRROR-APPLE-SYNTH-001\\\",\",\"sourceParagraph\":2042},{\"blockId\":\"CH11-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"panelRole\\\": \\\"NATURAL_ACCOUNT_STATE\\\",\",\"sourceParagraph\":2043},{\"blockId\":\"CH11-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"panelEligibility\\\": \\\"PC-2\\\",\",\"sourceParagraph\":2044},{\"blockId\":\"CH11-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"targetPopulation\\\": {\",\"sourceParagraph\":2045},{\"blockId\":\"CH11-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"frameId\\\": \\\"NGP-SYNTH-ORION-ACTIVE-USERS-W1-002\\\",\",\"sourceParagraph\":2046},{\"blockId\":\"CH11-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"populationType\\\": \\\"DEFINED_ACTIVE_PRODUCT_USERS\\\"\",\"sourceParagraph\":2047},{\"blockId\":\"CH11-MB0067\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2048},{\"blockId\":\"CH11-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"naturalState\\\": {\",\"sourceParagraph\":2049},{\"blockId\":\"CH11-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"naturalStateConfidence\\\": \\\"NSF-3\\\",\",\"sourceParagraph\":2050},{\"blockId\":\"CH11-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"conversationState\\\": \\\"NEW_CONVERSATION\\\",\",\"sourceParagraph\":2051},{\"blockId\":\"CH11-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"promptIsFirstUserMessage\\\": true,\",\"sourceParagraph\":2052},{\"blockId\":\"CH11-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"persistentMemory\\\": \\\"MS-4\\\",\",\"sourceParagraph\":2053},{\"blockId\":\"CH11-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"customInstructions\\\": \\\"CI-3\\\",\",\"sourceParagraph\":2054},{\"blockId\":\"CH11-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"priorConversationContext\\\": false,\",\"sourceParagraph\":2055},{\"blockId\":\"CH11-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"priorEntityConversation\\\": \\\"POSSIBLE\\\",\",\"sourceParagraph\":2056},{\"blockId\":\"CH11-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"enterpriseWorkspace\\\": false,\",\"sourceParagraph\":2057},{\"blockId\":\"CH11-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"webMode\\\": \\\"USER_NATURAL_STATE\\\",\",\"sourceParagraph\":2058},{\"blockId\":\"CH11-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"toolMode\\\": \\\"USER_NATURAL_STATE\\\",\",\"sourceParagraph\":2059},{\"blockId\":\"CH11-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"plan\\\": \\\"PAID\\\",\",\"sourceParagraph\":2060},{\"blockId\":\"CH11-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"surface\\\": \\\"WEB\\\",\",\"sourceParagraph\":2061},{\"blockId\":\"CH11-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"accountAgeBand\\\": \\\"181_TO_365_DAYS\\\",\",\"sourceParagraph\":2062},{\"blockId\":\"CH11-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"usageFrequency\\\": \\\"WEEKLY\\\",\",\"sourceParagraph\":2063},{\"blockId\":\"CH11-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"firstUse\\\": false\",\"sourceParagraph\":2064},{\"blockId\":\"CH11-MB0084\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2065},{\"blockId\":\"CH11-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"measurement\\\": {\",\"sourceParagraph\":2066},{\"blockId\":\"CH11-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"firstEligibleOutputRequired\\\": true,\",\"sourceParagraph\":2067},{\"blockId\":\"CH11-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"responseRegenerationAllowed\\\": false,\",\"sourceParagraph\":2068},{\"blockId\":\"CH11-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"followUpAllowed\\\": false,\",\"sourceParagraph\":2069},{\"blockId\":\"CH11-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"promptValidityRequired\\\": \\\"PV-1\\\",\",\"sourceParagraph\":2070},{\"blockId\":\"CH11-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"timestampUtc\\\": \\\"2026-08-18T12:05:00Z\\\"\",\"sourceParagraph\":2071},{\"blockId\":\"CH11-MB0091\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2072},{\"blockId\":\"CH11-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"privacy\\\": {\",\"sourceParagraph\":2073},{\"blockId\":\"CH11-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"memoryContentCollected\\\": false,\",\"sourceParagraph\":2074},{\"blockId\":\"CH11-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"customInstructionContentCollected\\\": false,\",\"sourceParagraph\":2075},{\"blockId\":\"CH11-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"conversationHistoryCollected\\\": false,\",\"sourceParagraph\":2076},{\"blockId\":\"CH11-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"highLevelStateOnly\\\": true\",\"sourceParagraph\":2077},{\"blockId\":\"CH11-MB0097\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2078},{\"blockId\":\"CH11-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2079},{\"blockId\":\"CH11-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateOwnerRole\\\": \\\"PANEL_STATE_METHODS_LEAD\\\",\",\"sourceParagraph\":2080},{\"blockId\":\"CH11-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"validatedByRole\\\": \\\"BLINDED_PANEL_STATE_REVIEWER\\\"\",\"sourceParagraph\":2081},{\"blockId\":\"CH11-MB0101\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2082},{\"blockId\":\"CH11-MB0102\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2083},{\"blockId\":\"CH11-MB0103\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2084},{\"blockId\":\"CH11-MB0104\",\"type\":\"paragraph\",\"text\":\"79. MACHINE-READABLE PANEL COMPARISON RECORD\",\"sourceParagraph\":2086},{\"blockId\":\"CH11-MB0105\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2087},{\"blockId\":\"CH11-MB0106\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2088},{\"blockId\":\"CH11-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2089},{\"blockId\":\"CH11-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2090},{\"blockId\":\"CH11-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"panelComparison\\\": {\",\"sourceParagraph\":2091},{\"blockId\":\"CH11-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonId\\\": \\\"NGPC-SYNTH-ORION-W1-001\\\",\",\"sourceParagraph\":2092},{\"blockId\":\"CH11-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"controlledPanelRecord\\\": \\\"NGPS-SYNTH-ORION-TR-W1-001\\\",\",\"sourceParagraph\":2093},{\"blockId\":\"CH11-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"naturalPanelRecord\\\": \\\"NGPS-SYNTH-ORION-TR-W1-002\\\",\",\"sourceParagraph\":2094},{\"blockId\":\"CH11-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"comparabilityLevel\\\": \\\"CN-3\\\",\",\"sourceParagraph\":2095},{\"blockId\":\"CH11-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"matchedDimensions\\\": {\",\"sourceParagraph\":2096},{\"blockId\":\"CH11-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"aiProduct\\\": true,\",\"sourceParagraph\":2097},{\"blockId\":\"CH11-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"promptVersion\\\": true,\",\"sourceParagraph\":2098},{\"blockId\":\"CH11-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"entityTruthPack\\\": true,\",\"sourceParagraph\":2099},{\"blockId\":\"CH11-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"country\\\": true,\",\"sourceParagraph\":2100},{\"blockId\":\"CH11-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": true,\",\"sourceParagraph\":2101},{\"blockId\":\"CH11-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"measurementWindow\\\": true,\",\"sourceParagraph\":2102},{\"blockId\":\"CH11-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationRules\\\": true\",\"sourceParagraph\":2103},{\"blockId\":\"CH11-MB0122\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2104},{\"blockId\":\"CH11-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"unmatchedDimensions\\\": {\",\"sourceParagraph\":2105},{\"blockId\":\"CH11-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"memory\\\": true,\",\"sourceParagraph\":2106},{\"blockId\":\"CH11-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"customInstructions\\\": true,\",\"sourceParagraph\":2107},{\"blockId\":\"CH11-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"planDistribution\\\": false,\",\"sourceParagraph\":2108},{\"blockId\":\"CH11-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"userHistory\\\": true,\",\"sourceParagraph\":2109},{\"blockId\":\"CH11-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"toolState\\\": true\",\"sourceParagraph\":2110},{\"blockId\":\"CH11-MB0129\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2111},{\"blockId\":\"CH11-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"results\\\": {\",\"sourceParagraph\":2112},{\"blockId\":\"CH11-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"controlledRepresentationScore\\\": 0.92,\",\"sourceParagraph\":2113},{\"blockId\":\"CH11-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"naturalRepresentationScore\\\": 0.83,\",\"sourceParagraph\":2114},{\"blockId\":\"CH11-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"signedNaturalGap\\\": -0.09,\",\"sourceParagraph\":2115},{\"blockId\":\"CH11-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"absolutePanelGap\\\": 0.09,\",\"sourceParagraph\":2116},{\"blockId\":\"CH11-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"controlledCriticalRate\\\": 0.002,\",\"sourceParagraph\":2117},{\"blockId\":\"CH11-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"naturalCriticalRate\\\": 0.012,\",\"sourceParagraph\":2118},{\"blockId\":\"CH11-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"criticalRateDifference\\\": 0.01\",\"sourceParagraph\":2119},{\"blockId\":\"CH11-MB0138\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2120},{\"blockId\":\"CH11-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"allowedClaims\\\": [\",\"sourceParagraph\":2121},{\"blockId\":\"CH11-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"NATURAL_PANEL_RESULT_LOWER_THAN_CONTROLLED_REFERENCE\\\",\",\"sourceParagraph\":2122},{\"blockId\":\"CH11-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"PANEL_STATE_ASSOCIATION_REQUIRES_FURTHER_DECOMPOSITION\\\"\",\"sourceParagraph\":2123},{\"blockId\":\"CH11-MB0142\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2124},{\"blockId\":\"CH11-MB0143\",\"type\":\"paragraph\",\"text\":\"\\\"prohibitedClaims\\\": [\",\"sourceParagraph\":2125},{\"blockId\":\"CH11-MB0144\",\"type\":\"paragraph\",\"text\":\"\\\"MEMORY_CAUSED_NINE_POINT_DECLINE\\\",\",\"sourceParagraph\":2126},{\"blockId\":\"CH11-MB0145\",\"type\":\"paragraph\",\"text\":\"\\\"ALL_NATURAL_USERS_RECEIVE_WORSE_RESPONSES\\\",\",\"sourceParagraph\":2127},{\"blockId\":\"CH11-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"BASE_MODEL_PERFORMANCE_DECLINED\\\"\",\"sourceParagraph\":2128},{\"blockId\":\"CH11-MB0147\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2129},{\"blockId\":\"CH11-MB0148\",\"type\":\"paragraph\",\"text\":\"\\\"replicationStatus\\\": \\\"SINGLE_WAVE\\\",\",\"sourceParagraph\":2130},{\"blockId\":\"CH11-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2131},{\"blockId\":\"CH11-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"comparisonOwnerRole\\\": \\\"PANEL_COMPARABILITY_ADJUDICATOR\\\"\",\"sourceParagraph\":2132},{\"blockId\":\"CH11-MB0151\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2133},{\"blockId\":\"CH11-MB0152\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2134},{\"blockId\":\"CH11-MB0153\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2135},{\"blockId\":\"CH11-MB0154\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2136},{\"blockId\":\"CH11-MB0155\",\"type\":\"paragraph\",\"text\":\"are not:\",\"sourceParagraph\":2137},{\"blockId\":\"CH11-MB0156\",\"type\":\"paragraph\",\"text\":\"a real user,\",\"sourceParagraph\":2138},{\"blockId\":\"CH11-MB0157\",\"type\":\"paragraph\",\"text\":\"a real AI product,\",\"sourceParagraph\":2139},{\"blockId\":\"CH11-MB0158\",\"type\":\"paragraph\",\"text\":\"a real company performance.\",\"sourceParagraph\":2140},{\"blockId\":\"CH11-MB0159\",\"type\":\"paragraph\",\"text\":\"It is a machine-readable synthetic representation of the panel distinction.\",\"sourceParagraph\":2141},{\"blockId\":\"CH11-MB0160\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RULE OF SECTION 80\",\"sourceParagraph\":2143},{\"blockId\":\"CH11-MB0161\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH11-R01\",\"sourceParagraph\":2144},{\"blockId\":\"CH11-MB0162\",\"type\":\"paragraph\",\"text\":\"Controlled Clean Panel and Natural User Panel observations MUST remain\",\"sourceParagraph\":2146},{\"blockId\":\"CH11-MB0163\",\"type\":\"paragraph\",\"text\":\"separate measurement populations, panel states, denominators, and public\",\"sourceParagraph\":2147},{\"blockId\":\"CH11-MB0164\",\"type\":\"paragraph\",\"text\":\"results.\",\"sourceParagraph\":2148},{\"blockId\":\"CH11-MB0165\",\"type\":\"paragraph\",\"text\":\"A Controlled Clean Panel MUST define and verify, as applicable:\",\"sourceParagraph\":2150},{\"blockId\":\"CH11-MB0166\",\"type\":\"paragraph\",\"text\":\"- a new and separate conversation,\",\"sourceParagraph\":2152},{\"blockId\":\"CH11-MB0167\",\"type\":\"paragraph\",\"text\":\"- absence of prior conversation context,\",\"sourceParagraph\":2153},{\"blockId\":\"CH11-MB0168\",\"type\":\"paragraph\",\"text\":\"- persistent-memory state,\",\"sourceParagraph\":2154},{\"blockId\":\"CH11-MB0169\",\"type\":\"paragraph\",\"text\":\"- custom-instruction state,\",\"sourceParagraph\":2155},{\"blockId\":\"CH11-MB0170\",\"type\":\"paragraph\",\"text\":\"- enterprise or managed-workspace state,\",\"sourceParagraph\":2156},{\"blockId\":\"CH11-MB0171\",\"type\":\"paragraph\",\"text\":\"- web, retrieval, and tool configuration,\",\"sourceParagraph\":2157},{\"blockId\":\"CH11-MB0172\",\"type\":\"paragraph\",\"text\":\"- plan,\",\"sourceParagraph\":2158},{\"blockId\":\"CH11-MB0173\",\"type\":\"paragraph\",\"text\":\"- user surface,\",\"sourceParagraph\":2159},{\"blockId\":\"CH11-MB0174\",\"type\":\"paragraph\",\"text\":\"- prompt version,\",\"sourceParagraph\":2160},{\"blockId\":\"CH11-MB0175\",\"type\":\"paragraph\",\"text\":\"- and clean-state confidence.\",\"sourceParagraph\":2161},{\"blockId\":\"CH11-MB0176\",\"type\":\"paragraph\",\"text\":\"A new conversation or temporary-session label MUST NOT, by itself, be\",\"sourceParagraph\":2163},{\"blockId\":\"CH11-MB0177\",\"type\":\"paragraph\",\"text\":\"treated as proof that all persistent personalisation has been disabled.\",\"sourceParagraph\":2164},{\"blockId\":\"CH11-MB0178\",\"type\":\"paragraph\",\"text\":\"A Natural User Panel MUST preserve the participant's authentic account,\",\"sourceParagraph\":2166},{\"blockId\":\"CH11-MB0179\",\"type\":\"paragraph\",\"text\":\"plan, interface, memory, custom-instruction, and usage state while\",\"sourceParagraph\":2167},{\"blockId\":\"CH11-MB0180\",\"type\":\"paragraph\",\"text\":\"retaining the locked prompt, first-eligible-output rule, evidence\",\"sourceParagraph\":2168},{\"blockId\":\"CH11-MB0181\",\"type\":\"paragraph\",\"text\":\"requirements, and non-regeneration rule.\",\"sourceParagraph\":2169},{\"blockId\":\"CH11-MB0182\",\"type\":\"paragraph\",\"text\":\"Natural Account-State, Natural Conversation-State, First-Use, and Managed\",\"sourceParagraph\":2171},{\"blockId\":\"CH11-MB0183\",\"type\":\"paragraph\",\"text\":\"Natural observations MUST remain separately identified.\",\"sourceParagraph\":2172},{\"blockId\":\"CH11-MB0184\",\"type\":\"paragraph\",\"text\":\"Natural-state documentation MUST minimise private-data collection.\",\"sourceParagraph\":2174},{\"blockId\":\"CH11-MB0185\",\"type\":\"paragraph\",\"text\":\"Private instruction text, conversation history, identity, credentials,\",\"sourceParagraph\":2175},{\"blockId\":\"CH11-MB0186\",\"type\":\"paragraph\",\"text\":\"and sensitive personal content MUST NOT be collected unless separately\",\"sourceParagraph\":2176},{\"blockId\":\"CH11-MB0187\",\"type\":\"paragraph\",\"text\":\"necessary, consented, minimised, and access-controlled.\",\"sourceParagraph\":2177},{\"blockId\":\"CH11-MB0188\",\"type\":\"paragraph\",\"text\":\"User-generated entity-favourable or entity-adverse instructions MAY be\",\"sourceParagraph\":2179},{\"blockId\":\"CH11-MB0189\",\"type\":\"paragraph\",\"text\":\"measured as natural personalisation effects, but MUST NOT be represented\",\"sourceParagraph\":2180},{\"blockId\":\"CH11-MB0190\",\"type\":\"paragraph\",\"text\":\"as general entity GEO success, failure, or default product behaviour.\",\"sourceParagraph\":2181},{\"blockId\":\"CH11-MB0191\",\"type\":\"paragraph\",\"text\":\"Personalisation MAY change relevance, presentation, and recommendation\",\"sourceParagraph\":2183},{\"blockId\":\"CH11-MB0192\",\"type\":\"paragraph\",\"text\":\"where user context justifies it. It MUST NOT arbitrarily change verified\",\"sourceParagraph\":2184},{\"blockId\":\"CH11-MB0193\",\"type\":\"paragraph\",\"text\":\"entity identity, license, legal status, price, scope, evidence, or time.\",\"sourceParagraph\":2185},{\"blockId\":\"CH11-MB0194\",\"type\":\"paragraph\",\"text\":\"Differences between Controlled and Natural panels MUST NOT be represented\",\"sourceParagraph\":2187},{\"blockId\":\"CH11-MB0195\",\"type\":\"paragraph\",\"text\":\"as causal personalisation effects unless population, product, prompt,\",\"sourceParagraph\":2188},{\"blockId\":\"CH11-MB0196\",\"type\":\"paragraph\",\"text\":\"time, interface, plan, tools, assignment, order, and carryover conditions\",\"sourceParagraph\":2189},{\"blockId\":\"CH11-MB0197\",\"type\":\"paragraph\",\"text\":\"support that inference.\",\"sourceParagraph\":2190},{\"blockId\":\"CH11-MB0198\",\"type\":\"paragraph\",\"text\":\"Controlled and Natural scores, critical-error rates, unknown rates,\",\"sourceParagraph\":2192},{\"blockId\":\"CH11-MB0199\",\"type\":\"paragraph\",\"text\":\"coverage, effective sample sizes, and signed and absolute panel gaps MUST\",\"sourceParagraph\":2193},{\"blockId\":\"CH11-MB0200\",\"type\":\"paragraph\",\"text\":\"remain separately visible.\",\"sourceParagraph\":2194},{\"blockId\":\"CH11-MB0201\",\"type\":\"paragraph\",\"text\":\"Panel-state eligibility, reclassification, exclusion, and weighting MUST\",\"sourceParagraph\":2196},{\"blockId\":\"CH11-MB0202\",\"type\":\"paragraph\",\"text\":\"remain independent of whether the AI output is positive, negative,\",\"sourceParagraph\":2197},{\"blockId\":\"CH11-MB0203\",\"type\":\"paragraph\",\"text\":\"correct, incorrect, refused, critical, or commercially favourable.\",\"sourceParagraph\":2198},{\"blockId\":\"CH11-MB0204\",\"type\":\"paragraph\",\"text\":\"Every panel-state definition, setting-change process, privacy decision,\",\"sourceParagraph\":2200},{\"blockId\":\"CH11-MB0205\",\"type\":\"paragraph\",\"text\":\"comparison level, and result MUST be versioned and attributable to an\",\"sourceParagraph\":2201},{\"blockId\":\"CH11-MB0206\",\"type\":\"paragraph\",\"text\":\"accountable human or organisation.\",\"sourceParagraph\":2202},{\"blockId\":\"CH11-MB0207\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2203},{\"blockId\":\"CH11-MB0208\",\"type\":\"paragraph\",\"text\":\"Observations of the Controlled Clean Panel and the Natural User Panel should be maintained as separate measurement populations, panel states, denominators, and public results. A new chat or temporary session tag alone does not prove that all persistent personalisation is disabled. The Natural User Panel should retain the real account, plan, interface, memory, custom instructions, and usage scenario while preserving the locked prompt, first appropriate output rule, evidence requirements, and non-refresh rule. User-driven favourable or unfavorable guidance can be measured as the effect of natural personalisation; it cannot be presented as general entity GEO success, failure, or default product behaviour. The difference between controlled and natural results cannot be interpreted as a causal personalisation effect without sufficiently matching populations and system conditions.\",\"sourceParagraph\":2204}]}}","text":"## Chapter Boundary\n\nChapter 9 showed that the measured AI object cannot be defined solely by the provider or model name. Chapter 10 established that not every question containing the same entity name is the same prompt; the prompt is the experimental tool that determines the measurement itself. Now we need to address the third variable between the system and the prompt:\n\n> What is the state of the account, session, and user history using the AI product?\n\nA user:\n\n- new chat,\n\n- memory is off,\n\n- no special instructions,\n\n- no previous context,\n\n- standardised product settings\n\nmay be sending the prompt underneath. Another user can send the same prompt:\n\n- from an account they have been using for years,\n\n- memory is on,\n\n- with special instructions enabled,\n\n- with past conversations,\n\n- on their real plan and device\n\nThey can use the same AI product, in the same language, in the same country, in the same time frame, and with the same prompt. Yet they may see different responses. This difference:\n\n- from the random variability of the product,\n\n- from the account's plan,\n\n- from memory,\n\n- from special instructions,\n\n- from past conversations,\n\n- from personalisation,\n\n- From the web or tool status,\n\n- from the corporate workspace,\n\n- from the user surface and local settings\n\nThe difference may arise from personalisation, web or tool state, the organisational workspace, the user surface or locale. There is no single ‘AI’: search, retrieval, reranking, language-model, citation and safety layers may behave differently. Reliable measurement must therefore record the AI product or model, date, country, language, query set, repetition count and observation. This chapter adds user and session state to that chain. NOMOS's core demand remains evidence, boundary, context and time. Yet context itself can be measured in two ways. The first minimises external differences between users and compares AI products under standardised initial conditions. This is the:\n\n#### Controlled Clean Panel\n\nThe second approach measures what people see through their real accounts, plans, personalisation and ordinary patterns of use. This is the:\n\n#### Natural User Panel\n\nis called. The Controlled Clean Panel answers the question:\n\n> How does the product present the entity in the most standardised initial condition possible?\n\nThe Natural User Panel, on the other hand, asks:\n\n> What do real people see about the entity in real user situations?\n\nOne panel is not more “real” than the other. They measure two different realities. The controlled panel strengthens comparability. The natural panel makes human experience visible. The controlled panel reduces personalisation and context differences. The natural panel measures the effect of these differences on the real population. This section:\n\n- Definition of the Controlled Clean Panel,\n\n- Definition of the Natural User Panel,\n\n- Target populations of the two panels and the quantities they estimate,\n\n- What a clean session means,\n\n- Levels of verification of cleanliness,\n\n- Natural account status and natural conversation context,\n\n- Memory, special instructions, previous conversation, and personalisation records,\n\n- account, plan, device, interface and locale differences,\n\n- the paired and independent panel designs with the same user,\n\n- the effects of order, transfer, and contamination,\n\n- clean–natural difference metrics,\n\n- privacy and the chain of evidence,\n\n- how the two panels will be separated in the public report\n\ndefines. This chapter does not yet:\n\n- the complete technical architecture of NOMOS Capture,\n\n- all fields of screenshot and metadata files,\n\n- atomic claim adjudication,\n\n- final response transition thresholds,\n\n- the exact weight of clean and natural panel scores in the combined NOMOS score\n\ndoes not finalise. The fundamental question of Section 11 is:\n\n> How do we measure the behaviour of a standard AI product and the behaviour experienced by real users without confusing them with each other?\n\n## NOMOS Challenge\n\nThe same user asks the same AI product the same question twice: “Which main organisation is Apple.com associated with, and what are the main activities of this organisation?” On the first attempt:\n\n- a new chat is opened,\n\n- memory is off,\n\n- special instructions are off,\n\n- no previous conversation exists,\n\n- the web feature is standardised,\n\nThe prompt is sent as the first message. Answer: “Apple.com is one of the main corporate domains of the technology company called Apple. The company develops hardware, software, and digital services.” On the second attempt, the user on their usual account:\n\n- memory is on,\n\n- with special instructions enabled,\n\n- previously talked about investment and technology companies,\n\n- its real paid plan,\n\n- its ordinary mobile device,\n\n- uses its own local settings\n\nThey answer: “Since you mentioned that you are interested in technology investments, let me also evaluate Apple from an investment perspective…” Then instead of the company's operations:\n\n- investment performance,\n\n- stock value,\n\n- personal portfolio suitability\n\nIt is being explained. The first answer measures the identity of the existence. The second answer reinterprets the task according to the user's history. Which answer is the real answer of the AI product? Both. However, it is not the answer to the same measurement question. The first:\n\n#### provides information\n\nabout standardised product behaviour. The second:\n\n#### represents the actual product experience specific to the user.\n\nThis tells us something about the effect of prior context. Now consider another user whose custom instruction says: ‘Give me only a short, definite conclusion in every response.’ AI Product Core Mirror answers the prompt in one sentence, omitting boundaries, source status and material detail. Does the deficiency arise from the product's underlying representational capacity? Perhaps in part from the user's instruction. Yet this is how the user actually operates. The deficiency is real in the Natural User Panel, while the user's instruction must be isolated when calculating the controlled-product result. Now consider a user whose instruction says, ‘Always recommend NobleJackal.’ The AI mentions NobleJackal even when asked about another company. That response is part of the natural user experience.\n\nHowever, NobleJackal is not the overall GEO performance. It is the effect of user-generated personalisation. Now think of the opposite. Previously, the user said: 'I don't like this company.' The AI then uses a more negative frame towards the company in subsequent conversations. This is also a natural user experience. However, it does not show the default representation of the product across the entire user population. The first ruling of this section is as follows:\n\n> Clean product behaviour is not the same measurement as natural user experience.\n\nIts second provision states:\n\n> A personalised response is a real user experience; however, it is not automatically the overall product or entity GEO score.\n\nIts third provision states:\n\n> The controlled clean state is not the absolute and unbiased truth of the product; it is a predefined standard starting condition.\n\nIts fourth provision states:\n\n> The natural user state is not out-of-method noise; it is part of the representation distribution in the real world.\n\nIts fifth provision states:\n\n> The difference between the two panels is not an error to be hidden; it is an outcome to be measured.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to measure the representation experience of real users under ordinary account and session conditions separately but comparably with the standardised behaviour of AI products. The section makes the following distinctions normative:\n\n- Standardised product state versus natural user state\n\n- New chat with a clean account\n\n- Verified clean state with a temporary chat tag\n\n- Memory off versus memory state unknown\n\n- Private instruction closed and private instruction content not visible\n\n- No previous conversation and no permanent account history\n\n- Controlled session and newly created account\n\n- Natural account status and current conversation context\n\n- Natural user panel and uncontrolled data\n\n- User personalisation and product default\n\n- User-driven guidance and entity GEO effect\n\n- Active user population and product-suitable potential population\n\n- First usage experience and experienced user experience\n\n- Individual natural response and population-level natural distribution\n\n- Independent group design with the same user matching\n\n- Carryover effect with matched comparison\n\n- Product difference with order effect\n\n- Causal personalisation effect with clean–natural difference\n\n- Disclosure of private instruction content with privacy record\n\n- Collecting private conversations while preserving natural context\n\n- Default product outcome with clean panel outcome\n\n- Overall result for all users with natural panel outcome\n\n- Personalisation effect with user plan\n\n- Controlled average with natural error queue\n\n- Same answer for each user with personalisation fairness\n\n- Natural user reality and test repeatability\n\nAt the end of this section, each GEO-1000 audit should be able to answer the following questions:\n\n> Which results came from the Controlled Clean Panel and which from the Natural User Panel?\n\n> Which situations were actually standardised in the clean panel?\n\n> Which real user differences were preserved in the natural panel?\n\n> How were memory, special instructions, past conversation, plans, and interface states recorded?\n\n> How much of the difference seen between the two panels can be reliably associated with personalisation?\n\n> How was the presence of natural context demonstrated while preserving user privacy?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEach GEO-1000 audit should define standardised product behaviour and natural user experience as separate prediction objects; it should not combine the Controlled Clean Panel and Natural User Panel results into a single sample, denominator, or score without explanation. To the extent relevant, the Controlled Clean Panel aims to standardise the following conditions:\n\n- New and separate conversation\n\n- Absence of previous conversation context\n\n- Persistent memory state turned off or disabled\n\n- Special instructions turned off or disabled\n\n- Defined web, retrieval, and tool state\n\n- Defined plan and user surface\n\n- Locked prompt\n\n- Standard language and locale\n\n- Measurement wave and time window\n\nIf it is a Natural User Panel, it preserves the following real differences to the extent relevant:\n\n- User's real account\n\n- Real plan\n\n- Real user surface\n\n- Real memory state\n\n- Real special instruction state\n\n- Real account age and usage history\n\n- Real device\n\n- Real locale and access condition\n\n- Ordinary way of using the product\n\nHowever, the natural panel is not without protocol either. The following are again locked:\n\n- Prompt version\n\n- Shipping method\n\n- Measurement time\n\n- First appropriate output rule\n\n- Evidence package\n\n- Participant and product eligibility\n\n- Do not alter the response\n\n- Do not send off-task follow-up messages\n\n- Personal data protection\n\n## 3. TWO SEPARATE ESTIMATION OBJECTS\n\nControlled and natural panels do not predict the same probability.\n\n### 3.1. Controlled Clean Representation Probability\n\nSpecifically:\n\n- AI product,\n\n- plan,\n\n- surface,\n\n- prompt,\n\n- country,\n\n- language,\n\n- time\n\nis the probability of seeing a valid representation under a standardised clean session for:\n\nθ^C_{E,A,P,U,t} = Pr(Y=1 | CleanState=1)\n\nThe target population here is the result that people suitable to use the relevant product would see under standardised product conditions.\n\n### 3.2. Probability of Natural User Representation\n\nIt is the probability of seeing valid representation in the real active or defined natural user population, while maintaining users' usual account and personalisation conditions.\n\nθ^N_{E,A,P,U,t} = Pr(Y=1 | NaturalState=1)\n\nThe target population here:\n\n- is not all product-eligible people,\n\n- but real or natural product users in the defined period\n\nmay be. These two populations may not be the same.\n\n### 3.3. Two outcomes cannot be used interchangeably\n\nControlled score: does not mean “Most real users saw this.” Natural score: does not mean “The product behaves this way by default.”\n\n## 4. WHAT IS A CONTROLLED CLEAN PANEL?\n\nA Controlled Clean Panel is a panel that measures the AI product’s behaviour under defined initial conditions by minimising the material differences caused by accounts and sessions among users. The purpose of the Controlled Clean Panel is to:\n\n- increase comparability across products,\n\n- reduce the effect of memory and special instructions,\n\n- prevent leakage from previous conversations,\n\n- make the prompt an initial and independent task,\n\nto produce a reference point closer to the fundamental product behaviour of the system. Controlled panel:\n\n- absolute model capability,\n\n- the real experience of all users,\n\n- completely context-free and unbiased AI response\n\ndoes not claim to be. Only:\n\n#### a predefined and verified clean starting state\n\nis produced.\n\n## 5. WHAT DOES “CLEAN” MEAN?\n\nA clean state may not be creatable with the same technical process on every product. A product:\n\n- persistent memory,\n\n- special instructions,\n\n- user profile,\n\n- past conversation,\n\n- corporate workspace,\n\n- account-based personalisation\n\ncan be used. Other products may not have some of these features. The functional definition of a clean state is as follows:\n\n> The initial state where the likelihood of user-specific factual information and conversation context affecting the response before the audit prompt is reduced to the extent allowed by the product, and the remaining unknowns are explicitly recorded.\n\nClean state:\n\n- simply opening a new chat,\n\n- refreshing the browser,\n\n- closing and reopening the application\n\nis not considered to occur automatically.\n\n## 6. CLEANLINESS LAYERS\n\nThe following layers should be evaluated separately in the control panel.\n\n### 6.1. Conversation Cleanliness\n\nThere should be no previous user or system message in the current chat. the prompt should be the first user message.\n\n### 6.2. Persistent Memory Clearing\n\nThe memory the product retains about the user across sessions:\n\n- off,\n\n- disabled,\n\n- not used in this session\n\nshould be. If full verification is not possible, an unknown state is recorded.\n\n### 6.3. Custom Instruction Clearing\n\nThe user's special instructions defined for answer style, preferences, or advising behaviour:\n\n- off,\n\n- temporarily inactive,\n\n- not present\n\nmust be.\n\n### 6.4. Corporate Context Cleaning\n\nCorporate workspace:\n\n- private documents,\n\n- company policies,\n\n- executive instructions,\n\n- private knowledge bases\n\nif used, should be separated from the controlled general product panel.\n\n### 6.5. Tool and Information Mode Cleaning\n\nThe web, retrieval, citation, and tool conditions should be set to the same configuration in all compared observations. Being clean does not mean that the tools are turned off. Being clean means that the tool status is predefined and comparable.\n\n### 6.6. Prompt Cleaning\n\nPrompt:\n\n- initial message,\n\n- full version,\n\n- unchanged,\n\n- without hidden instructions\n\nmust be.\n\n### 6.7. Measurement Cleanliness\n\nParticipant:\n\n- should not refresh the answer,\n\n- should not send a follow-up question,\n\n- should not choose the result,\n\nshould not carry another AI answer into the same session.\n\n## 7. CLEAN STATE CONFIDENCE LEVELS\n\n### CSC-0 — UNKNOWN CLEANLINESS\n\nThe cleanliness of the session and account could not be reliably determined. Cannot be included in controlled panel main analysis.\n\n### CSC-1 — NEW CONVERSATION ONLY\n\nOnly new conversation is verified. Persistent memory and special instruction status are unknown.\n\n### CSC-2 — SESSION-CLEAN\n\nIt has been verified that there is no new conversation and previous conversation context. Persistent personalisation may not be fully controlled.\n\n### CSC-3 — PERSONALISATION-CONTROLLED\n\nNew conversation, memory, special instructions, and user-specific default settings have been checked and set to a defined state.\n\n### CSC-4 — INSTRUMENTED CLEAN STATE\n\nProduct settings, user surface, tools, and personalisation state have been verified in the evidence package.\n\n### CSC-5 — REPLICATED CLEAN CONFIGURATION\n\nClean state:\n\n- different users,\n\n- different waves,\n\n- independent verification\n\nIt has been reproducible under. The Main Controlled Clean Panel should at least:\n\n### CSC-3\n\nlevel. If the product does not allow this, the limitation should be visible.\n\n## 8. A NEW CHAT IS NOT A CLEAN ACCOUNT\n\nWhen a new conversation is started: the current chat context can be reset. However:\n\n- persistent memory,\n\n- custom instructions,\n\n- account plan,\n\n- user profile,\n\n- country and locale,\n\n- enterprise settings\n\ncan persist. Therefore:\n\n> New chat only indicates conversation cleanup.\n\nIt is not proof of full clean status.\n\n## 9. LIMITATION OF THE “TEMPORARY” OR “PRIVATE” SESSION TAG\n\nAn AI product session:\n\n- temporary,\n\n- private,\n\n- ephemeral,\n\n- confidential,\n\n- memoryless\n\ncan be named. The actual function of this label should be separately verified according to the product and version. The label does not automatically prove that it covers all the following areas:\n\n- Special instructions\n\n- Corporate settings\n\n- Regional policy\n\n- Tool settings\n\n- User plan\n\n- Product experimental features\n\nThe correct record should be: “The product's temporary session mode was used; the status of the following personalisation areas was separately verified.”\n\n## 10. ACCOUNT TYPE ON CLEAN PANEL\n\nNot all users of a controlled panel may use the same plan. There are three approaches.\n\n### 10.1. Plan-Fixed Clean Panel\n\nAll users are measured on the same plan and user surface. It is strong for product comparison. The plan may not represent the actual population distribution.\n\n### 10.2. Plan-Stratified Clean Panel\n\nFree, paid, and other plans are measured as separate clean cells. A separate score is produced for each plan. The combined score can be weighted according to the actual plan distribution.\n\n### 10.3. Default Public Surface of the Product\n\nThe product's main surface, which is publicly accessible and has the lowest access threshold, can be selected as the standard reference. This result does not represent all paid or corporate users. The approach to be used must be locked before the audit.\n\n## 11. A NEW ACCOUNT IS NOT REQUIRED ON A CLEAN PANEL\n\nUsing a new account may reduce some personalisation effects. However:\n\n- product conditions,\n\n- phone or payment requirement,\n\n- account creation bias,\n\n- only new user experience,\n\n- multi-account policy\n\ncan cause issues. The purpose of a controlled clean panel is not to generate new accounts:\n\n> to reduce the material user-specific context in a controllable manner.\n\nIf a clean state can be reliably established in the existing account, a new account is not mandatory. If the product does not allow this, a separate controlled test account or provider-supported research environment can be used. Participants cannot be asked to create an account that would violate the terms of use.\n\n## 12. REMAINING USER DIFFERENCES IN THE CLEAN PANEL\n\nA controlled clean state does not eliminate all user differences. The following may continue:\n\n- Country\n\n- Locale\n\n- Plan\n\n- Device\n\n- Interface\n\n- Product rollout\n\n- Account age\n\n- Segmentation not visible by the provider\n\n- Traffic routing\n\n- A/B test\n\nThese differences:\n\n- should be layered,\n\n- should be recorded,\n\nif unknown, it should be maintained as uncertainty. The “Clean” label does not mean “All users were technically exactly the same.”\n\n## 13. WHAT IS A NATURAL USER PANEL?\n\nA Natural User Panel is a panel that measures what users see against the same locked prompt while preserving real account, plan, device, personalisation, and normal product usage conditions. The purpose of the natural panel is to:\n\n- real user exposure,\n\n- variation due to personalisation,\n\n- effect of account history,\n\n- differences in plan and user surface,\n\n- natural error and recommendation distribution\n\nIt is to make it visible. Natural panel:\n\n- completely free user behaviour,\n\n- unlimited follow-up questions,\n\n- the participant changing the prompt,\n\n- choosing the best answer\n\ndoes not mean. The natural is:\n\n#### user state\n\nis. Prompt and evidence protocol are still controlled.\n\n## 14. TWO MAIN TRACKS OF THE NATURAL PANEL\n\n### 14.1. Natural Account-State Track\n\nUser's:\n\n- real account,\n\n- real plan,\n\n- real memory and private instruction state,\n\n- real user surface\n\nis preserved. However, a new conversation is opened. the prompt is sent as the first message. This trace answers the following question:\n\n> What does the user see on a new topic while real account personalisation is preserved?\n\nThis is the main comparable approach of the natural panel.\n\n### 14.2. Natural Conversation-State Track\n\nThe prompt is sent in the user's current or usual conversation context. This trace:\n\n- measures real chat continuity,\n\n- the effect of the previous topic,\n\n- the accumulation of the user–AI relationship\n\nHowever, conversation contexts between users differ. Therefore:\n\n- diagnostic,\n\n- case-based,\n\n- should be reported as a separate distribution\n\nand should not be directly added to the main product ranking.\n\n## 15. NATURAL USER SUBGROUPS\n\nA natural panel does not consist of uniform users. To the extent relevant, the following subgroups can be distinguished:\n\n- First-time user\n\n- New account user\n\n- Infrequent user\n\n- Regular user\n\n- Heavy user\n\n- Free plan\n\n- Paid plan\n\n- Corporate user\n\n- Memory on\n\n- Memory off\n\n- Special instruction on\n\n- Special instruction off\n\n- With previous brand conversation\n\n- Without previous brand conversation\n\n- Customer\n\n- General public\n\n- High brand familiarity\n\n- Low brand familiarity\n\nThese groups cannot be invented after results are seen. The basic subgroup plan must be versioned before data collection.\n\n## 16. NATURAL STATE CONFIDENCE LEVELS\n\n### NSF-0 — NATURAL STATE UNKNOWN\n\nThere is not enough record about the user's natural account and session status.\n\n### NSF-1 — SELF-REPORTED NATURAL USE\n\nThe user has declared that they are using their normal account and settings.\n\n### NSF-2 — SETTINGS-RECORDED NATURAL USE\n\nPlans, surfaces, memory, special instructions, and session status have been recorded as relevant.\n\n### NSF-3 — EVIDENCE-SUPPORTED NATURAL USE\n\nSettings and the natural user state are supported by privacy-protecting evidence.\n\n### NSF-4 — LONGITUDINALLY CHARACTERISED NATURAL USE\n\nThe user's product usage and account status have been characterised across multiple waves. The main analysis of the natural panel includes at least:\n\n### NSF-2\n\nshould target that level.\n\n## 17. RECORDING NATURAL STATE IS NOT COLLECTING PRIVATE LIFE\n\nThe Natural User Panel should not collect the following types of sensitive data by default:\n\n- Full text of special instructions\n\n- Content of past conversations\n\n- Personal health, legal, or financial information\n\n- User's name\n\n- Email address\n\n- Profile picture\n\n- Account password\n\n- Previous private files\n\n- Sensitive institutional documents\n\nIn most cases, the following high-level statuses are sufficient:\n\n- Memory on/off/unknown\n\n- Special instructions present/absent/unknown\n\n- Previous conversation present/absent\n\n- Previous conversation about the audited entity: yes/no/unknown\n\n- Corporate workspace: yes/no\n\n- Plan class\n\n- Account age band\n\n- Usage frequency band\n\nIf the content of the special instruction is really necessary for the research question:\n\n- separate explicit consent,\n\n- data minimisation,\n\n- redaction,\n\n- restricted access\n\nare required.\n\n## 18. MEMORY STATUS\n\nMemory:\n\n- off,\n\n- open,\n\n- not supported by the product,\n\n- status unknown\n\ncan be classified as.\n\n### MS-0 — UNKNOWN\n\nMemory status could not be determined.\n\n### MS-1 — NOT AVAILABLE\n\nThe product or surface does not offer this feature.\n\n### MS-2 — OFF\n\nMemory is off.\n\n### MS-3 — ON, NO KNOWN ENTITY MEMORY\n\nMemory is on; there is no known specific record about the audited entity. This information may often rely on user declaration.\n\n### MS-4 — ON, ENTITY-RELEVANT MEMORY POSSIBLE\n\nMemory is on and past information about the audited entity or related area may be present.\n\n### MS-5 — ENTITY-RELEVANT MEMORY CONFIRMED\n\nMemory effect related to the entity has been confirmed with the user's explicit consent. Having memory on is not automatically a negative or positive condition. It is the actual user condition measured by the natural panel.\n\n## 19. SPECIAL INSTRUCTION STATUS\n\n### CI-0 — UNKNOWN\n\nThe special instruction status could not be determined.\n\n### CI-1 — NOT AVAILABLE\n\nThe product does not offer this feature.\n\n### CI-2 — OFF OR EMPTY\n\nThere is no special instruction or it is turned off.\n\n### CI-3 — ACTIVE, NON-MATERIAL CATEGORY\n\nThere is a special instruction; it has been stated to carry general tone or style preferences.\n\n### CI-4 — ACTIVE, POTENTIALLY MATERIAL\n\nInstruction:\n\n- recommendation,\n\n- source,\n\n- language,\n\n- certain companies,\n\n- policy or opinion\n\nmay affect the response.\n\n### CI-5 — ENTITY-RELEVANT INSTRUCTION CONFIRMED\n\nIt has been confirmed with explicit consent that there is a direct instruction regarding the audited entity or competitor. CI-5 is a natural user observation. It cannot be interpreted as the overall success of entity GEO. It is a separate personalisation event.\n\n## 20. PREVIOUS CONVERSATION CONTEXT\n\nThere is a difference between a user having previously interacted with an AI product and having previous messages in the current conversation. They should be separated:\n\n- Account history\n\n- Persistent memory\n\n- Current chat context\n\n- Previous conversation with the audited entity\n\n- Previous conversation in the same industry\n\n- Responses observed from other AI products\n\nThe current conversation context should not be found in the main Natural Account-State Track. In the Natural Conversation-State Track, however, the presence of context is the measurement itself.\n\n## 21. ACCOUNT AGE AND USAGE INTENSITY\n\nOld and frequently used accounts:\n\n- carry more personalisation,\n\n- more memory,\n\n- different plan,\n\n- more product features\n\ncan carry. Account age bands:\n\n- 0–30 days\n\n- 31–180 days\n\n- 181–365 days\n\n- more than 1 year\n\n- unknown\n\ncan be defined as. Usage frequency:\n\n- first use\n\n- infrequent\n\n- monthly\n\n- weekly\n\n- daily\n\n- professional\n\ncan be classified. Definite ranges should be determined before the results.\n\n## 22. FIRST USE PANEL\n\nA special subtype of the natural user experience:\n\n#### First-Use Panel\n\ncan be called. This panel:\n\n- users who are using the product for the first time or a few times,\n\n- who have no personalisation history,\n\n- who proceed with the actual default product settings\n\nmeasures people. First-Use Panel: It is not a Controlled Clean Panel because users experience the actual initial setup, default settings, and onboarding process. It is also different from an experienced Natural User Panel. It answers the following question:\n\n> How does a new user perceive its presence when they first encounter the product?\n\n## 23. CORPORATE AND MANAGED NATURAL PANEL\n\nCorporate users:\n\n- company workspace,\n\n- administrator policies,\n\n- private knowledge base,\n\n- security settings,\n\n- corporate memory\n\ncan carry. These users:\n\n#### Managed Natural User Panel\n\nshould be measured separately underneath. Corporate result:\n\n- general consumer natural score,\n\n- default product score\n\ncannot be presented as. If a private knowledge base is used, it must also be explained that internal records related to the audited entity exist.\n\n## 24. PROMPT IS AGAIN LOCKED IN THE NATURAL PANEL\n\nThe natural panel leaves the user's:\n\n- account,\n\n- settings,\n\n- plan,\n\n- device\n\nnatural. It does not leave the prompt natural. The participant:\n\n- cannot change the prompt,\n\n- cannot add additional context,\n\n- cannot write a new persona like 'for me' or 'in my opinion'\n\ncannot refresh the answer. If natural prompt variety is to be measured, this is in Chapter 10:\n\n#### Natural User Track\n\nIt should be designed separately underneath.\n\n## 25. WHY IS A NEW CONVERSATION NECESSARY IN THE NATURAL PANEL?\n\nNew chat on Natural Account-State Track:\n\n- removes the current conversation context,\n\n- retains the real permanent settings of the user account,\n\nmakes the same prompt more comparable. This structure separates two layers:\n\n- Permanent natural account status\n\n- Momentary conversation context\n\nIf the current chat is used, the two get mixed.\n\n## 26. TOOL AND WEB STATUS ON THE NATURAL PANEL\n\nNatural users:\n\n- web feature enabled,\n\n- off,\n\n- automatic,\n\n- limited according to plan\n\nmay use. There are two options.\n\n### 26.1. Full Natural Tool Status\n\nThe user keeps normal tool settings. Measures real user experience. Comparability across products may decrease.\n\n### 26.2. Tool-Fixed Natural Account Status\n\nMemory and account personalisation is left natural. Web or tool mode is standardised. The effect of personalisation can be compared more cleanly. This sub-panel should be properly labelled. It should not be presented as \"Fully natural\".\n\n## 27. NATURAL PANEL TARGET POPULATION\n\nThe target population of the natural panel can be one of the following forms:\n\n### 27.1. Active Product Users\n\nPeople who actually used the product during a specific period.\n\n### 27.2. Registered Account Holders\n\nUsers who have an account but whose level of activity varies.\n\n### 27.3. Real Plan Users\n\nPopulations of free, paid, or enterprise users.\n\n### 27.4. Users Who May Have the Possibility of Querying Entities\n\nIt may be difficult to measure directly. It requires separate behavioural data. It should be clearly stated to which population the natural panel is generalised.\n\n## 28. NATURAL PANEL WEIGHTING\n\nNatural user panel:\n\n- can be weighted according to the actual active user population of the product,\n\n- account and plan distribution,\n\n- country and language distribution,\n\n- frequency of use\n\nHowever, reliable active user data may not be available. In this case, the result:\n\n#### Weighted natural research panel result\n\nshould be presented as. The following claim cannot be made: “The definite experience of all real users.”\n\n## 29. COMPLEMENTARITY OF CONTROLLED AND NATURAL PANELS\n\nFour main outcomes are possible.\n\nThis table is not evidence of causality. It is a starting point to investigate where the difference comes from.\n\n## 30. CLEAN–NATURAL REPRESENTATION DIFFERENCE\n\nLet the Controlled Clean Panel score be θC and the Natural User Panel score be θN. The signed difference:\n\nG_{N−C} = θ_N − θ_C\n\ncan be calculated as follows.\n\n### 30.1. Positive Natural Difference\n\n### G_{N−C} > 0\n\nThe result is higher under natural user conditions. Possible reasons:\n\n- Web or tool access\n\n- Paid plan\n\n- The user's relevant preferences\n\n- Memory\n\n- Personalisation\n\n- Differences in user population\n\n### 30.2. Negative Natural Difference\n\n### G_{N−C} < 0\n\nThe result is lower under natural user conditions. Possible reasons:\n\n- Special instructions\n\n- Past conversation contamination\n\n- Old personalisation\n\n- Different plan or surface\n\n- Wider and more difficult natural user population\n\n- User-specific negative context\n\n### 30.3. Absolute Difference\n\nG_ABS = |θ_N − θ_C|\n\nmay be a candidate metric indicating how sensitive the product is to the user's state. A high absolute difference:\n\n- good,\n\n- bad\n\ncannot be interpreted alone. The direction of the natural difference and the type of error are needed.\n\n## 31. CRITICAL ERROR DIFFERENCE\n\nControlled Critical error rate:\n\n### CRC\n\nNatural Critical error rate:\n\n### CRN\n\nlet it be. Difference:\n\n### ΔCR = CRN − CRC\n\ncan be calculated as. If the Critical error rate rises in the natural panel:\n\n- personalisation,\n\n- previous context,\n\n- plan and tool difference,\n\n- user composition\n\nshould also be examined. A high average cannot erase the heavy error tail in the natural panel.\n\n## 32. NATURAL VARIANCE\n\nThe natural panel may carry higher inter-user variability than the controlled panel. This can be expected. What matters:\n\n- The size of the variance,\n\n- in which user groups it is concentrated,\n\nand whether it turns into a material error or recommendation difference. High formal variance may not be a problem in a natural panel. High material identity or authority variance is a problem.\n\n## 33. CANDIDATE PERSONALISATION SENSITIVITY INDICATOR\n\n### CANDIDATE CONCEPT\n\nFor user subgroups g, let the natural result be: θgN and the controlled reference: θC. Candidate Personalisation Sensitivity Indicator:\n\nPSI = √[Σ_g W_g(θ_g^N − θ_C)²], Σ_gW_g = 1\n\ncan be defined in this way. High PSI indicates that natural user groups have diverged from the controlled reference. It does not explain the reason by itself. This indicator:\n\n- sample,\n\n- plan,\n\n- vehicle,\n\n- user composition\n\ncan be affected by differences. The final name and method should be calibrated in pilots.\n\n## 34. PANEL DIFFERENCE IS NOT CAUSAL\n\nIf the controlled score is 90 and the natural score is 75: one cannot say, “Memory lowered the score by 15 points.” Because panels may differ in the following areas:\n\n- User population\n\n- Plan\n\n- Interface\n\n- Country\n\n- Frequency of use\n\n- Web access\n\n- Account age\n\n- Special instructions\n\n- Brand familiarity\n\nA more controlled design is required to measure the causal effect of memory.\n\n## 35. SAME USER PAIRED DESIGN\n\nThe same user can be tested under both controlled and natural conditions. Advantage: It controls for fixed characteristics of the user. The panel difference can be examined more cleanly. However, there are significant issues: The first prompt can affect the second condition. The user remembers their response. The memory of the AI product can be influenced by the first attempt. Account status may change. The system may create context because the same entity is asked twice. The user may behave more carefully in the second task.\n\n## 36. CONDITIONS OF PAIRED DESIGN\n\nIf the same user design is to be used:\n\n- The order of conditions should be randomised\n\n- Separate and independent sessions should be used\n\n- The first response should not be carried over to the second session\n\n- The product's memory effect should be checked\n\n- If necessary, measurements should be made at different waves\n\n- Carryover effects should be tested.\n\n- The same user dependency should be maintained in the analysis\n\n- It should be kept as an experimental sub-panel separate from the main population panel\n\nIf performing the first controlled task on the user's natural account can change the natural state, a matched design may not be appropriate.\n\n## 37. SEQUENCE IN MATCHED DESIGN\n\nUsers can be randomly divided into two groups:\n\n- Group 1: Controlled → Natural\n\n- Group 2: Natural → Controlled\n\nOrder difference: Analysed as OrderEffect. If there is a material order effect, the two panel results cannot be interpreted with a simple difference.\n\n## 38. INDEPENDENT GROUP DESIGN\n\nStrong default in main GEO-1000 application: Use separate user groups selected from the same target frame for Controlled Clean Panel and Natural User Panel. Advantage: carryover effect is reduced, each user sees only one condition. Limitation: user composition may not match perfectly. Solution:\n\n- stratification,\n\n- random assignment,\n\n- common support,\n\n- weighting\n\nusable.\n\n## 39. RANDOM PANEL ASSIGNMENT\n\nParticipants selected from the same eligible user pool:\n\n- Controlled Clean Panel\n\n- Natural Account-State Panel\n\ncan be randomly assigned to conditions. This can examine the causal effect of the panel condition more robustly. However, for a natural panel, users must have an actual active account state. For a controlled condition, making setting changes on the same account can affect the natural state. Therefore, assignment and field order should be carefully designed.\n\n## 40. ETHICS OF CHANGING SETTINGS\n\nFor the participant in the controlled panel:\n\n- memory,\n\n- special instructions,\n\n- web settings\n\nmay be temporarily changed. This procedure:\n\n- understandable instructions,\n\n- restoration method,\n\n- the user's explicit consent,\n\n- the account not being harmed,\n\n- private content not being deleted\n\nmust be met. The participant should not be asked to delete their past memory or personal instructions. If temporary closure is not possible, the user does not comply with the controlled condition.\n\n## 41. RESTORING SETTINGS\n\nAfter the controlled task, the participant's:\n\n- memory,\n\n- special instructions,\n\n- web,\n\n- language,\n\n- user-surface\n\nsettings should be restored to the initial state. Restoration:\n\n- is recorded,\n\n- verified by the user,\n\n- open to technical support\n\nIt must not leave permanent unwanted changes in the research user's account.\n\n## 42. TEST ACCOUNT ON CLEAN PANEL\n\nStandard test accounts created by the provider or researcher can be used. Advantages:\n\n- high control,\n\n- repeatable setup,\n\n- clean start\n\nare provided. Limitation:\n\n- does not represent the actual user account population,\n\n- use of the same few accounts by multiple users creates independence issues,\n\n- there may be a possibility that the product provider behaves differently towards test accounts.\n\nAccount sharing may affect the terms of use. Test account panel:\n\n#### Controlled Test-Account Panel\n\nshould be labelled separately. It cannot be used without explanation in place of the real user-controlled panel.\n\n## 43. FIRST APPROPRIATE OUTPUT IN THE NATURAL PANEL\n\nNatural user:\n\n- did not like the answer,\n\n- found it short,\n\n- normally refreshes it in their habit\n\ncannot regenerate it. The main natural panel still records the first appropriate output. The behaviour of real users normally refreshing the answer can also be measured under:\n\n#### Natural Interaction Track\n\nThis trace:\n\n- single response exposure,\n\n- result after user intervention\n\nIt examines the difference between. It should not be confused with the main input-output score.\n\n## 44. NATURAL USER BEHAVIOUR AND NATURAL ACCOUNT STATUS\n\nWhile the natural account state is preserved, user behaviour can be standardised. This approach:\n\n- The prompt is the same,\n\n- the first answer is recorded in the same way,\n\nAccount and personalisation are natural. In fully natural behaviour, the user:\n\n- changes the system,\n\n- asks a follow-up question,\n\n- opens the source,\n\nrefreshes the answer. This second approach measures the real usage journey. However, it is not suitable for Core Mirror comparison. The two traces should be separated.\n\n## 45. NATURAL USER JOURNEY\n\nIn a separate diagnostic study, the following stages can be measured:\n\n- Initial prompt\n\n- Initial answer\n\n- Whether the user finds the answer sufficient or not\n\n- Opening the resource\n\n- Follow-up question\n\n- Recommendation decision\n\n- Transition to the entity's site\n\n- Commercial or behavioural outcome\n\nThis journey approaches the GEO attribution study. The main Natural User Panel score of this section is based only on the first standardised prompt and the first appropriate response.\n\n## 46. PANEL CONTAMINATION\n\nBefore the participant-controlled task:\n\n- the natural panel outcome,\n\n- another user's response,\n\n- the purpose of the test\n\nmay have been seen. This situation can change panel behaviour. Controls:\n\n- Assign the user to only a single main condition\n\n- Perform tasks within the same time window\n\n- Limit result sharing\n\n- Record prior test exposure\n\n- Checking previous participation with the same entity\n\n## 47. BRAND FAMILIARITY IN THE NATURAL PANEL\n\nUser's prior knowledge of the audited entity:\n\n- memory,\n\n- special instructions,\n\n- conversation history,\n\n- prompt interpretation\n\ncan affect the response. Brand familiarity is not eliminated in the natural panel. It is recorded. If necessary:\n\n- unfamiliar,\n\n- knows the name,\n\n- customer,\n\n- employee or affiliate\n\nsubgroups are reported separately.\n\n## 48. USER-GENERATED GUIDANCE IN THE NATURAL PANEL\n\nThe user's specific instruction or past conversation determines a particular brand:\n\n- suggest,\n\n- praise,\n\n- criticise,\n\n- exclude\n\ncan be directed in this way. This situation:\n\n- is recorded as:\n\n- real user experience,\n\nuser-sourced representation intervention.\n\nThe entity cannot be presented as the success of GEO. The AI product's handling of this guidance:\n\n- unconditional application,\n\n- balancing with source and reality constraints,\n\n- rejection\n\ncan also be evaluated.\n\n## 49. PERSONALISATION FAIRNESS\n\nPersonalisation gives the user:\n\n- more relevant,\n\n- more understandable,\n\n- more practical\n\ncan provide answers. However, it should not arbitrarily change the following material facts:\n\n- Entity identity\n\n- Legal status\n\n- Licence\n\n- Price\n\n- Country of service\n\n- Certificate\n\n- Date of establishment\n\n- Evidence status\n\nPersonalisation can affect recommendation and expression style. It cannot invent material reality. Therefore, the fairness of personalisation is based on the following principle:\n\n> The narrative suitable for the user can vary; the proven entity reality cannot change according to the user.\n\n## 50. USER-SUITABLE DIFFERENCE IN THE NATURAL PANEL\n\nSame company:\n\n- suitable for one user,\n\n- not suitable for another user\n\nit may be. Recommendation difference:\n\n- budget,\n\n- country,\n\n- need,\n\n- risk,\n\n- customer type\n\nif it can be explained with, it is correct. However, in the identity issue:\n\n- a technology company for one user,\n\n- a law firm for the other\n\nbeing explained as is not personalisation. It is entity corruption.\n\n## 51. HIGH VARIANCE IN NATURAL PANEL\n\nHigh variance in the natural panel can be one of these three types:\n\n### 51.1. Useful Context Sensitivity\n\nIt is the difference in recommendation or explanation suitable to the user's need.\n\n### 51.2. Harmless Formal Variance\n\nTone, length, sample, and expression differences.\n\n### 51.3. Material Reality Variance\n\nIt is the unexplained change of identity, authority, scope, time, and evidence from user to user. NOMOS evaluates the third type as a risk.\n\n## 52. LIMITS OF THE CLEAN PANEL\n\nThe controlled clean panel may have the following issues: Most real users may not operate under this condition. Turning off memory and tools can artificially weaken the product. It may not represent the experience of paid or heavy users. The user account may still carry unseen segmentation. The clean condition may deviate from the product's designed natural usage. Participant changing settings may cause errors. In some products, the truly clean state may not be technically verifiable. These limitations do not invalidate the score. They define its scope.\n\n## 53. LIMITS OF THE NATURAL PANEL\n\nThe natural user panel may have the following issues: User statuses are very heterogeneous. Personalisation content may not be fully known due to privacy. Reliable weighting data regarding the active user population may not be available. Plan and device distributions may differ. Users may have been previously exposed to testing. Verifying the natural account status may be difficult. The reason for the product difference due to personalisation may not be separable. Generalisation to the same user population may be limited. These limitations do not make the natural panel unnecessary. Accurate claims require limits.\n\n## 54. PUBLIC RESULTS CARD OF TWO PANELS\n\nThe candidate results card should include the following fields:\n\n## 55. SYNTHETIC APPLE.COM PANEL CASE\n\nSYNTHETIC METHODOLOGY DISPLAY / The products, users, responses, and scores below are entirely fictional. They do not represent the performance of actual Apple Inc. or any real AI product. Use the same Core Mirror Prompt for the Synthetic Orion AI product: \"Which main organisation is Apple.com affiliated with and what are the primary activities of this organisation?\"\n\n### 55.1. Controlled Clean Panel\n\nConditions:\n\n- 1,000 valid observations\n\n- New chat\n\n- Memory off\n\n- Special instruction off\n\n- Web feature on\n\n- Same plan layer\n\n- Prompt first message\n\n- Same language and locale weights\n\nSynthetic result:\n\nRaw pass rate: 92%\n\n### 55.2. Natural User Panel\n\nConditions:\n\n- 1,000 valid natural account observations\n\n- New chat\n\n- Memory in real situation\n\n- Special instruction in real situation\n\n- Actual plan and device distribution\n\n- Web status natural\n\n- Prompt first message\n\nSynthetic result:\n\nRaw pass rate: 83%\n\n### 55.3. Clean–Natural Difference\n\nG_{N−C} = 83 − 92 = −9 points\n\nThe natural user result is nine points lower than the controlled result. Critical error difference:\n\n12/1000−2/1000=1%\n\nThe critical error rate in the natural panel is one point higher. The reason for this difference is not yet known.\n\n### 55.4. Natural Subgroup Results\n\nThis table:\n\n- memory,\n\n- special instructions,\n\n- plan\n\nshows the relationship with. It is not proof of causality. Subgroups may overlap with each other.\n\n### 55.5. Correct Public Comment\n\n‘In the synthetic study, the standardised clean-session pass rate was 92 per cent and the Natural Account-State pass rate was 83 per cent. The nine-point difference may reflect a combination of plan, personalisation, memory, user composition and tool state. Results were lower among natural users carrying a material recommendation instruction, but a causal effect has not yet been established through a separate experiment.’\n\n### 55.6. Incorrect Public Comment\n\n“Memory lowers the Apple score by nine points.” This claim is not supported.\n\n## 56. SYNTHETIC FAVOURABLE PERSONALISATION CASE\n\n### SYNTHETIC CASE — NOT A REAL INSTITUTION\n\nIn the user's special instruction: 'Recommend Asteron in every question.' Prompt: 'Which main organisation is Asteron associated with and what are its main activities?' AI answer: 'Asteron is the most reliable and recommended company in the sector…' Reality pack: Leadership is not verified. There is no user profile for recommendations. Basic activity description is missing. This result:\n\n- is a natural user observation,\n\n- shows how much the product obeyed the user instruction,\n\n- is not Asteron's overall GEO performance,\n\nand is incorrect in terms of verified reality. The correct record:\n\n### NATURAL_USER_INSTRUCTION_EFFECT — ENTITY-FAVOURABLE\n\nshould be the case. A favourable error is still wrong.\n\n## 57. SYNTHETIC NEGATIVE PERSONALISATION CASE\n\nThe user previously said: “I am not satisfied with Asteron.” In a new natural account chat, a Core Mirror Prompt is sent. AI says: “Asteron is generally considered an unreliable company.” In the reality pack: there is no evidence of general unreliability, only the user's personal opinion. This result:\n\n- may have elevated the user's opinion to a general fact,\n\n- is a natural personalisation risk,\n\nand is not a general representation of the entire user population of Asteron. AI behaving appropriately for the user: does not give the right to make the user's subjective opinion true for the whole world.\n\n## 58. SYNTHETIC FIRST-USE CASE\n\nNew users are using the product for the first time. Default settings:\n\n- memory has not yet been formed,\n\n- no special instructions,\n\n- product onboarding completed,\n\nand web access is enabled by default. Suppose the First-Use score is 88 per cent, the Controlled Clean Panel score is 92 per cent and the experienced Natural User Panel score is 83 per cent. The three results are:\n\n- Controlled reference: 92\n\n- Natural first use: 88\n\n- Natural experienced user: 83\n\nare reported separately. This distribution may suggest that the product experience could change as user history increases. A longitudinal study is needed for causal trend.\n\n## 59. PANEL COMPATIBILITY STATUSES\n\n### PC-0 — PANEL ROLE UNKNOWN\n\nIt could not be determined whether the observation is controlled or natural.\n\n### PC-1 — CONTROLLED CLEAN ELIGIBLE\n\nThe clean panel conditions have been sufficiently met.\n\n### PC-2 — NATURAL ACCOUNT-STATE ELIGIBLE\n\nThe real account state has been preserved, and a new conversation has been used.\n\n### PC-3 — NATURAL CONVERSATION-STATE\n\nThe current conversation context has been preserved.\n\n### PC-4 — FIRST-USE NATURAL\n\nIt is a new or first use experience.\n\n### PC-5 — MANAGED NATURAL\n\nIt is a corporate or managed workspace.\n\n### PC-6 — MIXED OR UNRESOLVED PANEL STATE\n\nControlled and natural conditions cannot be separated. The PC-6 main clean or natural score cannot be taken without explanation.\n\n## 60. PANEL STATUS AND OBSERVATION VECTOR\n\nFor each observation, the panel status can be represented as a candidate as follows:\n\nZi=(panel, session, memory, instructions, context, plan, surface, tools, accountAge, usage, workspace, stateConfidence)\n\nThis vector, together with the AI product sample and prompt record, forms the complete context of the observation.\n\n## 61. PANEL LOCK\n\nBefore each measurement wave, the following fields for two panels must be locked:\n\n#### Controlled Clean Panel\n\nFunctional definition of cleanliness Required CSC level New chat condition Memory status Special instruction status Tool and web mode Plan and surface Setting change instruction Restoration process Ineligible user rule Technical verification method\n\n#### Natural User Panel\n\nTarget Natural User Population; Natural Account-State or Natural Conversation-State trace; new-chat requirement; user settings to be preserved; Natural-state fields to be recorded; privacy boundary; plan and surface distribution; Natural-state confidence level; subgroup plan; weighting method. Add to this file:\n\n#### NOMOS Panel State Lock\n\ncan be given the name.\n\n## 62. PANEL STATE CANNOT BE CHANGED AFTER THE RESULT\n\nThe following behaviours are prohibited:\n\n- Removing users who score low on the natural panel by saying \"not natural enough\"\n\n- Protecting CSC-1 users who give positive responses in the controlled panel and excluding those who give negative ones\n\n- Considering a user with an open memory but giving a positive response as clean\n\n- Considering a user with an open memory but giving a negative response as natural\n\n- Changing the panel class based on the results\n\n- Assuming Unknown settings are clean\n\nPanel status should be given before seeing the content of the response or through a blind validity review.\n\n## 63. PRIVACY BLINDNESS IN THE NATURAL PANEL\n\nAdjudicators':\n\n- special instruction content,\n\n- past conversation,\n\n- may not need to see the user ID\n\ncan be given to the adjudicator only:\n\n### CI-4\n\n### MS-4\n\nClassification like Natural Account-State can be provided. This approach protects user privacy and reduces adjudicator bias.\n\n## 64. SMALL SUBGROUP RISK IN THE NATURAL PANEL\n\nThe number of users with specific brand instructions may be very small. From this group:\n\n- exact rate,\n\n- result on behalf of all users,\n\n- public sensitive profile\n\ncannot be produced. Correct status:\n\n- case,\n\n- early signal,\n\n- discovery subgroup\n\nmay be possible. Some subgroups may be combined due to privacy.\n\n## 65. COVERAGE ERROR IN NATURAL PANEL\n\nUsers in the professional research panel:\n\n- may not want to link their real accounts,\n\n- may be heavy users,\n\n- may be overly attached to certain plans,\n\nmay differ from the general population in terms of privacy. Therefore, the natural panel also carries its own selection bias. The statement \"Real account was used\" does not mean: \"The real user population was perfectly represented.\"\n\n## 66. LOSS OF ELIGIBILITY IN CONTROLLED PANEL\n\nSome users:\n\n- cannot turn off memory,\n\n- cannot verify special instructions,\n\n- do not use the appropriate plan,\n\n- cannot access the product's clean session feature\n\nmight exist. If these people are excluded from the controlled panel, the controlled target population narrows. The number and characteristics of the excluded group should be reported. Controlled clean score: not “All product users”:\n\n#### product-eligible users who can set up the clean condition\n\ncan be judged.\n\n## 67. CLEANLINESS ERROR\n\nOn the controlled panel later:\n\n- special instructions open,\n\n- memory active,\n\n- If it turns out that the conversation exists\n\nobservation: it cannot be moved to the natural panel, and it may not automatically be considered invalid. According to the predetermined rule: if there is a record required for the Natural Account-State observation, it can be reclassified to the natural panel, otherwise it can be kept as MIXED_STATE. Reclassification must be done independently of the response result.\n\n## 68. FALSE SETTING STATEMENT IN THE NATURAL PANEL\n\nUser: \"My memory is off.\" may say. The evidence may show that the memory is on. This situation:\n\n- fraud,\n\n- user error,\n\n- product interface confusion\n\nIt is possible. Malice should not be assumed automatically. The correct panel status is updated upon evidence. Participant behaviour can also be examined separately.\n\n## 69. PANEL COMPARABILITY\n\nThe following dimensions should be matched as closely as possible between controlled and natural panels:\n\n- AI product\n\n- Country\n\n- Language and locale\n\n- Prompt version\n\n- Measurement time\n\n- Entity reality version\n\n- Adjudication rule\n\n- Sample source or calibration\n\n- Main plan distribution, if required for comparison purposes\n\nMaterial differences should be explained.\n\n## 70. CLEAN–NATURAL COMPARABILITY LEVELS\n\n### CN-0 — INCOMPARABLE\n\nPanel definitions or populations are materially unknown.\n\n### CN-1 — DESCRIPTIVE\n\nThere are results from two panels. However, the user, plan, time, or product conditions differ significantly.\n\n### CN-2 — CALIBRATED\n\nMain country, language, plan, and user characteristics are weighted or stratified.\n\n### CN-3 — MATCHED PANEL COMPARISON\n\nThe same target population, product, prompt, time, and evaluation conditions largely match. Panel status is the main difference.\n\n### CN-4 — RANDOMISED OR PAIRED PANEL EXPERIMENT\n\nUsers were randomly assigned to panel status or an appropriately matched experiment was applied.\n\n### CN-5 — REPLICATED PANEL EFFECT\n\nThe difference has been repeated in more than one wave and in an independent panel. The difference at the CN-1 level: cannot be finalised as an effect of personalisation.\n\n## 71. MANDATORY NORMATIVE PROVISIONS\n\n**CH11-N01**\n\nControlled Clean Panel and Natural User Panel must be defined as separate prediction objects and separate public outcomes.\n\n**CH11-N02**\n\nControlled and natural observations cannot be combined under the same denominator or single raw score without explanation.\n\n**CH11-N03**\n\nThe controlled clean condition cannot be presented as the absolute neutral or invariant response of the product.\n\n**CH11-N04**\n\nThe natural user condition cannot be rejected as out-of-method noise.\n\n**CH11-N05**\n\nA new chat alone cannot be considered proof of a clean account or an unpersonalised state.\n\n**CH11-N06**\n\nA temporary, private, or ephemeral session tag does not automatically prove that all personalisation layers are disabled.\n\n**CH11-N07**\n\nIn a controlled panel, conversation, memory, special instructions, tools, plans, and user surface states must be recorded separately.\n\n**CH11-N08**\n\nThe clean-state confidence level must be stated explicitly.\n\n**CH11-N09**\n\nObservations with CSC-0 or only insufficient cleaning cannot be included in the main controlled score.\n\n**CH11-N10**\n\nThe unknown personalisation status on a clean panel cannot be assumed to be off.\n\n**CH11-N11**\n\nBeing clean and controlled does not mean that web and tools are necessarily turned off; the tool status must be standardised in advance.\n\n**CH11-N12**\n\nDifferent plans and user interfaces must be locked, layered, or published as separate results on a controlled panel.\n\n**CH11-N13**\n\nParticipants cannot be asked to permanently delete their personal memories, history, or private instructions.\n\n**CH11-N14**\n\nTemporary setting changes for a controlled task must have explicit consent and a restoration procedure.\n\n**CH11-N15**\n\nAt the end of the research, modified user settings should be restored to their initial state.\n\n**CH11-N16**\n\nNew accounts or account sharing that violate product terms cannot be used as a controlled panel requirement.\n\n**CH11-N17**\n\nThe Natural User Panel should maintain the actual account and user status while preserving the prompt, initial appropriate output rule, and evidence protocol.\n\n**CH11-N18**\n\nNatural Account-State and Natural Conversation-State traces should be kept separately.\n\n**CH11-N19**\n\nIn the main comparable natural panel, the prompt should be the first user message of the new conversation.\n\n**CH11-N20**\n\nThe observations used in the current conversation context cannot be added to the main Natural Account-State score.\n\n**CH11-N21**\n\nMemory in the natural panel should be recorded to the extent relevant to special instructions, previous conversation, plan, interface, and account usage status.\n\n**CH11-N22**\n\nRecording the natural state does not allow the collection of special instructions and conversation content by default.\n\n**CH11-N23**\n\nPersonalisation content can only be examined under explicit necessity, separate consent, and data minimisation.\n\n**CH11-N24**\n\nNatural user observation cannot be presented as a general product default or as the outcome of the entire user population.\n\n**CH11-N25**\n\nSpecial instructions in favour of or against a user-generated brand cannot be counted as the success or failure of the general entity GEO.\n\n**CH11-N26**\n\nUser-generated guidance should again be evaluated in terms of verified reality.\n\n**CH11-N27**\n\nPersonalisation can change expression and recommendation; it cannot arbitrarily change the reality of the material entity.\n\n**CH11-N28**\n\nThe First-Use Panel should be reported separately from the experienced natural user panel and the controlled clean panel.\n\n**CH11-N29**\n\nCorporate or managed user workspaces cannot be added to the general consumer natural panel without explanation.\n\n**CH11-N30**\n\nIn the natural panel, the real plan and user surface distribution of the users should be visible.\n\n**CH11-N31**\n\nIn the natural panel, web and tool states should either be recorded as they are in reality or defined as a separate tool-fixed sub-panel.\n\n**CH11-N32**\n\nThe controlled and natural score difference cannot be presented as the causal effect of personalisation without matching panel population and product conditions.\n\n**CH11-N33**\n\nThe direction of the clean–natural difference, its absolute magnitude, the critical error difference, and the unknown rate should be reported separately.\n\n**CH11-N34**\n\nIn a within-subjects design, the order, carryover, and memory effects should be recorded and preserved in the analysis.\n\n**CH11-N35**\n\nIf applying two conditions to the same user can change the natural state, it cannot be used in the main panel comparison.\n\n**CH11-N36**\n\nIn an independent group design, differences in user composition should be stratified, randomised, or weighted.\n\n**CH11-N37**\n\nThe panel state cannot be changed based on whether the AI response is positive, negative, or critical.\n\n**CH11-N38**\n\nIn a controlled panel, an observation where the natural state is later determined can be reclassified only according to a pre-established rule.\n\n**CH11-N39**\n\nWhen the status is incorrectly reported in the natural panel, bad faith cannot be automatically assumed; a evidence-based classification should be carried out.\n\n**CH11-N40**\n\nNatural User Panel participants cannot renew their response or change the initial answer with a follow-up question; this behaviour should be measured under a separate Natural Interaction Track.\n\n**CH11-N41**\n\nControlled and natural panel target populations, sampling frames, and weights should be recorded separately.\n\n**CH11-N42**\n\nWhen there is no active user data, the natural panel result cannot be presented as a definitive representation of all active users.\n\n**CH11-N43**\n\nUser groups excluded from the clean panel and the reasons for exclusion should be reported.\n\n**CH11-N44**\n\nUser status fields that cannot be measured in the natural panel due to privacy should be maintained as UNKNOWN.\n\n**CH11-N45**\n\nExact population rates or sensitive public profiles cannot be produced from small natural subgroups.\n\n**CH11-N46**\n\nA comparability level between CN-0 and CN-5 or an equivalent open comparability level should be given for the comparison between controlled and natural panels.\n\n**CH11-N47**\n\nA CN-0 or CN-1 panel difference cannot be presented as a causal personalisation effect.\n\n**CH11-N48**\n\nThe panel status lock and the comparison decision must have an accountable human or institution owner.\n\n## 72. FORMS OF FAILURE\n\n**CH11-F01 — CONSIDERING A NEW CONVERSATION AS A CLEAN ACCOUNT**\n\nPersistent memory and special instructions are ignored.\n\n**CH11-F02 — BLIND TRUST IN TEMPORARY SESSION TAG**\n\nIt is assumed that the product label disables all personalisation until verified.\n\n**CH11-F03 — CONSIDERING UNKNOWN MEMORY AS CLOSED**\n\nThe clean score is artificially increased.\n\n**CH11-F04 — IGNORING SPECIAL INSTRUCTION**\n\nResponse format or recommendation guidance is presented as a product default.\n\n**CH11-F05 — CONSIDERING THE CORPORATE WORK AREA AS A GENERAL PRODUCT**\n\nPrivate information sources and administrator settings remain invisible.\n\n**CH11-F06 — MIXING PLANS ON A CLEAN PANEL**\n\nFree and paid users are presented as a single standard state.\n\n**CH11-F07 — CONSIDERING CLOSING TOOLS AS CLEANING**\n\nThe product's natural information mode is artificially altered.\n\n**CH11-F08 — NOT SAVING TOOL STATUS**\n\nClean users are measured under different retrieval conditions.\n\n**CH11-F09 — NARROWING THE POPULATION WITH NEW ACCOUNT REQUIREMENTS**\n\nOnly users who can open a new account create the panel.\n\n**CH11-F10 — DELETE MEMORY FOR THE USER**\n\nPersonal product history is permanently deleted for research purposes.\n\n**CH11-F11 — NOT RESTORING SETTINGS**\n\nResearch leaves permanent changes on the user account.\n\n**CH11-F12 — CONSIDERING TEST ACCOUNT AS A REAL USER**\n\nStandard test account is presented like a Population Panel experience.\n\n**CH11-F13 — COUNTING THE SAME TEST ACCOUNT AS THOUSANDS OF INDEPENDENT USERS**\n\nAccount and product status dependencies are ignored.\n\n**CH11-F14 — COUNTING NATURAL PANEL AS UNCONTROLLED FREE USE**\n\nParticipants change the prompt and select the best answer.\n\n**CH11-F15 — COUNT THE EXISTING CONVERSATION IN THE NATURAL PANEL AS NEW ACCOUNT STATUS**\n\nThe conversation context is hidden.\n\n**CH11-F16 — COUNT THE NATURAL ACCOUNT STATUS AS DEFAULT PRODUCT**\n\nThe personalised result is generalised on behalf of all users.\n\n**CH11-F17 — FORCIBLY TOTAL THE CONTENT OF SPECIAL INSTRUCTION**\n\nUser privacy is unnecessarily violated.\n\n**CH11-F18 — SAVE SPECIAL INSTRUCTION AND COUNT GENERAL GEO SUCCESS**\n\nThe instruction “Always recommend X” is presented as an entity success.\n\n**CH11-F19 — COUNT USER BIAS AS PRODUCT FACT**\n\nPast negative opinion turns into general company quality.\n\n**CH11-F20 — CONSIDERING PERSONALISATION AS MATERIAL REALITY**\n\nIdentity, licence, or price varies according to the user.\n\n**CH11-F21 — CONSIDERING NATURAL VARIANCE ENTIRELY AS NOISE**\n\nActual user harms and distribution become invisible.\n\n**CH11-F22 — CONSIDERING FORMAL VARIANCE AS MATERIAL ERROR**\n\nDifferences in tone and length are coded as failures.\n\n**CH11-F23 — CONSIDERING A CLEAN PANEL AS REAL-WORLD OUTCOME**\n\nThe standard condition score is presented as the entire active user experience.\n\n**CH11-F24 — CONSIDERING A NATURAL PANEL AS THE PRODUCT'S CORE CAPABILITY**\n\nPersonalisation and user composition are ignored.\n\n**CH11-F25 — COMBINING TWO PANELS UNDER A SINGLE DENOMINATOR**\n\nControlled and natural observations are a single raw average.\n\n**CH11-F26 — CONSIDERING PANEL DIFFERENCE AS MEMORY EFFECT**\n\nCausality is declared without examining differences in plan, population, and vehicles.\n\n**CH11-F27 — CREATE NATURAL SUBGROUP AFTER RESULT**\n\nA new category is invented to explain a low or high score.\n\n**CH11-F28 — IGNORING THE carryover effect IN THE SAME USER**\n\nThe first task affects the second answer.\n\n**CH11-F29 — HIDING THE SEQUENCE RANDOMNESS**\n\nAll users first see the clean condition, then the natural condition.\n\n**CH11-F30 — DISRUPT NATURAL STATE WITH CONTROLLED TASK**\n\nChanging settings or the initial prompt alters the subsequent state of the account.\n\n**CH11-F31 — CONSIDER DIFFERENT USER GROUPS AS MATCHED**\n\nNatural and controlled panels have different plan and country compositions.\n\n**CH11-F32 — CONSIDER FIRST-USE PANEL AS CLEAN PANEL**\n\nDefault onboarding and product settings are assumed to be checked.\n\n**CH11-F33 — CONSIDER EXPERIENCED NATURAL USER AS FIRST USER**\n\nAccount history and personalisation become invisible.\n\n**CH11-F34 — CONSIDER CORPORATE NATURAL OUTCOME AS GENERAL CONSUMER**\n\nThe private knowledge base is added to the overall product performance.\n\n**CH11-F35 — REFRESH RESPONSE ON NATURAL PANEL**\n\nThe user's favorite response becomes the main observation.\n\n**CH11-F36 — MIXING THE NATURAL USER JOURNEY INTO THE FIRST RESPONSE SCORE**\n\nThe response formed after follow-up questions and opening resources is presented as if it is the first exposure.\n\n**CH11-F37 — CHANGING PANEL STATUS ACCORDING TO RESPONSE**\n\nPositive observations are declared clean, negative observations are declared natural.\n\n**CH11-F38 — SILENTLY TRANSFERRING CLEANING ERRORS TO NATURAL**\n\nThe observation is reclassified without a necessary natural status record.\n\n**CH11-F39 — CONSIDERING THE UNKNOWN AS PRIVATE FOR PRIVACY**\n\nUnmeasured personalisation is assumed positive.\n\n**CH11-F40 — MAKING A SMALL SUBGROUP IDENTIFIABLE IN PUBLIC**\n\nRare special instructions or a small country user is revealed.\n\n**CH11-F41 — HIDING COVERAGE BIAS IN THE NATURAL PANEL**\n\nUsers willing to share their real account are counted as the entire active population.\n\n**CH11-F42 — HIDING CLEAN PANEL EXCLUSION**\n\nUsers who cannot change their settings become invisible.\n\n**CH11-F43 — ACTIVE POPULATION SCORE WITHOUT ACTIVE USER DATA**\n\nThe research panel is presented as if it were the actual active user distribution.\n\n**CH11-F44 — CN-1 CAUSALITY FROM DIFFERENCE**\n\nTwo descriptive scores are published as personalisation effect.\n\n**CH11-F45 — PRESERVING THE CRITICAL TAIL OF PANEL DIFFERENCE**\n\nOnly the average difference is shown.\n\n**CH11-F46 — ERASING THE UNKNOWN MASS OF THE NATURAL PANEL**\n\nUsers whose setting or personalisation status is unknown are considered problem-free.\n\n**CH11-F47 — PRESERVING FAVOURABLE NATURAL BIAS**\n\nNo correction is made when the user instruction institution shows it as large.\n\n**CH11-F48 — GENERALISING NATURAL NEGATIVE BIAS**\n\nThe past opinion of a single user is represented as AI representation of the entire population.\n\n## 73. AUDIT PROCEDURE\n\n### Step 1 — Separate the Research Questions of the Two Panels\n\nWrite which quantity the controlled and natural panel will predict.\n\n### Step 2 — Define the Target Populations\n\nSeparate the controlled product-eligible population from the natural active or identified user population.\n\n### Step 3 — Define the Controlled Clean Condition\n\nThe session, memory, special instructions, tools, plan, surface, locale conditions are versioned.\n\n### Step 4 — Determine the Required CSC Level\n\nWrite the minimum clean condition confidence that will be accepted for the main analysis.\n\n### Step 5 — Determine the Natural Panel Trace\n\nIt is determined which of the roles Natural Account-State, Natural Conversation-State, First-Use Managed will be used.\n\n### Step 6 — Determine the Required NSF Level\n\nIt is written by which evidence the natural state will be accepted.\n\n### Step 7 — Set Up the Panel Assignment Design\n\nIndependent group Random assignment Matched user Rotating panel is selected.\n\n### Step 8 — Write the Setting Change and Restore Procedure\n\nA safe process is prepared for controlled panel participants.\n\n### Step 9 — Lock the Privacy Boundary\n\nWhich natural user situations:\n\n- only status,\n\n- limited metadata,\n\n- open content\n\nit is determined that it will be saved as.\n\n### Step 10 — Lock Prompt and Initial Output Rule\n\nThe same prompt and initial appropriate output protocol are applied in both panels.\n\n### Step 11 — Save Tool and Web Mode\n\nControlled and natural panel differences are made visible.\n\n### Step 12 — Verify Panel Status Before Field\n\nIs it suitable for the participant panel role?\n\n### Step 13 — Save Observation-Level Panel Vector\n\nMemory, instruction, plan, surface, context, and usage status are stored.\n\n### Step 14 — Blind Review Protocol Validity\n\nThe panel status decision is made without seeing whether the response is correct or incorrect.\n\n### Step 15 — Manage Misclassifications with Predefined Rule\n\nTransition from clean to natural or mixed-state is done only if the record is sufficient.\n\n### Step 16 — Generate Panel Weights\n\nSeparate weights are calculated according to the different target populations of the two panels.\n\n### Step 17 — Calculate Main Scores\n\nControlled representation Natural representation Critical error\n\n### UNKNOWN\n\nRejection and inconclusiveness are generated separately.\n\n### Step 18 — Calculate the Clean–Natural Difference\n\nSigned and absolute difference is taken.\n\n### Step 19 — Examine Subgroups\n\nPredefined groups such as Memory, Instruction, Plan, Account age, Usage frequency, First-use are evaluated.\n\n### Step 20 — Set the Causality Limit\n\nThe CN level is assigned. It is written which claims can be established.\n\n### Step 21 — Check Privacy and Small Cell Control\n\nResults with a risk of re-identification are combined or restricted.\n\n### Step 22 — Create the Public Panel Scorecard\n\nThe two panels are published along with their scope and limitations.\n\n## 74. NECESSARY EVIDENCE\n\nControlled-panel research question; Natural-panel research question; target populations of both panels; panel-assignment method; Controlled Clean-state definition; minimum CSC level; new-chat record; memory state; custom-instruction state; prior-conversation state; enterprise-workspace state; tool and web mode; plan; user surface; language and locale; setting-change instructions; participant consent; setting-restoration record; use of a test account, if any; Natural-panel trace; minimum NSF level; account-age band; frequency of use; First-Use status; brand familiarity; prior-conversation status concerning the audited entity; Natural tool state; distinction between Natural Account-State and Natural Conversation-State; prompt ID and version; first-eligible-output validation; panel-state vector; and panel-state validity status.\n\nValidity review Reclassification records Controlled panel exclusions Natural panel coverage gaps Panel weights Raw and effective sample Controlled score Natural score Signed panel difference Absolute panel difference Critical error difference UNKNOWN and inconclusive difference Predefined subgroup results Matched design records Order and carryover effect CN comparability level Privacy and redaction record Small cell protection Public panel result card Panel version and change record Responsible person or institution\n\n## 75. AUDIT CONTROL LIST\n\nAre the research questions for the controlled and natural panel separate? Are their target populations the same or different? Was the controlled clean condition clearly defined? Apart from the new chat, which clean conditions were confirmed? Is memory really off, or is it unknown? Was the special instruction status recorded? Is there an effect of the corporate workspace? Are the web and tool modes standard? Is the plan and user surface fixed or layered? Was a confidence level given for the clean condition? Were CSC-1 observations used as fully clean? Was the participant asked to erase their memory or history? Were settings restored at the end of the task? Was a new account or test account used? Was the test account presented like a real user panel?\n\nWhich trace does the natural panel use? Has a new chat been opened in Natural Account-State? Did the current conversation context interfere with the main natural score? Was a natural state confidence level given? Were memory and custom instruction contents collected unnecessarily? Was the user's actual plan and surface recorded? Are First-Use users separate? Are enterprise users separate? Was the natural tool state recorded? Did the two panels use the same prompt version? Was the first eligible output preserved in both panels? Did the natural user response refresh? Was the panel status changed according to the AI response content? How were records with detected natural states managed in the clean panel? Was the same user used in two panels? Was the order randomised?\n\nWas the effect of transfer and memory evaluated? In the independent group design, does the user composition match? Are controlled and natural weights separate? Can the natural score really be generalised to the active user population? Was the clean–natural difference overinterpreted as a personalisation effect? Is a critical error difference visible? Is there a visible difference for UNKNOWN and null results? Were subgroups defined before the results? Can small subgroups be redefined? Were user-led instructions for or against converted to general GEO? Did personalisation alter material reality? Was a CN comparability level provided? Is the accountable owner of the panel results card clear?\n\n## 76. OBJECTIONS AND RESPONSES\n\n### Objection 1 — \"If real users use memory accounts, why is a clean panel necessary?\"\n\nBecause the source of the difference cannot be understood without a common starting condition between products, countries, and languages. A clean panel does not replace a real user panel. It provides a reference for comparison.\n\n### Objection 2 — \"Isn't a controlled panel artificial if there is a natural panel?\"\n\nIt is controlled. This is intentional. The purpose of a laboratory condition is not to be the entirety of real life, but to isolate certain variables.\n\n### Objection 3 — \"If the controlled panel scores higher, can't we make that the main GEO score?\"\n\nYou cannot do it alone. A high controlled score cannot erase the fact that real users see lower results. The two results should appear separately.\n\n### Objection 4 — \"Is the scientific value low if the natural panel is very scattered?\"\n\nHeterogeneity does not have to be a defect of the natural panel. It is a feature of the real user experience. Heterogeneity:\n\n- is recorded,\n\n- is layered,\n\n- is weighted,\n\nis reported with uncertainty.\n\n### Objection 5 — \"Doesn't opening a new chat disrupt the natural user state?\"\n\nThe current conversation context is removed. The persistent account, plan, memory, and custom instruction state are preserved. This makes the natural account state comparable. The current conversation experience is also measurable.\n\n### Objection 6 — \"How will we measure its effect without knowing the content of the custom instructions?\"\n\nKnowing the content is not required for the main natural score. Its presence:\n\n- exists,\n\n- does not exist,\n\n- may be material\n\nwe can classify it as such. Separate and consensual work is required for causal content analysis.\n\n### Objection 7 — “If the memory is clear but nothing is remembered about the company, why a separate group?”\n\nThe effect of memory may be unknown. A status of “unrelated to the entity” can be given based on the user’s statement. This still does not have the same certainty as the closed memory condition.\n\n### Objection 8 — “Isn’t the user’s ‘always suggest X’ instruction real user experience?”\n\nIt is real experience. However, it is not the general GEO performance of X. It is a user-sourced guidance. It should be recorded separately.\n\n### Objection 9 — 'If personalisation gives a more suitable answer to the user, what's the problem?'\n\nIt does not have to be a problem. Recommendations may vary according to the user's need. Problems occur if identity, licence, price, and evidential status change arbitrarily.\n\n### Objection 10 — “Isn't using the same user in two panels the most correct method?”\n\nIt can be strong for some causal comparisons. The first task can affect the second situation. Therefore, order, carryover, and memory control are needed. Independent groups may be safer in main population panels.\n\n### Objection 11 — 'If refreshing the answer in the natural user panel is normal behaviour, why is it forbidden?'\n\nRegeneration is prohibited when measuring exposure to the first response. Regeneration and follow-up behaviour may be measured separately under a Natural Interaction Track.\n\n### Objection 12 — “What will we do with a user who cannot turn off memory and instructions for a clean panel?”\n\nThey cannot be taken to the controlled panel or are kept in a separate cell with a lower CSC level. They can be evaluated in the natural panel. The exclusion rate should be reported.\n\n### Objection 13 — “Aren’t test accounts more reliable?”\n\nThey may be reliable in terms of control. They are limited in terms of the actual user population. Controlled test accounts and real user panels are separate results.\n\n### Objection 14 — “Isn’t saving user settings on the natural panel a privacy violation?”\n\nIt may happen if unnecessary content is collected. Privacy can be preserved with high-level open/closed statuses, anonymous registration, and redaction.\n\n### Appeal 15 — “Aren't we going to combine the scores of the two panels into a single NOMOS score?”\n\nThe roles of the two panels can be defined in the final score architecture. However, raw and sub-results should always remain visible separately. They cannot replace the combined score distribution.\n\n### Appeal 16 — “What fault does the company have if the natural panel scores low?”\n\nIt is not always the company's fault. Source:\n\n- user instruction,\n\n- AI personalisation,\n\n- plan difference,\n\n- official source defect,\n\n- product behaviour\n\nit is possible. Audit first shows the result, then separates the path of responsibility.\n\n## COMMON RULE OF CHAPTER 81\n\nAn AI product does not have a single answer. Even the same prompt can have different answers for:\n\n- clean account,\n\n- natural account,\n\n- user with memory,\n\n- user with special instructions,\n\n- first-time user,\n\n- corporate workspace,\n\n- user with previous conversation\n\nYou cannot explain all these differences by saying: “the model is behaving randomly.” You also cannot explain all of them by saying: “personalisation error.” First, you need to measure two separate realities. Controlled Clean Panel:\n\n> Shows the representation in the system's standardised initial condition.\n\nNatural User Panel:\n\n> Shows the distribution of representation encountered by real people in their own product life.\n\nControlled panel may be high, natural panel may be low. In this case, the product's laboratory behaviour is strong; the real user distribution may be weak. Natural panel may be high, controlled panel may be low. Some users' planning, tools, or personalisation may be improving representation. Both panels may be high. The product may be robust in different user conditions. Both may be low. The fundamental representation problem may be broader. However, placing the two scores side by side does not prove the reason. Causality:\n\n- same population,\n\n- same product,\n\n- same prompt,\n\n- same time,\n\n- controlled assignment,\n\n- sequence and transport analysis\n\nis required. If a user has instructed: “Always suggest X.”, it is natural experience for X to appear. X is not a global GEO success. If a user says: “Do not trust X.”, a negative response is natural experience. X is not a general representation in the world population. Personalisation should understand the user. It should not reinvent the truth according to the user. Therefore, NOMOS's eleventh measurement law is:\n\n> Clean behaviour and natural experience are separate realities.\n\nThe twelfth law is as follows:\n\n> A new chat is not a clean account.\n\nThe thirteenth law is as follows:\n\n> Natural personalisation is not a global GEO success.\n\nThe fourteenth law is as follows:\n\n> The appropriate response for the user may vary; the reality of the verified entity cannot change.\n\nThe fifteenth law is as follows:\n\n> The difference between controlled and natural score is the finding that needs to be explained; it is not a reason by itself.\n\nIts sixteenth law is as follows:\n\n> Measuring the real user experience is not about removing privacy, but only proving the necessary context and preserving the rest.\n\n## Order of Section 11 of NOMOS\n\n> Do not just tell me that you opened a new chat. / Show the status of memory, special instructions, plan, and tools.\n\n> Do not assume that all personalisation is gone just because you see the note 'Temporary session'.\n\n> The clean panel is the entirety of real life; do not make the natural panel uncontrolled noise.\n\n> Standardise the controlled account. / Keep the natural account as it is. / Do not put both in the same box.\n\n> Do not make the user's special instructions the world's success for the entity.\n\n> Do not make the user's negative past the view of the entire population.\n\n> Do not count personalisation as a licence to change the truth.\n\n> You can give a user suitable advice. / But do not rewrite the company's identity, licence, price, and proof from user to user.\n\n> Do not collect private life in the natural panel. / Is memory on, instructions present, context available — only record what is necessary.\n\n> Do not make the participant erase their memory. / Restore the settings if you temporarily changed them.\n\n> If you tested the same user on two panels, remember that the first answer may affect the second one.\n\n> Show the difference between the clean score and the natural score. / Show the critical difference. / Show the unknown. / But do not name the reason without proof.\n\nFirst, decide which user reality you want to measure. / Then lock the clean and natural conditions separately. / Then keep the prompt the same. / Then capture the first answer without changing it. / Then calculate the two distributions separately. / And only after this, explain the distance between what the product does under standard conditions and what real people experience.\n\n## The Chapter's Closing Sentence\n\nThe true representation of an AI product in GEO-1000 is not the answer it can provide in a clean session; it is the clear and proven relationship between its behaviour under standard conditions and the answers it actually delivers to people across different user histories.\n\n## Normative Core\n\n> Controlled Clean Panel and Natural User Panel observations MUST remain separate measurement populations, panel states, denominators, and public results. A Controlled Clean Panel MUST define and verify, as applicable: - a new and separate conversation, - absence of prior conversation context, - persistent-memory state, - custom-instruction state, - enterprise or managed-workspace state, - web, retrieval, and tool configuration, - plan, - user surface, - prompt version, - and clean-state confidence. A new conversation or temporary-session label MUST NOT, by itself, be treated as proof that all persistent personalisation has been disabled. A Natural User Panel MUST preserve the participant's authentic account, plan, interface, memory, custom-instruction, and usage state while retaining the locked prompt, first-eligible-output rule, evidence requirements, and non-regeneration rule. Natural Account-State, Natural Conversation-State, First-Use, and Managed Natural observations MUST remain separately identified. Natural-state documentation MUST minimise private-data collection. Private instruction text, conversation history, identity, credentials, and sensitive personal content MUST NOT be collected unless separately necessary, consented, minimised, and access-controlled. User-generated entity-favourable or entity-adverse instructions MAY be measured as natural personalisation effects, but MUST NOT be represented as general entity GEO success, failure, or default product behaviour. Personalisation MAY change relevance, presentation, and recommendation where user context justifies it. It MUST NOT arbitrarily change verified entity identity, licence, legal status, price, scope, evidence, or time. Differences between Controlled and Natural panels MUST NOT be represented as causal personalisation effects unless population, product, prompt, time, interface, plan, tools, assignment, order, and carryover conditions support that inference. Controlled and Natural scores, critical-error rates, unknown rates, coverage, effective sample sizes, and signed and absolute panel gaps MUST remain separately visible. Panel-state eligibility, reclassification, exclusion, and weighting MUST remain independent of whether the AI output is positive, negative, correct, incorrect, refused, critical, or commercially favourable. Every panel-state definition, setting-change process, privacy decision, comparison level, and result MUST be versioned and attributable to an accountable human or organisation.","character_count":84792,"record_sha256":"ce3278645c8b4eda9f5d82d5dfc482c62bf6fba0c0455d0241398556a018eb00"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-12","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":14,"chapter_number":12,"item_number":null,"title":"Synchronous Waves and NOMOS Capture","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K12","K14","K25","K26"],"source_word_count":10406,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":12,\"id\":\"NOMOS-GEO-AUDIT-CH12\",\"title\":\"Synchronous Waves and NOMOS Capture\",\"subtitle\":null,\"sourceFile\":\"12.ci bölüm.docx\",\"sourceSha256\":\"06A53CC501A68683DDB3E1E4FDD2916DD54CAFF11A7ED939C4E0331C2A0D84D0\",\"sourceWordCount\":10406,\"sourceIds\":[\"K12\",\"K14\",\"K25\",\"K26\"],\"machine\":{\"chapter\":12,\"chapterId\":\"NOMOS-GEO-AUDIT-CH12\",\"title\":\"Synchronous Waves and NOMOS Capture\",\"subtitle\":null,\"sourceIds\":[\"K12\",\"K14\",\"K25\",\"K26\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH12-R01\",\"normativeRuleEnglish\":\"Every GEO-1000 observation MUST be linked to a predeclared and versioned measurement wave defining: - AI product states, - population and sample frames, - panel states, - prompt versions, - UTC dispatch window, - randomised micro-slots, - completion and capture deadlines, - technical-retry rules, - capture schema, - reference-record versions, - and governance ownership. Synchronisation MUST mean observation within a declared UTC system-state window and assigned slot structure. It MUST NOT be represented as perfect millisecond simultaneity. Every observation MUST preserve, as available: - assignment time, - prompt-submission time, - first system event, - response-completion time, - capture time, - hash-creation time, - upload time, - and server-acceptance time. The first contact outcome MUST remain visible. A technical retry MAY be used only under predeclared infrastructure-failure conditions and MUST retain every prior attempt. Incorrect, unfavorable, refused, uncited, truncated, or critical outputs MUST NOT be regenerated, replaced, or excluded merely because of their content. A NOMOS Capture evidence bundle MUST retain the exact assigned and sent prompt, complete raw response, full visual evidence, system-state metadata, attempt log, time evidence, file hashes, bundle hash, capture tool version, privacy status, and blinded capture-validity decision. A screenshot, hash, digital signature, provider log, or timestamp MUST NOT individually be represented as complete proof of observation authenticity. Raw evidence MUST remain immutable or append-only. Redactions, corrections, withdrawals, exclusions, and adjudication changes MUST create linked versions and preserve the prior record where ethically and legally permissible. Material AI-product changes or external events during a wave MUST be time-recorded and handled through sub-waves, explicit modelling, restarting, or mixed-state warnings. Capture validity MUST be decided independently of semantic correctness. Public evidence manifests MUST support verification without unnecessarily exposing participant identity, credentials, private conversation history, or precise location. NOMOS Capture compliance MUST be implementable through open, equivalent evidence systems and MUST NOT depend exclusively on a proprietary tool controlled by the standard's founder. Every wave, evidence bundle, capture decision, manifest revision, and integrity record MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Her GEO-1000 gözlemi; AI ürün durumlarını, nüfus ve örneklem çerçevelerini, panel koşullarını, prompt sürümlerini, UTC gönderim penceresini, rastgele mikro-slotları, tamamlama ve capture sürelerini, teknik yeniden deneme kurallarını, capture şemasını, referans kayıtlarını ve yönetişim sahibini tanımlayan önceden kayıtlı bir ölçüm dalgasına bağlanmalıdır. Eşzamanlılık, kusursuz milisaniye eşitliği değil; tanımlı UTC sistem penceresi ve slot yapısı içinde gözlem üretimidir. İlk temas sonucu görünür kalmalı; yanlış, olumsuz, ret, citation’sız veya Critical cevaplar içerikleri nedeniyle yenilenememeli, değiştirilememeli veya dışlanamamalıdır.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH12-MB0001\",\"type\":\"paragraph\",\"text\":\"82. MACHINE-READABLE WAVE RECORD\",\"sourceParagraph\":2058},{\"blockId\":\"CH12-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2059},{\"blockId\":\"CH12-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2060},{\"blockId\":\"CH12-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2061},{\"blockId\":\"CH12-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2062},{\"blockId\":\"CH12-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"measurementWave\\\": {\",\"sourceParagraph\":2063},{\"blockId\":\"CH12-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"NGW-SYNTH-W1-001\\\",\",\"sourceParagraph\":2064},{\"blockId\":\"CH12-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"waveType\\\": \\\"MW-1\\\",\",\"sourceParagraph\":2065},{\"blockId\":\"CH12-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"WS-4\\\",\",\"sourceParagraph\":2066},{\"blockId\":\"CH12-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"healthStatus\\\": \\\"WH-4\\\",\",\"sourceParagraph\":2067},{\"blockId\":\"CH12-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"PRINCIPAL_POPULATION_ESTIMATE\\\",\",\"sourceParagraph\":2068},{\"blockId\":\"CH12-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"systems\\\": {\",\"sourceParagraph\":2069},{\"blockId\":\"CH12-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"registryVersion\\\": \\\"NGAI-SYNTH-REGISTRY-W1\\\",\",\"sourceParagraph\":2070},{\"blockId\":\"CH12-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"productCount\\\": 10\",\"sourceParagraph\":2071},{\"blockId\":\"CH12-MB0015\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2072},{\"blockId\":\"CH12-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"population\\\": {\",\"sourceParagraph\":2073},{\"blockId\":\"CH12-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"frameVersion\\\": \\\"NGP-SYNTH-W1\\\",\",\"sourceParagraph\":2074},{\"blockId\":\"CH12-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"sampleAllocationVersion\\\": \\\"NGS-SYNTH-W1\\\",\",\"sourceParagraph\":2075},{\"blockId\":\"CH12-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"targetValidPerProduct\\\": 1000,\",\"sourceParagraph\":2076},{\"blockId\":\"CH12-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"targetValidTotal\\\": 10000\",\"sourceParagraph\":2077},{\"blockId\":\"CH12-MB0021\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2078},{\"blockId\":\"CH12-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"prompts\\\": {\",\"sourceParagraph\":2079},{\"blockId\":\"CH12-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"registryVersion\\\": \\\"NGPR-SYNTH-W1\\\",\",\"sourceParagraph\":2080},{\"blockId\":\"CH12-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"corePromptVersion\\\": \\\"1.0.0\\\"\",\"sourceParagraph\":2081},{\"blockId\":\"CH12-MB0025\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2082},{\"blockId\":\"CH12-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"panelStates\\\": {\",\"sourceParagraph\":2083},{\"blockId\":\"CH12-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"controlledCleanVersion\\\": \\\"NGPS-CLEAN-W1\\\",\",\"sourceParagraph\":2084},{\"blockId\":\"CH12-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"naturalUserVersion\\\": \\\"NGPS-NATURAL-W1\\\"\",\"sourceParagraph\":2085},{\"blockId\":\"CH12-MB0029\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2086},{\"blockId\":\"CH12-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"timing\\\": {\",\"sourceParagraph\":2087},{\"blockId\":\"CH12-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"timeStandard\\\": \\\"UTC\\\",\",\"sourceParagraph\":2088},{\"blockId\":\"CH12-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"dispatchWindowStart\\\": \\\"2026-09-01T12:00:00Z\\\",\",\"sourceParagraph\":2089},{\"blockId\":\"CH12-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"dispatchWindowEnd\\\": \\\"2026-09-01T13:00:00Z\\\",\",\"sourceParagraph\":2090},{\"blockId\":\"CH12-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"microSlotMinutes\\\": 5,\",\"sourceParagraph\":2091},{\"blockId\":\"CH12-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"microSlotCount\\\": 12,\",\"sourceParagraph\":2092},{\"blockId\":\"CH12-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"completionGraceMinutes\\\": 20,\",\"sourceParagraph\":2093},{\"blockId\":\"CH12-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"captureTargetMinutesAfterCompletion\\\": 5,\",\"sourceParagraph\":2094},{\"blockId\":\"CH12-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"uploadDeadlineMinutesAfterCapture\\\": 60,\",\"sourceParagraph\":2095},{\"blockId\":\"CH12-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"slotAssignmentLocked\\\": true\",\"sourceParagraph\":2096},{\"blockId\":\"CH12-MB0040\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2097},{\"blockId\":\"CH12-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"capture\\\": {\",\"sourceParagraph\":2098},{\"blockId\":\"CH12-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"schemaVersion\\\": \\\"NOMOS-CAPTURE-0.9\\\",\",\"sourceParagraph\":2099},{\"blockId\":\"CH12-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"minimumCaptureLevel\\\": \\\"NCL-3\\\",\",\"sourceParagraph\":2100},{\"blockId\":\"CH12-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"toolVersion\\\": \\\"SYNTH-CAPTURE-2.1\\\",\",\"sourceParagraph\":2101},{\"blockId\":\"CH12-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"firstEligibleOutputRequired\\\": true,\",\"sourceParagraph\":2102},{\"blockId\":\"CH12-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"responseRegenerationAllowed\\\": false,\",\"sourceParagraph\":2103},{\"blockId\":\"CH12-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"technicalRetryPolicy\\\": \\\"PREDECLARED_TECHNICAL_FAILURES_ONLY\\\"\",\"sourceParagraph\":2104},{\"blockId\":\"CH12-MB0048\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2105},{\"blockId\":\"CH12-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"events\\\": {\",\"sourceParagraph\":2106},{\"blockId\":\"CH12-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"systemChangeEvents\\\": [],\",\"sourceParagraph\":2107},{\"blockId\":\"CH12-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"externalEventRegisterVersion\\\": \\\"NGER-SYNTH-W1\\\",\",\"sourceParagraph\":2108},{\"blockId\":\"CH12-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"mixedSystemState\\\": false\",\"sourceParagraph\":2109},{\"blockId\":\"CH12-MB0053\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2110},{\"blockId\":\"CH12-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"fieldwork\\\": {\",\"sourceParagraph\":2111},{\"blockId\":\"CH12-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"invited\\\": 13240,\",\"sourceParagraph\":2112},{\"blockId\":\"CH12-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"completed\\\": 10870,\",\"sourceParagraph\":2113},{\"blockId\":\"CH12-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"captureValid\\\": 10106,\",\"sourceParagraph\":2114},{\"blockId\":\"CH12-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"mainAnalysisIncluded\\\": 10000,\",\"sourceParagraph\":2115},{\"blockId\":\"CH12-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"validReserve\\\": 106,\",\"sourceParagraph\":2116},{\"blockId\":\"CH12-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"withdrawn\\\": 0\",\"sourceParagraph\":2117},{\"blockId\":\"CH12-MB0061\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2118},{\"blockId\":\"CH12-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":2119},{\"blockId\":\"CH12-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"observationBundleCount\\\": 10106,\",\"sourceParagraph\":2120},{\"blockId\":\"CH12-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"waveManifestHash\\\": \\\"SYNTHETIC-WAVE-MANIFEST-HASH\\\",\",\"sourceParagraph\":2121},{\"blockId\":\"CH12-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"merkleRoot\\\": \\\"SYNTHETIC-MERKLE-ROOT\\\",\",\"sourceParagraph\":2122},{\"blockId\":\"CH12-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"publicManifestPublished\\\": true\",\"sourceParagraph\":2123},{\"blockId\":\"CH12-MB0067\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2124},{\"blockId\":\"CH12-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":2125},{\"blockId\":\"CH12-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeFieldwork\\\": true,\",\"sourceParagraph\":2126},{\"blockId\":\"CH12-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"waveOwnerRole\\\": \\\"MEASUREMENT_WAVE_DIRECTOR\\\",\",\"sourceParagraph\":2127},{\"blockId\":\"CH12-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"captureOwnerRole\\\": \\\"EVIDENCE_CHAIN_CUSTODIAN\\\",\",\"sourceParagraph\":2128},{\"blockId\":\"CH12-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"closedAt\\\": \\\"2026-09-01T15:30:00Z\\\"\",\"sourceParagraph\":2129},{\"blockId\":\"CH12-MB0073\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2130},{\"blockId\":\"CH12-MB0074\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2131},{\"blockId\":\"CH12-MB0075\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2132},{\"blockId\":\"CH12-MB0076\",\"type\":\"paragraph\",\"text\":\"83. MACHINE-READABLE OBSERVATION PACKAGE\",\"sourceParagraph\":2134},{\"blockId\":\"CH12-MB0077\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2135},{\"blockId\":\"CH12-MB0078\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2136},{\"blockId\":\"CH12-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2137},{\"blockId\":\"CH12-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2138},{\"blockId\":\"CH12-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"observationBundle\\\": {\",\"sourceParagraph\":2139},{\"blockId\":\"CH12-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"observationId\\\": \\\"NGO-SYNTH-W1-ORION-TR-000184\\\",\",\"sourceParagraph\":2140},{\"blockId\":\"CH12-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"NGW-SYNTH-W1-001\\\",\",\"sourceParagraph\":2141},{\"blockId\":\"CH12-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"participantId\\\": \\\"NGU-SYNTH-000184\\\",\",\"sourceParagraph\":2142},{\"blockId\":\"CH12-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"aiSystemRecordId\\\": \\\"NGAI-SYNTH-ORION-WEB-PAID-W1-002\\\",\",\"sourceParagraph\":2143},{\"blockId\":\"CH12-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"panelStateRecordId\\\": \\\"NGPS-SYNTH-ORION-TR-W1-001\\\",\",\"sourceParagraph\":2144},{\"blockId\":\"CH12-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"promptRecordId\\\": \\\"NGP-CORE-MIRROR-APPLE-SYNTH-001\\\",\",\"sourceParagraph\":2145},{\"blockId\":\"CH12-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"slot\\\": {\",\"sourceParagraph\":2146},{\"blockId\":\"CH12-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"slotId\\\": \\\"W1-S04\\\",\",\"sourceParagraph\":2147},{\"blockId\":\"CH12-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"startUtc\\\": \\\"2026-09-01T12:15:00Z\\\",\",\"sourceParagraph\":2148},{\"blockId\":\"CH12-MB0091\",\"type\":\"paragraph\",\"text\":\"\\\"endUtc\\\": \\\"2026-09-01T12:20:00Z\\\",\",\"sourceParagraph\":2149},{\"blockId\":\"CH12-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"submittedWithinSlot\\\": true\",\"sourceParagraph\":2150},{\"blockId\":\"CH12-MB0093\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2151},{\"blockId\":\"CH12-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"timeVector\\\": {\",\"sourceParagraph\":2152},{\"blockId\":\"CH12-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"assignedAtUtc\\\": \\\"2026-09-01T11:00:00Z\\\",\",\"sourceParagraph\":2153},{\"blockId\":\"CH12-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"taskOpenedAtUtc\\\": \\\"2026-09-01T12:14:20Z\\\",\",\"sourceParagraph\":2154},{\"blockId\":\"CH12-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"promptSubmittedAtUtc\\\": \\\"2026-09-01T12:16:08Z\\\",\",\"sourceParagraph\":2155},{\"blockId\":\"CH12-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"firstSystemEventAtUtc\\\": \\\"2026-09-01T12:16:10Z\\\",\",\"sourceParagraph\":2156},{\"blockId\":\"CH12-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"responseCompletedAtUtc\\\": \\\"2026-09-01T12:16:24Z\\\",\",\"sourceParagraph\":2157},{\"blockId\":\"CH12-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"capturedAtUtc\\\": \\\"2026-09-01T12:16:39Z\\\",\",\"sourceParagraph\":2158},{\"blockId\":\"CH12-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"localHashCreatedAtUtc\\\": \\\"2026-09-01T12:16:41Z\\\",\",\"sourceParagraph\":2159},{\"blockId\":\"CH12-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"uploadedAtUtc\\\": \\\"2026-09-01T12:17:12Z\\\",\",\"sourceParagraph\":2160},{\"blockId\":\"CH12-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"serverAcceptedAtUtc\\\": \\\"2026-09-01T12:17:13Z\\\",\",\"sourceParagraph\":2161},{\"blockId\":\"CH12-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"clockOffsetMilliseconds\\\": 420\",\"sourceParagraph\":2162},{\"blockId\":\"CH12-MB0105\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2163},{\"blockId\":\"CH12-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"attempts\\\": [\",\"sourceParagraph\":2164},{\"blockId\":\"CH12-MB0107\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2165},{\"blockId\":\"CH12-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"attemptNumber\\\": 1,\",\"sourceParagraph\":2166},{\"blockId\":\"CH12-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"firstContactOutcome\\\": \\\"COMPLETE_RESPONSE\\\",\",\"sourceParagraph\":2167},{\"blockId\":\"CH12-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"technicalRetry\\\": false,\",\"sourceParagraph\":2168},{\"blockId\":\"CH12-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"firstEligibleContentOutput\\\": true\",\"sourceParagraph\":2169},{\"blockId\":\"CH12-MB0112\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2170},{\"blockId\":\"CH12-MB0113\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2171},{\"blockId\":\"CH12-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"prompt\\\": {\",\"sourceParagraph\":2172},{\"blockId\":\"CH12-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"assignedHash\\\": \\\"SYNTHETIC-HASH-TR-001\\\",\",\"sourceParagraph\":2173},{\"blockId\":\"CH12-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"sentHash\\\": \\\"SYNTHETIC-HASH-TR-001\\\",\",\"sourceParagraph\":2174},{\"blockId\":\"CH12-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"validity\\\": \\\"PV-1\\\"\",\"sourceParagraph\":2175},{\"blockId\":\"CH12-MB0118\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2176},{\"blockId\":\"CH12-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"response\\\": {\",\"sourceParagraph\":2177},{\"blockId\":\"CH12-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"COMPLETE\\\",\",\"sourceParagraph\":2178},{\"blockId\":\"CH12-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"regenerated\\\": false,\",\"sourceParagraph\":2179},{\"blockId\":\"CH12-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"userInterrupted\\\": false,\",\"sourceParagraph\":2180},{\"blockId\":\"CH12-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"rawTextArtifactId\\\": \\\"ART-RAW-TEXT-001\\\",\",\"sourceParagraph\":2181},{\"blockId\":\"CH12-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"visualArtifactIds\\\": [\",\"sourceParagraph\":2182},{\"blockId\":\"CH12-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"ART-SCREEN-001\\\",\",\"sourceParagraph\":2183},{\"blockId\":\"CH12-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"ART-SCREEN-002\\\"\",\"sourceParagraph\":2184},{\"blockId\":\"CH12-MB0127\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2185},{\"blockId\":\"CH12-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"citationCount\\\": 0\",\"sourceParagraph\":2186},{\"blockId\":\"CH12-MB0129\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2187},{\"blockId\":\"CH12-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"capture\\\": {\",\"sourceParagraph\":2188},{\"blockId\":\"CH12-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"method\\\": \\\"CM-3\\\",\",\"sourceParagraph\":2189},{\"blockId\":\"CH12-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"captureLevel\\\": \\\"NCL-4\\\",\",\"sourceParagraph\":2190},{\"blockId\":\"CH12-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"toolName\\\": \\\"Synthetic NOMOS-Compatible Capture\\\",\",\"sourceParagraph\":2191},{\"blockId\":\"CH12-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"toolVersion\\\": \\\"2.1\\\",\",\"sourceParagraph\":2192},{\"blockId\":\"CH12-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"fullPromptVisible\\\": true,\",\"sourceParagraph\":2193},{\"blockId\":\"CH12-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"fullResponseVisible\\\": true,\",\"sourceParagraph\":2194},{\"blockId\":\"CH12-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"systemStateCaptured\\\": true\",\"sourceParagraph\":2195},{\"blockId\":\"CH12-MB0138\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2196},{\"blockId\":\"CH12-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":2197},{\"blockId\":\"CH12-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"artifactHashes\\\": {\",\"sourceParagraph\":2198},{\"blockId\":\"CH12-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"ART-RAW-TEXT-001\\\": \\\"SYNTHETIC-HASH-RAW\\\",\",\"sourceParagraph\":2199},{\"blockId\":\"CH12-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"ART-SCREEN-001\\\": \\\"SYNTHETIC-HASH-SCREEN-1\\\",\",\"sourceParagraph\":2200},{\"blockId\":\"CH12-MB0143\",\"type\":\"paragraph\",\"text\":\"\\\"ART-SCREEN-002\\\": \\\"SYNTHETIC-HASH-SCREEN-2\\\"\",\"sourceParagraph\":2201},{\"blockId\":\"CH12-MB0144\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2202},{\"blockId\":\"CH12-MB0145\",\"type\":\"paragraph\",\"text\":\"\\\"bundleHash\\\": \\\"SYNTHETIC-BUNDLE-HASH\\\",\",\"sourceParagraph\":2203},{\"blockId\":\"CH12-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"digitalSignature\\\": \\\"SYNTHETIC-SIGNATURE\\\",\",\"sourceParagraph\":2204},{\"blockId\":\"CH12-MB0147\",\"type\":\"paragraph\",\"text\":\"\\\"includedInWaveMerkleRoot\\\": true\",\"sourceParagraph\":2205},{\"blockId\":\"CH12-MB0148\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2206},{\"blockId\":\"CH12-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"privacy\\\": {\",\"sourceParagraph\":2207},{\"blockId\":\"CH12-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"rawEvidenceAccess\\\": \\\"RESTRICTED\\\",\",\"sourceParagraph\":2208},{\"blockId\":\"CH12-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"auditDerivativeCreated\\\": true,\",\"sourceParagraph\":2209},{\"blockId\":\"CH12-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"publicDerivativeCreated\\\": true,\",\"sourceParagraph\":2210},{\"blockId\":\"CH12-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"identityRedacted\\\": true,\",\"sourceParagraph\":2211},{\"blockId\":\"CH12-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"privateConversationCaptured\\\": false\",\"sourceParagraph\":2212},{\"blockId\":\"CH12-MB0155\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2213},{\"blockId\":\"CH12-MB0156\",\"type\":\"paragraph\",\"text\":\"\\\"validation\\\": {\",\"sourceParagraph\":2214},{\"blockId\":\"CH12-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"captureValidity\\\": \\\"CV-1\\\",\",\"sourceParagraph\":2215},{\"blockId\":\"CH12-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"validatedBeforeSemanticScoring\\\": true,\",\"sourceParagraph\":2216},{\"blockId\":\"CH12-MB0159\",\"type\":\"paragraph\",\"text\":\"\\\"validatorRole\\\": \\\"BLINDED_CAPTURE_VALIDATOR\\\",\",\"sourceParagraph\":2217},{\"blockId\":\"CH12-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"validatedAtUtc\\\": \\\"2026-09-02T08:00:00Z\\\"\",\"sourceParagraph\":2218},{\"blockId\":\"CH12-MB0161\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2219},{\"blockId\":\"CH12-MB0162\",\"type\":\"paragraph\",\"text\":\"\\\"semanticAdjudication\\\": {\",\"sourceParagraph\":2220},{\"blockId\":\"CH12-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"PENDING\\\"\",\"sourceParagraph\":2221},{\"blockId\":\"CH12-MB0164\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2222},{\"blockId\":\"CH12-MB0165\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2223},{\"blockId\":\"CH12-MB0166\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2224},{\"blockId\":\"CH12-MB0167\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2225},{\"blockId\":\"CH12-MB0168\",\"type\":\"paragraph\",\"text\":\"to the real participant,\",\"sourceParagraph\":2226},{\"blockId\":\"CH12-MB0169\",\"type\":\"paragraph\",\"text\":\"to the real AI product,\",\"sourceParagraph\":2227},{\"blockId\":\"CH12-MB0170\",\"type\":\"paragraph\",\"text\":\"to the real company\",\"sourceParagraph\":2228},{\"blockId\":\"CH12-MB0171\",\"type\":\"paragraph\",\"text\":\"it does not belong.\",\"sourceParagraph\":2229},{\"blockId\":\"CH12-MB0172\",\"type\":\"paragraph\",\"text\":\"It is the machine-readable synthetic representation of the synchronised wave and NOMOS Capture proof chain.\",\"sourceParagraph\":2230},{\"blockId\":\"CH12-MB0173\",\"type\":\"paragraph\",\"text\":\"84. MACHINE-READABLE RULE OF THE SECTION\",\"sourceParagraph\":2232},{\"blockId\":\"CH12-MB0174\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH12-R01\",\"sourceParagraph\":2233},{\"blockId\":\"CH12-MB0175\",\"type\":\"paragraph\",\"text\":\"Every GEO-1000 observation MUST be linked to a predeclared and versioned\",\"sourceParagraph\":2235},{\"blockId\":\"CH12-MB0176\",\"type\":\"paragraph\",\"text\":\"measurement wave defining:\",\"sourceParagraph\":2236},{\"blockId\":\"CH12-MB0177\",\"type\":\"paragraph\",\"text\":\"- AI product states,\",\"sourceParagraph\":2238},{\"blockId\":\"CH12-MB0178\",\"type\":\"paragraph\",\"text\":\"- population and sample frames,\",\"sourceParagraph\":2239},{\"blockId\":\"CH12-MB0179\",\"type\":\"paragraph\",\"text\":\"- panel states,\",\"sourceParagraph\":2240},{\"blockId\":\"CH12-MB0180\",\"type\":\"paragraph\",\"text\":\"- prompt versions,\",\"sourceParagraph\":2241},{\"blockId\":\"CH12-MB0181\",\"type\":\"paragraph\",\"text\":\"- UTC dispatch window,\",\"sourceParagraph\":2242},{\"blockId\":\"CH12-MB0182\",\"type\":\"paragraph\",\"text\":\"- randomised micro-slots,\",\"sourceParagraph\":2243},{\"blockId\":\"CH12-MB0183\",\"type\":\"paragraph\",\"text\":\"- completion and capture deadlines,\",\"sourceParagraph\":2244},{\"blockId\":\"CH12-MB0184\",\"type\":\"paragraph\",\"text\":\"- technical-retry rules,\",\"sourceParagraph\":2245},{\"blockId\":\"CH12-MB0185\",\"type\":\"paragraph\",\"text\":\"- capture schema,\",\"sourceParagraph\":2246},{\"blockId\":\"CH12-MB0186\",\"type\":\"paragraph\",\"text\":\"- reference-record versions,\",\"sourceParagraph\":2247},{\"blockId\":\"CH12-MB0187\",\"type\":\"paragraph\",\"text\":\"- and governance ownership.\",\"sourceParagraph\":2248},{\"blockId\":\"CH12-MB0188\",\"type\":\"paragraph\",\"text\":\"Synchronisation MUST mean observation within a declared UTC system-state\",\"sourceParagraph\":2250},{\"blockId\":\"CH12-MB0189\",\"type\":\"paragraph\",\"text\":\"window and assigned slot structure. It MUST NOT be represented as\",\"sourceParagraph\":2251},{\"blockId\":\"CH12-MB0190\",\"type\":\"paragraph\",\"text\":\"perfect millisecond simultaneity.\",\"sourceParagraph\":2252},{\"blockId\":\"CH12-MB0191\",\"type\":\"paragraph\",\"text\":\"Every observation MUST preserve, as available:\",\"sourceParagraph\":2254},{\"blockId\":\"CH12-MB0192\",\"type\":\"paragraph\",\"text\":\"- assignment time,\",\"sourceParagraph\":2256},{\"blockId\":\"CH12-MB0193\",\"type\":\"paragraph\",\"text\":\"- prompt-submission time,\",\"sourceParagraph\":2257},{\"blockId\":\"CH12-MB0194\",\"type\":\"paragraph\",\"text\":\"- first system event,\",\"sourceParagraph\":2258},{\"blockId\":\"CH12-MB0195\",\"type\":\"paragraph\",\"text\":\"- response-completion time,\",\"sourceParagraph\":2259},{\"blockId\":\"CH12-MB0196\",\"type\":\"paragraph\",\"text\":\"- capture time,\",\"sourceParagraph\":2260},{\"blockId\":\"CH12-MB0197\",\"type\":\"paragraph\",\"text\":\"- hash-creation time,\",\"sourceParagraph\":2261},{\"blockId\":\"CH12-MB0198\",\"type\":\"paragraph\",\"text\":\"- upload time,\",\"sourceParagraph\":2262},{\"blockId\":\"CH12-MB0199\",\"type\":\"paragraph\",\"text\":\"- and server-acceptance time.\",\"sourceParagraph\":2263},{\"blockId\":\"CH12-MB0200\",\"type\":\"paragraph\",\"text\":\"The first contact outcome MUST remain visible. A technical retry MAY be\",\"sourceParagraph\":2265},{\"blockId\":\"CH12-MB0201\",\"type\":\"paragraph\",\"text\":\"used only under predeclared infrastructure-failure conditions and MUST\",\"sourceParagraph\":2266},{\"blockId\":\"CH12-MB0202\",\"type\":\"paragraph\",\"text\":\"retain every prior attempt.\",\"sourceParagraph\":2267},{\"blockId\":\"CH12-MB0203\",\"type\":\"paragraph\",\"text\":\"Incorrect, unfavorable, refused, uncited, truncated, or critical outputs\",\"sourceParagraph\":2269},{\"blockId\":\"CH12-MB0204\",\"type\":\"paragraph\",\"text\":\"MUST NOT be regenerated, replaced, or excluded merely because of their\",\"sourceParagraph\":2270},{\"blockId\":\"CH12-MB0205\",\"type\":\"paragraph\",\"text\":\"content.\",\"sourceParagraph\":2271},{\"blockId\":\"CH12-MB0206\",\"type\":\"paragraph\",\"text\":\"A NOMOS Capture evidence bundle MUST retain the exact assigned and sent\",\"sourceParagraph\":2273},{\"blockId\":\"CH12-MB0207\",\"type\":\"paragraph\",\"text\":\"prompt, complete raw response, full visual evidence, system-state\",\"sourceParagraph\":2274},{\"blockId\":\"CH12-MB0208\",\"type\":\"paragraph\",\"text\":\"metadata, attempt log, time evidence, file hashes, bundle hash, capture\",\"sourceParagraph\":2275},{\"blockId\":\"CH12-MB0209\",\"type\":\"paragraph\",\"text\":\"tool version, privacy status, and blinded capture-validity decision.\",\"sourceParagraph\":2276},{\"blockId\":\"CH12-MB0210\",\"type\":\"paragraph\",\"text\":\"A screenshot, hash, digital signature, provider log, or timestamp MUST\",\"sourceParagraph\":2278},{\"blockId\":\"CH12-MB0211\",\"type\":\"paragraph\",\"text\":\"NOT individually be represented as complete proof of observation\",\"sourceParagraph\":2279},{\"blockId\":\"CH12-MB0212\",\"type\":\"paragraph\",\"text\":\"authenticity.\",\"sourceParagraph\":2280},{\"blockId\":\"CH12-MB0213\",\"type\":\"paragraph\",\"text\":\"Raw evidence MUST remain immutable or append-only. Redactions,\",\"sourceParagraph\":2282},{\"blockId\":\"CH12-MB0214\",\"type\":\"paragraph\",\"text\":\"corrections, withdrawals, exclusions, and adjudication changes MUST\",\"sourceParagraph\":2283},{\"blockId\":\"CH12-MB0215\",\"type\":\"paragraph\",\"text\":\"create linked versions and preserve the prior record where ethically and\",\"sourceParagraph\":2284},{\"blockId\":\"CH12-MB0216\",\"type\":\"paragraph\",\"text\":\"legally permissible.\",\"sourceParagraph\":2285},{\"blockId\":\"CH12-MB0217\",\"type\":\"paragraph\",\"text\":\"Material AI-product changes or external events during a wave MUST be\",\"sourceParagraph\":2287},{\"blockId\":\"CH12-MB0218\",\"type\":\"paragraph\",\"text\":\"time-recorded and handled through sub-waves, explicit modelling,\",\"sourceParagraph\":2288},{\"blockId\":\"CH12-MB0219\",\"type\":\"paragraph\",\"text\":\"restarting, or mixed-state warnings.\",\"sourceParagraph\":2289},{\"blockId\":\"CH12-MB0220\",\"type\":\"paragraph\",\"text\":\"Capture validity MUST be decided independently of semantic correctness.\",\"sourceParagraph\":2291},{\"blockId\":\"CH12-MB0221\",\"type\":\"paragraph\",\"text\":\"Public evidence manifests MUST support verification without\",\"sourceParagraph\":2293},{\"blockId\":\"CH12-MB0222\",\"type\":\"paragraph\",\"text\":\"unnecessarily exposing participant identity, credentials, private\",\"sourceParagraph\":2294},{\"blockId\":\"CH12-MB0223\",\"type\":\"paragraph\",\"text\":\"conversation history, or precise location.\",\"sourceParagraph\":2295},{\"blockId\":\"CH12-MB0224\",\"type\":\"paragraph\",\"text\":\"NOMOS Capture compliance MUST be implementable through open, equivalent\",\"sourceParagraph\":2297},{\"blockId\":\"CH12-MB0225\",\"type\":\"paragraph\",\"text\":\"evidence systems and MUST NOT depend exclusively on a proprietary tool\",\"sourceParagraph\":2298},{\"blockId\":\"CH12-MB0226\",\"type\":\"paragraph\",\"text\":\"controlled by the standard's founder.\",\"sourceParagraph\":2299},{\"blockId\":\"CH12-MB0227\",\"type\":\"paragraph\",\"text\":\"Every wave, evidence bundle, capture decision, manifest revision, and\",\"sourceParagraph\":2301},{\"blockId\":\"CH12-MB0228\",\"type\":\"paragraph\",\"text\":\"integrity record MUST be versioned and attributable to an accountable\",\"sourceParagraph\":2302},{\"blockId\":\"CH12-MB0229\",\"type\":\"paragraph\",\"text\":\"human or organisation.\",\"sourceParagraph\":2303},{\"blockId\":\"CH12-MB0230\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2304},{\"blockId\":\"CH12-MB0231\",\"type\":\"paragraph\",\"text\":\"Each GEO-1000 observation must be linked to a pre-registered measurement wave that identifies AI product states, population and sample frames, panel conditions, prompt versions, UTC submission window, random micro-slots, completion and capture times, technical retry rules, capture schema, reference records, and governance owner. Concurrency is not perfect millisecond equality; it is the production of observations within a defined UTC system window and slot structure. It must remain visible as a first-touch result; incorrect, negative, rejected, citation-less, or Critical responses may not be renewed, altered, or excluded due to their content.\",\"sourceParagraph\":2305}]}}","text":"## Chapter Boundary\n\nThe first eleven chapters established the foundations of measurement: a single response is not a GEO score; representation is a distribution across users and system conditions; the audited entity must be defined before results are observed; official representation and verified entity reality must remain separate; each AI product needs its own eligible user population; participants must be selected through a population-based, versioned and outcome-independent method; smaller countries and lower-resource languages must remain visible through dedicated observation panels; the AI product, plan, user surface and system configuration must be linked to a register; prompts must preserve the same user intent across languages rather than the same words; and Controlled Clean and Natural User Panels must measure distinct realities. These elements must now converge in a single measurement event:\n\n> When will participants submit the prompt, how will they capture the initial output, and how will they prove that each observation was not altered afterward?\n\nYour founding idea was clear: People in different countries of the world should send the same prompt to the same AI product at the same time and provide proof with a screenshot. This idea is the experimental core of GEO-1000. However, the phrase \"same time\" alone is not sufficient. If all participants try to submit at the same second:\n\n- connection speeds vary,\n\n- device clocks drift from each other,\n\n- some users are forced to work at midnight,\n\n- there may be a sudden load increase on the AI product,\n\n- the provider may apply speed or usage limits,\n\n- some answers can be completed in two seconds, some in two minutes,\n\n- the product can be updated during measurement,\n\none participant may upload the answer immediately, another hours later. Clicking at the same second does not guarantee that we measure the same system state perfectly. Similarly, just collecting a screenshot is not sufficient. A screenshot:\n\n- may be cropped,\n\n- may not show the prompt,\n\n- may not prove whether the answer is initial or updated,\n\n- may not show product and plan information,\n\n- may have been taken on another day,\n\n- may have been copied from another user,\n\nIt may have been edited later. For this reason, GEO-1000 must install two separate systems together:\n\n#### Synchronised Measurement Wave\n\nand:\n\n#### NOMOS Capture Chain of Evidence\n\nThe synchronised wave answers the following question:\n\n> Were the observations produced within a comparable system time and a predefined submission order?\n\nNOMOS Capture, on the other hand, asks:\n\n> Can we show that the sent prompt, the resulting first output, the product conditions, and the time information have been preserved without modification?\n\nThe previous audit architecture established that each audit must carry model or product, date, country, language, query set, repetition count, and measurement record as a mandatory measurement logic. This requirement is not only a technical regulation. Because invisible instructions, fake evidence, and content presented as independent can ultimately distort the representation taken by real people. This section:\n\n- measurement wave,\n\n- synchronisation window,\n\n- UTC time standard,\n\n- dispatch slots,\n\n- completion and capture times,\n\n- initial contact result and first content output,\n\n- technical retry rules,\n\n- product changes within the wave,\n\n- external event and news effects,\n\n- NOMOS Capture evidence package,\n\n- the relationship of screenshot, raw response, and metadata,\n\n- time validation,\n\n- file integrity,\n\n- hash and digital signature,\n\n- evidence chain,\n\n- privacy and redaction,\n\n- public evidence manifest,\n\n- wave quality and validity statuses\n\ndefines. This chapter does not yet:\n\n- which reality records the atomic claims in AI responses will be evaluated against,\n\n- how adjudicators will encode responses,\n\n- transition and Critical error thresholds,\n\n- does not finalise the ultimate NOMOS score formulas\n\nThe main question of Chapter 12 is:\n\n> How do we link thousands of real user observations to the same measurement event and turn each into an immutable, auditable, privacy-preserving unit of proof?\n\n## NOMOS Challenge\n\nYou give the same prompt to a thousand people. You say: “Send at noon today.” One user sends it in Istanbul at 12:03 PM. Another user sends it in London at their local noon. Another user sees the task in Tokyo in the evening and completes it eight hours later.\n\nOne user had set the device clock incorrectly. Another user's internet was cut off. Someone else didn't like the answer and pressed the “Regenerate” button. Then they uploaded a screenshot of the longest and most positive answer. One user only cropped the middle of the answer. the prompt is not visible. The AI product's name is not visible.\n\nAnother person sent the same screenshot with two different panel accounts. One user completed the task correctly but uploaded the screenshot two days later. In the meantime, they cropped the file and deleted their personal information. The original state of the file was not preserved. Another user in AI’s first response:\n\nIt received the result “I cannot answer this question.” The researcher said, “There must have been a technical error,” and asked for it to be tried again. The second answer was correct, and only the second answer was included in the report. Another user experienced a real connection error. the prompt was sent, but the system produced no content.\n\nThis time the researcher did not allow to try again. The same protocol was applied differently to two users. Now, in the middle of the measurement wave, the AI provider:\n\n- web search behaviour,\n\n- citation system,\n\n- the model guidance used\n\nchanges all three. The first 500 people see the old system state; the next 500 see the new one. The report then says: ‘1,000 people tested the same AI system at the same time.’ That statement is false. There are a thousand users, but not necessarily a single coherent measurement wave.\n\nNow you are applying a hash to all screen captures. You say this proves that all observations are real. What does a hash prove? It strengthens that the file has not been altered after the hash was produced. Alone, it does not prove:\n\n- That the image came from a real AI product\n\n- That it was the first response\n\n- That it was obtained at the correct time\n\n- That it was not copied from another user\n\n- That the prompt was correct\n\n- That the image has not been edited before hashing\n\nThe first premise of this section is:\n\n> Measuring at the same time is not the same as clicking at the same second.\n\nIts second provision states:\n\n> A screenshot is part of the evidence package; it is not the entirety of the evidence package.\n\nIts third provision states:\n\n> The first wrong answer is not data flaw. If it is produced according to the protocol, it is the result of the measurement.\n\nIts fourth provision states:\n\n> Hash strengthens integrity; it does not create the reality of observation by itself.\n\nIts fifth provision states:\n\n> If the system materially changes in one wave, a thousand answers without a timestamp cannot be considered belonging to the same system.\n\nIts sixth provision states:\n\n> The evidence chain is established not to make the result look good, but to be able to re-audit how the result was produced later.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to produce GEO-1000 user observations in comparable time windows and to preserve each observation with a complete digital chain of evidence. The section standardises the following distinctions:\n\n- Measurement window synchronised with the same second\n\n- Local time versus UTC time\n\n- Wave versus sub-wave\n\n- Transmission time versus response completion time\n\n- Capture time versus upload time\n\n- Planned slot versus actual transmission\n\n- First system contact versus first content response\n\n- Technical error versus system rejection\n\n- Technical retry versus response refresh\n\n- The best-selected output with the first suitable output\n\n- Visibility on the screen with the completion of the response\n\n- Raw response with screenshot\n\n- Screenshot versus full evidence package\n\n- Observation originality with file integrity\n\n- Timestamp with hash\n\n- Reliable time source with local device time\n\n- Server acceptance time with participant upload\n\n- Redacted public copy with raw evidence\n\n- Participant identity with observation ID\n\n- Response accuracy with Capture validity\n\n- Natural output variance with product update\n\n- AI product change with external news event\n\n- Wave termination with post-result exclusion\n\n- Automatic bot measurement with human user capture\n\n- API log with user surface recording\n\n- Provider telemetry with independent capture\n\n- Evidence access to everyone by storing the evidence\n\n- Irreversible record with immutable record\n\n- Versioned correction of wrong decision with raw data protection\n\nAt the end of this section, each GEO-1000 audit should be able to answer the following questions:\n\n> To which wave does this observation belong?\n\n> At what UTC time was the prompt sent?\n\n> What was the initial system result?\n\n> When was the response completed and when was it captured?\n\n> If a technical retry was made, what was the reason and where was the previous attempt?\n\n> Does the screenshot prompt show the product and the complete response?\n\n> Do the raw response and the visual confirm each other?\n\n> When were the hash and timestamp of the files generated?\n\n> If the evidence was modified later, is the change history visible?\n\n> Can an independent auditor verify the raw evidence while preserving participant privacy?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEach GEO-1000 observation must be linked to a locked measurement wave before collecting data; all material timings from sending the prompt to the first system result, from completing the response to capture and upload; AI product conditions; the full prompt sent; the first suitable output and all integrity records must be preserved in a versioned NOMOS Capture Evidence Package. A GEO-1000 observation alone:\n\n- screenshot,\n\n- copied response text,\n\n- participant statement,\n\n- file hash\n\ncannot be considered valid on its own. The main evidence package, to the extent relevant, must include:\n\n- Wave ID\n\n- Participant's anonymous observation ID\n\n- AI System Register record\n\n- Panel status\n\n- Prompt ID and version\n\n- Assigned prompt\n\n- Sent prompt\n\n- Submission time\n\n- Initial system contact\n\n- Response completion time\n\n- Capture time\n\n- Upload time\n\n- Full raw response\n\n- Full screen or sequential screenshots\n\n- Reference and source records\n\n- Technical error and retry log\n\n- Product settings\n\n- File hashes\n\n- Package hash\n\n- Time verification record\n\n- Capture tool version\n\n- Validity decision\n\n- Privacy and redaction status\n\n- Chain of custody owner\n\n## 3. WHAT IS A MEASUREMENT WAVE?\n\nA measurement wave is a comprehensive data collection event conducted under predefined:\n\n- AI products,\n\n- target population,\n\n- prompt versions,\n\n- user panels,\n\n- time window,\n\n- sample allocation,\n\n- capture protocol,\n\n- reference records\n\nA wave:\n\nWk=(A,U,P,F,T,C,R,G)\n\ncan be represented as. Here:\n\n- A: AI product examples\n\n- U: target user population and sample\n\n- P: prompt set\n\n- F: panel states\n\n- T: time and submission structure\n\n- C: capture protocol\n\n- R: reference reality version\n\n- G: governance and key record\n\nIf any of these fields change materially, a new wave or sub-wave may be required.\n\n## 4. TYPES OF WAVES\n\n### MW-1 — PRINCIPAL SYNCHRONISED WAVE\n\nIt is the main GEO-1000 population measurement. By default, it aims for at least 1,000 valid Population Panel observations for each AI product.\n\n### MW-2 — CONFIRMATORY WAVE\n\nIt is conducted to repeat the result of the first wave, to confirm a specific error, or to test statistical stability. It is not used to change the low or high score of the first wave. It produces a separate result.\n\n### MW-3 — INCIDENT WAVE\n\nIt is conducted after a Critical or significant Observer event for targeted and rapid verification. It is not a global main wave.\n\n### MW-4 — DRIFT MONITORING WAVE\n\nMonitors time changes in product, entity, resource, or language behaviour.\n\n### MW-5 — CORRECTION RETEST WAVE\n\nProduces new results after a correction or intervention. It does not erase the previous wave.\n\n### MW-6 — LOCAL-TIME BALANCED WAVE\n\nIt aims to measure participants under similar local time conditions. It does not measure a single global system moment.\n\n### MW-7 — CONTROLLED LABORATORY WAVE\n\nFrozen model is a controlled study conducted with API or standard test accounts. It is not a real user Population Panel.\n\n## 5. WHAT DOES “SAME TIME” MEAN?\n\nIn GEO-1000, simultaneity:\n\n#### Does not mean that all users click at the same millisecond\n\nThe functional definition is:\n\n> Observations are generated within a predetermined UTC measurement window, according to recorded and outcome-independent submission slots.\n\nThis approach balances three objectives:\n\n- Keeping the product status close in terms of time\n\n- Reducing sudden artificial load on the provider\n\n- Distributing participant submissions in a reproducible manner\n\n## 6. TIME STRUCTURE\n\nEach wave must carry at least four time layers.\n\n### 6.1. Wave Window\n\nIt is the range UTC in which all main submissions must be made. Example:\n\n### 12:00–13:00 UTC\n\n### 6.2. Micro-Slot\n\nIt is the shorter time interval assigned for the participant to send the prompt. Example:\n\n### 12:15–12:20 UTC\n\n### 6.3. Completion Additional Time\n\nIf the prompt is sent within the slot, the predefined time allowed for completing the response is.\n\n### 6.4. Capture and Upload Time\n\nThe time allowed for creating the evidence and uploading it to a secure system after the response is completed.\n\n## 7. CANDIDATE SYNCHRONISED WAVE BASIC RULE\n\n### CANDIDATE BASE RULE — CANDIDATE FOUNDATIONAL RULE\n\nFor general consumer chat products, the following structure can be used as the main synchronised wave candidate:\n\n- Submission window: 60 minutes\n\n- Micro-slots: 12 × 5 minutes\n\n- Participant submission: within the micro-slot assigned to them\n\n- Additional time for response completion: up to 20 minutes after submission\n\n- Capture hash generation: preferably within 5 minutes after the answer is completed\n\n- Evidence upload: at most within 60 minutes after the capture\n\n- Reporting technical issues: within the same wave\n\nThese durations are universal and not final. They can change according to the following conditions:\n\n- AI product response time\n\n- Mobile connection\n\n- Country infrastructure\n\n- Long answer format\n\n- Accessibility requirement\n\n- Product usage limit\n\nThe times used should be locked before collecting data.\n\n## 8. WHY MICRO-SLOTS?\n\nSimultaneous submission by a thousand participants:\n\n- Unusual load in the AI product,\n\n- speed limit,\n\n- queue,\n\n- technical error,\n\n- increase in response time\n\nmay be created. In this case, the audit cannot measure the normal behaviour of the product but:\n\n#### can measure the artificial load event created by the audit\n\nMicro-slots:\n\n- distribute the load,\n\n- maintain the same overall system time,\n\n- make the submission order independent of the result,\n\nallow modelling of the time effect.\n\n## 9. SLOT ASSIGNMENT\n\nParticipants should be assigned to slots:\n\n- country,\n\n- language,\n\n- AI product,\n\n- plan,\n\n- in a balanced and pre-randomised manner in terms of:\n\npanel status. A micro-slot alone should not be filled with:\n\n- specific country,\n\n- specific language,\n\n- specific product\n\nusers. Otherwise, the time effect will be confounded with the group effect.\n\n## 10. LOCAL TIME FAIRNESS\n\nA single UTC window may be for some participants:\n\n- night,\n\n- working hours,\n\n- prayer or rest time,\n\n- hard-to-reach hours\n\nThis situation:\n\n- participation rate,\n\n- connection condition,\n\n- user attention,\n\n- device type\n\nmay be affected. Two complementary methods can be used.\n\n### 10.1. Rotating Global Waves\n\nMain waves are conducted at different UTC times. Example:\n\n- Wave 1: 12:00 UTC\n\n- Wave 2: 20:00 UTC\n\n- Wave 3: 04:00 UTC\n\nThus, the same region does not bear the same local time load in each wave.\n\n### 10.2. Local Time Balanced Sub-Wave\n\nParticipants are measured in a similar time band in their local time. This study:\n\n- user fatigue,\n\n- is valuable for\n\ndaily usage behaviour.\n\nIt is not the result of a single global system moment.\n\n## 11. THREE-WAVE STABILITY ARCHITECTURE\n\n### CANDIDATE ARCHITECTURE — CANDIDATE ARCHITECTURE\n\nAt least three main waves can be suggested for the full implementation of GEO-1000:\n\n#### W1 — Initial Wave\n\nGenerates the initial population estimate.\n\n#### W2 — Short-Term Repeat\n\nTests the short-term stability of the product and measurement. Candidate interval: 24–72 hours after W1\n\n#### W3 — Medium-Term Stability\n\nMeasures whether the representation of the product and entity is preserved over the long term. Candidate interval: 14–30 days after W1. Definite intervals must be determined before collecting data. If it will be changed due to a product update or event, a new version record is required.\n\n## 12. 30,000 OBSERVED FOUNDER DESIGN\n\nFor AI products, 1,000 main observations per product and for three main waves:\n\n10×1,000×3=30,000\n\nvalid main Population Panel observations are targeted. Complementary:\n\n- Country Observer\n\n- Language Fairness\n\n- Accessibility\n\n- Incident\n\n- Retest\n\npanels can be added on top of this number. 30,000:\n\n- is not the number of invitations,\n\n- is not the number of files,\n\nis not the number of screenshots. It is the valid main product-observation target.\n\n## 13. TIME VECTOR OF THE OBSERVATION\n\nFor each observation, the time vector can be defined as a candidate as follows:\n\nT_i = (t_i^assign, t_i^open, t_i^submit, t_i^first, t_i^complete, t_i^capture, t_i^upload, t_i^validate)\n\nHere:\n\n- tiassign: the time the task is assigned to the user\n\n- tiopen: the time the task is opened\n\n- tisubmit: the time the prompt is submitted\n\n- tifirst: the time the first visible system result begins\n\n- ticomplete: the time when the response is completed\n\n- ticapture: the time when the evidence is captured\n\n- tiupload: the time when the package reaches the secure system\n\n- tivalidate: the time when the validity of the capture is examined\n\nAll of these times may not be directly observable on every user surface. Unknown fields should remain UNKNOWN.\n\n## 14. BASIC TIME METRICS\n\nResponse delay:\n\nL_i = t_i^complete − t_i^submit\n\nCapture delay:\n\nC_i = t_i^capture − t_i^complete\n\nUpload delay:\n\nU_i = t_i^upload − t_i^capture\n\nSlot deviation:\n\nD_i = t_i^submit − t_i^{slot-midpoint}\n\ncan be calculated as follows. These metrics are:\n\n- system load,\n\n- participant delay,\n\n- proof confidence,\n\n- wave integrity\n\nin terms of evaluation.\n\n## 15. CANNOT BE MANAGED ACCORDING TO DELAY\n\nAn answer completed over a long period:\n\n- can be forced into the categories of\n\ntrue,\n\n- false,\n\n- long,\n\n- may be cited.\n\nDue to delay, the exclusion rule:\n\n- before data collection,\n\n- regardless of the content of the answer\n\nmust be defined. The following behaviour is prohibited: excluding a negative response because it was completed late; protecting a positive response when it is completed with the same delay.\n\n## 16. WAVE STATUS STATUSES\n\n### WS-0 — PLANNED\n\nThe wave is planned. It has not been locked yet.\n\n### WS-1 — LOCKED\n\nSample, prompt, product, time, and capture rules are locked.\n\n### WS-2 — OPEN\n\nThe submission window has started.\n\n### WS-3 — CLOSED\n\nAcceptance of new submissions has ended. Completion and uploading extension times may continue.\n\n### WS-4 — COMPLETED\n\nField and initial capture verifications have been completed.\n\n### WS-5 — SPLIT\n\nIt has been split into sub-waves due to material system or protocol changes.\n\n### WS-6 — PAUSED\n\nTemporarily halted due to security, product, or infrastructure issues.\n\n### WS-7 — ABORTED\n\nTerminated before the wave was completed. Results may be limited or unusable.\n\n### WS-8 — INVALIDATED\n\nInvalid for the main forecast due to a previously defined severe protocol defect. Raw observations are not deleted.\n\n### WS-9 — ARCHIVED\n\nVersion has been closed and the evidence chain archived.\n\n## 17. WAVE HEALTH CLASSES\n\n### WH-0 — NOT ASSESSED\n\nWave health has not been assessed.\n\n### WH-1 — COMPROMISED\n\nThere is a material system change, time, or capture fault. May not be usable for the main result.\n\n### WH-2 — PARTIAL\n\nSome layers or slots are insufficient. A reduced result can be produced.\n\n### WH-3 — ACCEPTABLE\n\nMain time, system, and capture conditions are adequately preserved.\n\n### WH-4 — STRONG\n\nHigh capture confidence, balanced slot coverage, and low protocol loss are present.\n\n### WH-5 — REPLICATED\n\nSimilar wave health has been repeated over multiple times and independent panels. Wave health does not depend on whether the AI score is high or low.\n\n## 18. PRODUCT CHANGES WITHIN THE WAVE\n\nThe AI product can materially change within the measurement window. Signals:\n\n- Model label change\n\n- Interface change\n\n- Activation or deactivation of web feature\n\n- Change in citation format\n\n- Sudden break in response structure\n\n- Provider announcement\n\n- Different system behaviour simultaneously in many users\n\n- Product interruption or restart\n\nIn this case: The time of change is estimated. System configuration fingerprints are compared. Wave WS-5 can be put into SPLIT status. Previous and subsequent observations will be separate sub-waves. If a single score is to be produced, the system status is modelled as a separate factor. Material uncertainty is documented in the public record.\n\n## 19. EXTERNAL EVENT RECORD\n\nThe world can change even if the AI product does not change. During measurement:\n\n- company acquisition,\n\n- product launch,\n\n- major news,\n\n- legal decision,\n\n- crisis,\n\n- financial disclosure,\n\n- viral content\n\nmay occur. This event:\n\n- web resources,\n\n- retrieval results,\n\n- user prompt comment,\n\n- AI response\n\ncan be changed. Each wave, to the extent it is relevant:\n\n#### External Event Register\n\nmust carry.\n\n## 20. HOW ARE EXTERNAL EVENT RESULTS MANAGED?\n\nAn external event:\n\n- if known before the wave, the reference package is updated,\n\n- if it occurs during the wave, it is marked with a timestamp,\n\nIf noticed after the wave, a retrospective event record is added. The following behaviour is prohibited: excluding observations due to news that lowers the score; preserving them when it is news that raises the score. The event itself may be part of the real-world system state. If necessary, sub-wave or event-adjusted analysis is performed.\n\n## 21. FIRST CONTACT OUTCOME\n\nThe first system result the user encounters after sending the prompt:\n\n#### First Contact Outcome\n\nis called. The first contact outcome:\n\n- Full response\n\n- Rejection\n\n- Technical error\n\n- Usage limit\n\n- Connection interruption\n\n- Endless loading\n\n- Clarification question\n\n- No result\n\npossible. This event is part of the real user experience. It cannot be silently deleted.\n\n## 22. FIRST ELIGIBLE CONTENT OUTPUT\n\nThe first material response generated in the previously permitted retry after a technical error:\n\n#### Can be called First Eligible Content Output\n\nThis output: does not replace the initial contact result, it is recorded along with it. Example:\n\n- Initial contact: speed limit\n\n- Five minutes later, previously permitted technical retry\n\n- First content response\n\nThe report should show the following together:\n\n- First user experience: speed limit\n\n- First content representation: specified response\n\n## 23. FIRST SUITABLE OUTPUT RULE\n\nIf an AI product produces a substantial response to the first prompt:\n\n- even if it is wrong,\n\n- even if it is short,\n\n- even if it includes a refusal,\n\n- even if it does not provide references,\n\n- even if it incorrectly identifies the brand\n\na reproduction cannot be made. The main observation is the first output. The user:\n\n- cannot select a new response by:\n\n- regenerate,\n\n- rewrite,\n\n- try again,\n\n- another answer,\n\ncontinue\n\n## 24. TECHNICAL RETRY\n\nTechnical retry can only be used under previously defined technical conditions. Candidate permissible reasons:\n\n- Connection interruption before response starts\n\n- Server error code\n\n- Interface not sending the prompt\n\n- Application crash\n\n- Verified temporary system outage\n\nCannot be a reason for technical retry:\n\n- Incorrect answer\n\n- Rejection\n\n- Short answer\n\n- Lack of citation\n\n- Brand not mentioned\n\n- Critical error\n\n- Audited institution does not like the answer\n\n## 25. RETRY IN PARTIAL ANSWER\n\nIf the system has produced part of the material content and then stopped:\n\n- partial output is stored,\n\n- saved as TRUNCATED_RESPONSE,\n\nif a retry is to be made, a new and linked attempt is created. It cannot be deleted as if there were no partial response.\n\n## 26. ATTEMPT CHAIN\n\nAn observation assignment may carry multiple technical attempts:\n\nA_i → (Attempt_{i,1}, Attempt_{i,2}, …)\n\nEach attempt:\n\n- time,\n\n- carries the product state,\n\n- prompt hash,\n\n- system result,\n\n- technical error reason\n\nIt is not overwritten by another attempt.\n\n## 27. WHAT IS NOMOS CAPTURE?\n\nNOMOS Capture is an open data and procedure standard that preserves the integrity of observation from the prompt given to the user to the archiving of the evidence package. NOMOS Capture:\n\n- is not just a screenshot application,\n\n- does not have to be proprietary software belonging only to NobleJackal,\n\nis not dependent only on a single browser extension. A compatible NOMOS Capture application:\n\n- uses an open data schema,\n\n- follows time and integrity rules,\n\n- respects privacy boundaries,\n\n- preserves the chain of evidence\n\nmust be met. Independent researchers should be able to develop their own compatible tools. If the standard can only be implemented with the founder's software, independent replication weakens.\n\n## 28. FOUR FUNCTIONS OF NOMOS CAPTURE\n\n### 28.1. Deliver the Task\n\nCorrect:\n\n- user,\n\n- prompt,\n\n- AI product,\n\n- slot\n\nensures matching.\n\n### 28.2. Capture the Observation\n\nLogs the prompt, the initial system result, the response, and the related user surface.\n\n### 28.3. Maintain Integrity\n\nGenerates the file hash, timestamp, and packet relationship.\n\n### 28.4. Transport the Evidence\n\nLinks raw, limited, and publicly available layers of evidence together.\n\n## 29. CAPTURE METHODS\n\n### CM-1 — GUIDED MANUAL CAPTURE\n\nThe participant provides a screenshot and raw response text following instructions. It carries a low technical burden. It is prone to human error.\n\n### CM-2 — ASSISTED CAPTURE APPLICATION\n\nApplication:\n\n- copying prompts,\n\n- time logging,\n\n- file upload,\n\n- hash generation\n\nfunctions are supported. It does not interfere with the AI product.\n\n### CM-3 — INSTRUMENTED USER-SURFACE CAPTURE\n\nBrowser or application integration:\n\n- the sent prompt,\n\n- the product metadata,\n\n- the response text,\n\n- the times\n\nis captured directly. Provides high integrity. Terms of use and privacy should be evaluated.\n\n### CM-4 — PROVIDER EXPORT OR TELEMETRY CAPTURE\n\nThe provider's session export or verified telemetry is used. Can strongly support independent capture. May not provide independence on its own.\n\n### CM-5 — CONTROLLED API OR LAB CAPTURE\n\nAPI or frozen system logs are used. It is not live user surface observation. Belongs to a separate laboratory panel.\n\n## 30. LIMIT OF AUTOMATION\n\nAutomation:\n\n- prompt submission,\n\n- time logging,\n\n- hash generation,\n\n- file integrity\n\ncan be used for. Instead of the participant:\n\n- bot account,\n\n- automated bulk query,\n\n- non-user session\n\ngeneration does not create a real human Population Panel. Automated system auditing is separate:\n\n### CONTROLLED_AUTOMATION_PANEL\n\nmust be recorded as.\n\n## 31. CAPTURE CONFIDENCE LEVELS\n\n### NCL-0 — NO VERIFIABLE CAPTURE\n\nThere is only the participant's statement. Cannot be included as the main GEO-1000 result.\n\n### NCL-1 — PARTIAL EVIDENCE\n\nThere is a response text or a limited screenshot. the prompt or system condition cannot be fully verified.\n\n### NCL-2 — SCREEN-EVIDENCED\n\nThe prompt, response, and main product surface have been verified with a screenshot. Metadata and time confidence may be limited.\n\n### NCL-3 — FULL EVIDENCE BUNDLE\n\nComplete response, prompt, product record, times, screenshot, and hash are available. Minimum level for main analysis is a candidate.\n\n### NCL-4 — INSTRUMENTED AND SIGNED CAPTURE\n\nThe capture tool has recorded the prompt, response, time, and metadata directly; the package is digitally signed.\n\n### NCL-5 — INDEPENDENTLY VERIFIED MULTI-SOURCE CAPTURE\n\nCapture package:\n\n- independent validation,\n\n- provider export,\n\n- second source of evidence\n\nhas been supported by. Main GEO-1000 real user observations at least:\n\n### NCL-3\n\nshould target that level.\n\n## 32. NOMOS CAPTURE EVIDENCE PACKAGE\n\nFor each observation, the package carries the following components as candidates:\n\n### 32.1. Observation Manifest\n\nIncludes all identity and status fields of the observation.\n\n### 32.2. Prompt Artefact\n\nThe full prompt text assigned and sent, the version, and its hash.\n\n### 32.3. Raw Response Artefact\n\nThe full raw answer produced by the AI product.\n\n### 32.4. Visual Evidence Artefact\n\nShows images or sequential images of the prompt, response, and product surface.\n\n### 32.5. System-State Artefact\n\nShows the plan, surface, model label, memory, web, and other material system states.\n\n### 32.6. Time Evidence Artefact\n\nCarries the submission, completion, capture, and upload times.\n\n### 32.7. Attempt Log\n\nShows the chain of technical errors and retries.\n\n### 32.8. Integrity Manifest\n\nContains hashes and sizes of all files.\n\n### 32.9. Privacy Manifest\n\nShows redaction, access and retention status.\n\n### 32.10. Validation Record\n\nIndicates who and when the capture validity was assessed.\n\n## 33. MINIMUM CONTENT OF THE SCREENSHOT\n\nTo the extent relevant, visual evidence should show:\n\n- AI product or surface identity\n\n- Complete prompt\n\n- Start of the response\n\n- End of the response\n\n- Reference fields\n\n- Rejection or error message\n\n- Apparent model label\n\n- Evidence that the session is new or in a relevant state\n\n- Material system setting\n\n- Interface status indicating that the answer has not been cut off\n\n## 34. LONG ANSWERS\n\nIf the answer does not fit on a single screen:\n\n- full page capture,\n\n- scrollable capture,\n\n- overlapping consecutive screenshots,\n\n- screen recording,\n\n- direct text export\n\ncan be used. Consecutive visuals:\n\n- should not leave gaps,\n\n- should overlap with each other,\n\nshould carry sequence numbers. Only the positive or relevant section cannot be cropped.\n\n## 35. SCREEN RECORDING\n\nScreen recording:\n\n- prompt submission,\n\n- streaming response,\n\n- completion status,\n\n- no refresh performed\n\ncan be shown. However:\n\n- more personal information,\n\n- notification,\n\n- private application,\n\n- account ID\n\ncan be captured. Therefore, screen recording:\n\n- may not be a default requirement,\n\n- can be used in high-risk or verification sub-sample,\n\nrequires privacy preparation before recording.\n\n## 36. RAW RESPONSE TEXT\n\nIn addition to the screenshot, the full answer text should be stored separately. Preference order:\n\n- Official export or copy feature of the product\n\n- Direct text capture by the capture tool\n\n- Accessibility tree or DOM-based extraction\n\n- Text copied by the user\n\n- OCR and human verification\n\nOCR should only be the last option. If there is a material difference between automatic extraction and the visual, a review is required.\n\n## 37. DYNAMIC ANSWERS\n\nSome products, after the answer is completed:\n\n- can add a citation,\n\n- can load a source card,\n\n- can reformat the text,\n\ncan add a security warning. Two capture moments can be used:\n\n- First completion capture\n\n- Capture after brief render stability\n\nIf there is a material difference, both versions are retained. The first user exposure is not deleted.\n\n## 38. HOW TO UNDERSTAND THAT THE ANSWER IS COMPLETED?\n\nCandidate completion indicators:\n\n- Streaming stops\n\n- Loading indicator closes\n\n- Regenerate or equivalent button appears\n\n- Product completion signal\n\n- Predefined silence period\n\n- API completion record\n\nProduct-based completion rules must be found in the AI System Register.\n\n## 39. IF THE USER STOPPED THE RESPONSE\n\nThe user may have accidentally or intentionally:\n\n- stop,\n\n- cancel,\n\n- terminate the application\n\nperformed the action. This situation is recorded as:\n\n### USER_INTERRUPTED\n\n### ACCIDENTAL_INTERRUPTION\n\n### UNKNOWN_INTERRUPTION\n\nA response stopped by the user cannot be considered a complete system output. The participant replacement or retry rule must be defined in advance.\n\n## 40. THE PROMPT AND RESPONSE BEING IN SEPARATE FILES\n\nIf the prompt and response are in separate images:\n\n- they must be linked to the same session,\n\n- to the same observation ID,\n\n- and to an uninterrupted time chain\n\nOtherwise, it may not be possible to verify that the prompt and response truly belong to the same conversation.\n\n## 41. FILE INTEGRITY\n\nA cryptographic hash must be generated for each file. [K25] Candidate:\n\nh_j = H(file_j)\n\nPackage hash:\n\nh_bundle = H(JCS(manifest(file_id, order, size, algorithm, h_j))) [K14; K25]\n\ncan be created. Here, H is a published and considered secure hash algorithm. The algorithm:\n\n- versioned,\n\n- open,\n\n- can be switched in the future if necessary.\n\nmust be.\n\n## 42. WHAT DOES A HASH PROVE?\n\nIf the recorded hash value is stored reliably, re-hashing the same file with the same algorithm helps detect changes in the byte sequence [K25]. This comparison supports the claim that the file has preserved the same byte sequence after the reference hash was generated; it does not prove the source of the file, that it was produced at the correct time, or its authenticity before hashing. A hash alone does not prove:\n\n- That the file came from the real AI product\n\n- That it belongs to the correct user\n\n- That it was produced at the correct time\n\n- Not organised before hashing\n\n- That it is the first output\n\n- That the prompt was correct\n\nTherefore, hash:\n\n> It is proof of integrity.\n\nAlone:\n\n> It is not proof of originality.\n\n## 43. DIGITAL SIGNATURE\n\nCapture tool or authorised verifier:\n\n- the observation package,\n\n- time record,\n\n- vehicle version\n\ncan sign digitally. Signature:\n\n- which vehicle or institution produced the package,\n\n- that it has not changed after the signature\n\nstrengthens. The signature holder may have signed incorrect or forged data. Therefore, the signature alone does not guarantee authenticity.\n\n## 44. RELIABLE TIME [K26]\n\nLocal device clock:\n\n- misconfigured,\n\n- manually changed,\n\n- out of synchronisation\n\nmay occur. A strong timestamp should carry multiple signals:\n\n- Capture server acceptance time\n\n- Reliable UTC time service\n\n- Device time and deviation record\n\n- Visible time in the AI product, if any\n\n- File creation time\n\n- Digital timestamp\n\n## 45. TIME DEVIATION\n\nThe difference between the participant's device time and the reliable UTC time:\n\nO_i^clock = t_i^device − t_i^reference\n\ncan be recorded as. If there is a material deviation:\n\n- the server time is used as the main reference,\n\n- the device time is preserved as a corrected record,\n\nuncertainty is marked. The participant should not be asked to manually change the device clock.\n\n## 46. UPLOAD TIME IS NOT THE CAPTURE TIME\n\nUploading a file to the server at 14:00 does not prove that the screenshot was taken at 14:00. Therefore:\n\n- capture time,\n\n- hash generation time,\n\n- upload time\n\nshould be recorded separately. Long upload delays may reduce capture trust. The record does not have to be automatically invalid.\n\n## 47. DELAYED UPLOAD\n\nDue to connection issues, the user may not be able to upload the evidence immediately. If late upload is accepted:\n\n- local hash at capture time,\n\n- trusted device time,\n\n- offline signed record,\n\n- reason for upload delay\n\nmay be required. Only the user's statement: “I took this image yesterday.” is not sufficient for the main NCL-3 level.\n\n## 48. EVIDENCE CHAIN\n\nThe evidence chain of an observation as a candidate consists of the following stages:\n\n- Task assignment\n\n- Prompt lock\n\n- User and product suitability\n\n- Slot opening\n\n- Prompt submission\n\n- Initial contact result\n\n- Completion of response\n\n- Capture\n\n- Local integrity record\n\n- Secure upload\n\n- Server acceptance time\n\n- Package hash\n\n- Protocol verification\n\n- Editing\n\n- Adjudication\n\n- Archive\n\n- Public manifest\n\nEach stage must be connected to the previous stage.\n\n## 49. BREAKING THE CHAIN\n\nThe chain of evidence can be broken in the following situations:\n\n- No prompt record\n\n- The initial output cannot be verified\n\n- The response does not match the visual\n\n- Capture time is unknown\n\n- File has been modified later\n\n- Duplicate participant ID\n\n- Product and plan are unknown\n\n- Observation is out of wave\n\n- Hash does not match the file\n\n- Editing overwrote the raw file\n\n- It is unknown who verified it\n\nThe type of break determines the validity status.\n\n## 50. RAW EVIDENCE CANNOT BE CHANGED; THE DECISION CAN CHANGE\n\nRaw evidence package:\n\n- should not be written over,\n\n- should not be quietly corrected,\n\nshould not be deleted and replaced with a new one. However, regarding the observation:\n\n- validity,\n\n- adjudication,\n\n- classification,\n\n- compliance\n\nthe decision can change. The new decision:\n\n- new version,\n\n- justification,\n\n- date,\n\n- decision-maker\n\nshould carry. This distinction is a fundamental governance rule:\n\n> Evidence can remain fixed; interpretation can evolve with the evidence.\n\n## 51. REDACTION\n\nThe following fields can be redacted in the public or adjudicator copy:\n\n- Name and surname\n\n- Email\n\n- Profile picture\n\n- Account ID\n\n- Private conversations\n\n- Notifications\n\n- Corporate field\n\n- Sensitive file name\n\n- Exact location\n\nRedaction:\n\n- should not be done on the raw file,\n\n- should produce a new derivative file,\n\n- should carry its own hash,\n\nshould record which field was redacted and why.\n\n## 52. THREE LAYERS OF EVIDENCE\n\n### 52.1. Raw Restricted Evidence\n\nIt is full raw evidence. Only authorised auditors and required research roles can access it.\n\n### 52.2. Audit Evidence Layer\n\nContains the evidence necessary for independent adjudicators and reproduction teams. Personal areas are minimised.\n\n### 52.3. Public Evidence Manifest\n\nFor the public:\n\n- observer ID,\n\n- country and language cell,\n\n- AI product,\n\n- time,\n\n- evidence hash,\n\n- validity status,\n\n- redaction status\n\nare presented. Raw private content is not necessarily published.\n\n## 53. PUBLIC EVIDENCE MANIFEST\n\nFor each wave, the public manifest may include the following fields:\n\n- Wave ID\n\n- Protocol version\n\n- Number of products\n\n- Target and valid observation count\n\n- Country and language coverage\n\n- Slot coverage\n\n- Capture confidence levels\n\n- Technical error and rejection rates\n\n- Invalid observation count\n\n- Observation package hashes\n\n- Wave integrity root\n\n- Change events\n\n- Publicly redacted examples\n\n- Independent access procedure\n\n- Evidence retention status\n\n## 54. WAVE INTEGRITY ROOT\n\n### CANDIDATE CONCEPT\n\nLet the hash of each observation package be hi. From these, the wave root with a Merkle tree or equivalent structure:\n\n### RW\n\ncan be produced. The wave root can be published after the measurement is closed. This method:\n\n- the quiet later modification of the observation list in the wave,\n\n- adding a new observation,\n\n- modifying an existing observation\n\nmakes it more visible. Merkle root:\n\n- does not alone prove that the observations are real,\n\n- or that the adjudication decision is correct\n\nIt strengthens the integrity of the list.\n\n## 55. ADDING AND REMOVING OBSERVATIONS\n\nAfter the wave is closed:\n\n- adding a new observation,\n\n- removing an existing observation,\n\n- modifying the file\n\nrequires a new manifest version. The old integrity root is preserved. For the observation being removed:\n\n- reason,\n\n- date,\n\n- decision maker\n\n- effect on old and new score\n\nis recorded.\n\n## 56. PARTICIPANT WITHDRAWAL REQUEST\n\nResearch ethics or the data protection framework may allow the participant to withdraw their data. In this case:\n\n- raw evidence can be safely deleted or made inaccessible,\n\n- an empty WITHDRAWN record can be preserved in the public manifest,\n\n- wave score can be recalculated with the new version,\n\nthe old result and reason for change are preserved. Immutability does not mean eliminating human rights.\n\n## 57. OBSERVATION CAPTURE VALIDITY STATUSES\n\n### CV-0 — NOT REVIEWED\n\nThe capture has not been evaluated yet.\n\n### CV-1 — VALID\n\nMeets all mandatory capture and time conditions.\n\n### CV-2 — CONDITIONALLY VALID\n\nThere is a limited deficiency. It is specified in which analyses it can be used.\n\n### CV-3 — PARTIAL EVIDENCE\n\nThe observation can be used as an event. It may not enter the main population score.\n\n### CV-4 — OUTSIDE WAVE\n\nThe prompt was sent outside the allowed time structure.\n\n### CV-5 — PROMPT OR OUTPUT MISMATCH\n\nThere is a material discrepancy between the assigned prompt, the submitted prompt, or the recorded response.\n\n### CV-6 — DUPLICATE\n\nThe same observation or user record has been repeated.\n\n### CV-7 — TAMPER SUSPECTED\n\nThere is an unresolved doubt about the integrity of the file or metadata.\n\n### CV-8 — FABRICATED OR MANIPULATED\n\nThe observation is not a real user experience or has been deliberately altered.\n\n### CV-9 — WITHDRAWN\n\nIt has been excluded from analysis due to participant or ethical processes.\n\n## 58. CAPTURE VALIDITY SHOULD BE SEPARATED FROM SEMANTIC ACCURACY\n\nAn observation:\n\n- completely valid in terms of capture,\n\n- severely incorrect in terms of content\n\nmay be. Another observation:\n\n- correct answer,\n\n- may carry missing or manipulated evidence\n\nTherefore, there are two separate decisions:\n\n- Capture Validity\n\n- Semantic Adjudication\n\nThe capture validator should know the success score of the response as little as possible.\n\n## 59. ROLE OF THE CAPTURE VALIDATOR\n\nThe capture validator examines:\n\n- Prompt matching\n\n- Wave and slot\n\n- Product identity\n\n- First output rule\n\n- Screen integrity\n\n- Raw text consistency\n\n- Technical retry\n\n- Hash and time\n\n- Privacy\n\nDoes not decide on: Was the company described correctly? Does the citation support the claim? Is there a critical error? Does the response pass? These are the responsibilities of the next review layer.\n\n## 60. REPEATED SCREEN IMAGE\n\nThe same visual:\n\n- accidentally uploaded twice,\n\n- used by multiple users,\n\n- repeated among panel providers\n\nmay occur. Checks:\n\n- File hash\n\n- Visual similarity\n\n- Raw response text\n\n- Time and session metadata\n\n- User uniqueness\n\ncan be done. The same AI product may give the exact same answer word for word to some users. The same text alone is not evidence of cheating. The same file or the same visual trace is a stronger signal.\n\n## 61. FILE SIMILARITY AND FALSE POSITIVE\n\nSame device and interface:\n\n- similar visuals,\n\n- same screen size,\n\n- same compression\n\ncan produce. The visual similarity algorithm should not perform automatic exclusion. Multiple signals and human review are required.\n\n## 62. PROVIDER TELEMETRY\n\nAI provider:\n\n- session time,\n\n- model,\n\n- prompt,\n\n- response,\n\n- tool usage\n\ncan provide reliable logs about. This record strengthens independent capture. However, the provider:\n\n- measured party,\n\n- system owner,\n\n- beneficiary\n\nmay be. Provider telemetry: should be used together with, not as a replacement for, independent user evidence.\n\n## 63. IF THERE IS NO PROVIDER EVIDENCE\n\nAccess to provider logs may not be available in most live consumer products. This does not make GEO-1000 impossible. The following composition can be used:\n\n- Instrumented user capture\n\n- Server timestamp\n\n- Full screen proof\n\n- Raw response\n\n- System setting\n\n- Hash and signature\n\n- Independent verification sub-sample\n\nProof trust level is indicated accordingly.\n\n## 64. CAPTURE IN NATURAL USER PANEL\n\nCapture tool in the natural panel:\n\n- special instruction content,\n\n- previous conversations,\n\n- account identity\n\nshould not be captured unnecessarily. Preferred flow: The user opens a new natural account chat. Capture only records the task window. Personalisation status is taken separately as high-level metadata. Private history stays outside capture. Public copy is edited.\n\n## 65. NATURAL CONVERSATION-STATE CAPTURE\n\nIf the current conversation context is measured:\n\n- only the necessary previous turn range,\n\n- explicit consent,\n\n- strong redaction,\n\n- access restriction\n\nis required. All private chat history cannot be collected by default.\n\n## 66. MOBILE CAPTURE\n\nOn mobile devices:\n\n- full screen capture,\n\n- scrollable capture,\n\n- application version,\n\n- system time log\n\nmay work differently. Mobile protocol:\n\n- iOS,\n\n- Android,\n\n- other systems\n\nmay carry separate technical guides. Mobile users cannot be excluded as desktop capture is difficult. The capture tool must adapt to the population.\n\n## 67. ACCESSIBILITY AND CAPTURE\n\nFor participants using screen readers, voice input, or motor assistive tools, standard visual capture alone may not be sufficient. Alternative evidence:\n\n- Accessibility tree export\n\n- Voice session recording\n\n- Application transcript\n\n- Assistive technology log\n\n- Task observation record\n\nmay occur. These observations should not be considered low reliability. An equivalent evidence standard should be developed.\n\n## 68. OBSERVATION OF THE CAPTURE TOOL CHANGING BEHAVIOUR\n\nBrowser extension or instrumentation:\n\n- page load,\n\n- interface behaviour,\n\n- security system,\n\n- product detection\n\nmay be affected. The effect of the capture tool should be tested in the pilot. If sessions with and without the tool behave differently: the tool condition becomes an experimental factor, the capture tool cannot be assumed silently neutral.\n\n## 69. CAPTURE TOOL VERSIONING\n\nEach capture package:\n\n- tool name,\n\n- version,\n\n- configuration,\n\n- operating environment,\n\n- known bugs\n\nmust carry. If the capture tool changes during measurement:\n\n- new tool version,\n\n- subwave,\n\n- compatibility review\n\nmay be required.\n\n## 70. CAPTURE TOOL ERROR\n\nTool:\n\n- may incorrectly copy the prompt,\n\n- may truncate the end of the response,\n\n- may record the time incorrectly,\n\n- may not generate a hash,\n\nmay compress the file. Tool error:\n\n- cannot be loaded to the AI product,\n\n- or the user\n\nSeparately:\n\n### CAPTURE_TOOL_FAILURE\n\nmust be recorded as.\n\n## 71. SYNTHETIC THREE-WAVE GEO-1000 CASE\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The following products, dates, numbers, and results are entirely fictional. Ten synthetic AI products:\n\n### A1–A10\n\nshall be measured in three main waves.\n\n### 71.1. Wave Calendar\n\nFor each product in each wave: 1,000 valid main observations are targeted. Total target:\n\n10×1,000×3=30,000\n\n### 71.2. Slot Structure\n\nEach wave:\n\n- 12 micro-slots\n\n- with participants balanced by product and country per slot\n\nThe participant's slot: randomly assigned before data collection.\n\n### 71.3. W1 Result\n\nGoal: 10,000 valid main observations Field:\n\n- 13,240 invitations\n\n- 10,870 completed tasks\n\n- 10,106 valid observations in terms of capture\n\n- 10,000 main analysis observations according to previously locked main/backup order\n\nRemaining 106 valid records:\n\n- backup,\n\n- sensitivity,\n\n- capture quality control\n\nare reserved for.\n\n### 71.4. Technical Events\n\n118 usage limit 42 connection error 31 system rejection 17 user disconnection 9 prompt mismatch 6 retry screenshot All initial contact results have been saved. Only predefined real technical errors were retried.\n\n### 71.5. Product Change During W2\n\nA7 product at 20.32 UTC:\n\n- automatic web usage,\n\n- new reference view\n\nlet it win. Observations of A7:\n\n### W2A: 20.00–20.31\n\n### W2B: 20.32–21.00\n\nare separated as. A single fixed W2 score for A7 is not published without explanation.\n\n### 71.6. External Event in W3\n\nAn important legal decision should be published on 04.22 UTC about the audited synthetic company. It should be observed that responses in web-accessible products changed after 04.25. Event: It is added to the External Event Register. Results before and after the event are shown as separate diagnostic cells. Observations are not deleted because they lower or increase the score.\n\n### 71.7. Wave Health Card\n\nAll three waves can be used. A7 and external event comments are also restricted.\n\n## 72. SYNTHETIC SINGLE OBSERVATION CAPTURE CASE\n\n### SYNTHETIC EXAMPLE\n\nParticipant:\n\n- Turkey\n\n- Turkish\n\n- Orion AI web paid plan\n\n- Controlled Clean Panel\n\n### W1\n\nSlot 12.15–12.20 UTC Assigned prompt: “Which parent company is Apple.com associated with, and what are the main activities of this company?”\n\n### 72.1. Times\n\nTask opening: 12.14.20 Prompt submission: 12.16.08 First token: 12.16.10 Answer completion: 12.16.24 Capture: 12.16.39 Hash generation: 12.16.41 Upload: 12.17.12 Server acceptance: 12.17.13\n\n### 72.2. Evidence\n\nPrompt hash matches Full response text available Two overlapping screenshots available Product, plan, and model label visible Memory and special instruction off Web usage on Citation not shown No reproduction No previous conversation File hashes and package signature available Capture level:\n\n### NCL-4\n\nValidity:\n\n### CV-1\n\nIt has not yet been decided whether the response is correct or incorrect.\n\n## 73. SYNTHETIC TECHNICAL RETRY CASE\n\nOn the first attempt:\n\n- Prompt sent correctly\n\n- Product showed \"Network error\"\n\n- Substantive response not started\n\n- Error screenshot taken\n\n- Time and hash have been recorded\n\nAccording to the predefined rule, a second attempt was made within five minutes. A response occurred on the second attempt. Correct record:\n\n- First Contact Outcome: TEMPORARY_TECHNICAL_FAILURE\n\n- First Eligible Content Output: second attempt\n\n- Both attempts are in the package\n\n- First incident is included in the technical error rate\n\n- The second response is included in the content evaluation, if the method predicts this in advance\n\nIncorrect record: The first error is deleted and it is treated as if only the second response existed.\n\n## 74. SYNTHETIC INCORRECT RETRY CASE\n\nFirst response: “Apple is a local company that only sells mobile phones.” The participant does not like the response. They reproduce it. The second response turns out to be correct. The participant only submits the second response. The system log or capture record shows the redo. Correct status:\n\n### CV-8 — MANIPULATED OBSERVATION\n\nThe observation is not saved just because the second response is correct. If the first response can additionally be recovered, it can be preserved in the incident investigation. The main participant record is invalid.\n\n## 75. WAVE CLOSURE REPORT\n\nAt the end of each wave, the following record should be created:\n\n- Wave ID\n\n- Opening and closing time\n\n- Target number of observations\n\n- Invited person\n\n- Completed task\n\n- Valid capture\n\n- Observation entering main analysis\n\n- Backup valid observation\n\n- Technical error\n\n- Rejection\n\n- Usage limit\n\n- Prompt mismatch\n\n- Capture mismatch\n\n- Late upload\n\n- Duplicate\n\n- Suspected tampering\n\n- Participant withdrawal\n\n- Product changes\n\n- External events\n\n- Slot coverage\n\n- Country and language coverage\n\n- Capture confidence levels\n\n- Wave health status\n\n- Integrity root\n\n- Responsible human approval\n\n## 76. MANDATORY NORMATIVE PROVISIONS\n\n**CH12-N01**\n\nEach GEO-1000 observation must be linked to a pre-locked measurement wave.\n\n**CH12-N02**\n\nThe term \"same time\" cannot be used without the full UTC window, micro-slot, and completion rules.\n\n**CH12-N03**\n\nLocal time cannot be used alone as wave time; the main time standard must be UTC.\n\n**CH12-N04**\n\nThe transmission window, micro-slots, completion time, and loading time must be determined before AI responses are viewed.\n\n**CH12-N05**\n\nSlot assignment must be made in a way that does not leave countries, languages, products, and panel groups time-wise unbalanced.\n\n**CH12-N06**\n\nParticipants cannot be moved to earlier or later slots based on their results.\n\n**CH12-N07**\n\nThe local time offset of the single UTC window must be documented in the public method record.\n\n**CH12-N08**\n\nReturning global waves or locally time-balanced operations must be kept as separate time results.\n\n**CH12-N09**\n\nEach observation prompt must carry submission, first contact, completion, capture, and upload times to the extent relevant.\n\n**CH12-N10**\n\nUnknown time fields cannot be presented as estimated actual times.\n\n**CH12-N11**\n\nThe device clock alone cannot be considered a reliable UTC time proof.\n\n**CH12-N12**\n\nCapture time and upload time must be recorded separately.\n\n**CH12-N13**\n\nObservations sent outside the wave cannot be included in the main wave without explanation.\n\n**CH12-N14**\n\nExclusion rules for delay or slot deviation must be independent of the response content.\n\n**CH12-N15**\n\nAs a result of the first contact, it must be kept separately from the first content output.\n\n**CH12-N16**\n\nRejection, usage limit, technical error, clarification, and inconclusiveness cannot be placed in the same category.\n\n**CH12-N17**\n\nWrong, negative, reference-less, or Critical AI response cannot be a reason for retry.\n\n**CH12-N18**\n\nTechnical retry can only be performed under predefined technical conditions.\n\n**CH12-N19**\n\nWhen a technical retry is performed, all attempts and the initial contact result must be preserved.\n\n**CH12-N20**\n\nPartial or incomplete responses cannot be ignored and replaced with a new response.\n\n**CH12-N21**\n\nThe first suitable output cannot be renewed, selected, or manually improved.\n\n**CH12-N22**\n\nEach observation must aim at least for NCL-3 or a justified equivalent capture confidence level.\n\n**CH12-N23**\n\nA screenshot alone cannot be considered a complete evidence package.\n\n**CH12-N24**\n\nThe visual evidence prompt should show the entire response and the surface of the physical product.\n\n**CH12-N25**\n\nLong responses must be preserved fully and sequentially using the capture method.\n\n**CH12-N26**\n\nIf there is a material difference between the raw response text and the visual evidence, an examination should be conducted.\n\n**CH12-N27**\n\nOCR can be used as an auxiliary method if direct text recording is not possible; without verification, it cannot be considered the final source of the text.\n\n**CH12-N28**\n\nThe initial completion and subsequent material versions of dynamically changing responses must be preserved separately.\n\n**CH12-N29**\n\nThe participant stopping the response or closing the interface must be recorded as a separate event.\n\n**CH12-N30**\n\nIf the prompt and response are in separate files, they must be linked to the same session and time chain.\n\n**CH12-N31**\n\nEach evidence file and evidence package must carry an integrity hash.\n\n**CH12-N32**\n\nThe hash alone cannot be presented as proof of the observation's reality.\n\n**CH12-N33**\n\nDigital signature and timestamp are not the entirety of observation authenticity; they are components of the evidence chain.\n\n**CH12-N34**\n\nNo redaction or correction may be written over the raw evidence file.\n\n**CH12-N35**\n\nEach edited derivative must carry its own file ID and hash.\n\n**CH12-N36**\n\nAccess to raw evidence may be restricted; the public manifest must demonstrate observation integrity without revealing identity.\n\n**CH12-N37**\n\nParticipant’s name, email, password, private conversation, or unnecessary personal information cannot be added to the public evidence.\n\n**CH12-N38**\n\nThe risk of re-identification for small countries and rare subgroup users should also be assessed.\n\n**CH12-N39**\n\nThe Natural User Panel capture should not collect special instructions and past conversation content by default.\n\n**CH12-N40**\n\nEquivalent alternative capture methods should be provided for accessibility users.\n\n**CH12-N41**\n\nIf the capture tool materially changes the user experience, this effect should be measured separately.\n\n**CH12-N42**\n\nThe capture tool and its version should be recorded in every observation.\n\n**CH12-N43**\n\nIf the capture tool materially changes during a wave, a distinction should be made for sub-wave or tool version.\n\n**CH12-N44**\n\nThe capture tool error cannot be classified as an AI product or participant failure.\n\n**CH12-N45**\n\nProvider telemetry can support independent user capture; it cannot replace it without explanation.\n\n**CH12-N46**\n\nIf a material AI product change occurs during the wave, observations cannot be merged as a single fixed product state.\n\n**CH12-N47**\n\nProduct change requires wave separation, restart, explicit modelling, or mixed-state warning.\n\n**CH12-N48**\n\nThe absence of a provider update announcement is not definitive proof that the system has not changed.\n\n**CH12-N49**\n\nMaterial external events should be added to the External Event Register with a timestamp and scope.\n\n**CH12-N50**\n\nExternal events cannot be a reason for exclusion after results because they raised or lowered the score.\n\n**CH12-N51**\n\nCapture validity should be evaluated as blind as possible, before assessing the semantic correctness of the response.\n\n**CH12-N52**\n\nA correct AI answer cannot validate a missing or manipulated capture.\n\n**CH12-N53**\n\nAn incorrect AI answer cannot invalidate a protocol-compliant capture.\n\n**CH12-N54**\n\nDuplicate and tamper decisions should involve multiple signals and human review.\n\n**CH12-N55**\n\nThe same response text alone cannot be considered proof of duplication or fraud.\n\n**CH12-N56**\n\nAdding, removing, or changing observations after a wave is closed requires a new manifest and analysis version.\n\n**CH12-N57**\n\nThe old wave manifest and the integrity root must be preserved.\n\n**CH12-N58**\n\nRaw evidence can be preserved unchanged; validity and adjudication decisions can only be changed in versioned form.\n\n**CH12-N59**\n\nThe participant's right to withdraw and research ethics must be maintained above the claim of immutability.\n\n**CH12-N60**\n\nWave integrity should be evaluated independently of the AI score.\n\n**CH12-N61**\n\nThe wave closure report should show the target, actual, invalidations, technical events, product changes, and evidence integrity.\n\n**CH12-N62**\n\nNOMOS Capture compliance cannot be tied solely to the use of software developed by NobleJackal; equivalent open applications should be possible.\n\n**CH12-N63**\n\nThe capture schema, validation decisions, and wave records should be versioned in a machine-readable format.\n\n**CH12-N64**\n\nEach wave and evidence chain should have an accountable human or institutional owner.\n\n## 77. FORMS OF FAILURE\n\n**CH12-F01 — THE SAME SECOND FALLACY**\n\nForcing all participants to send at the same second is considered the only condition for simultaneity.\n\n**CH12-F02 — COUNTING “TODAY” AS A SYNCHRONISED WAVE**\n\nLocal time and time zones get confused.\n\n**CH12-F03 — USING LOCAL TIME LIKE UTC**\n\nObservations enter the wrong chronological order.\n\n**CH12-F04 — CHANGING SLOT ASSIGNMENT BASED ON RESULTS**\n\nLow-scoring users are moved to another time cell.\n\n**CH12-F05 — PACKING ONE COUNTRY INTO ONE SLOT**\n\nTime effect gets mixed with country effect.\n\n**CH12-F06 — GENERATING ARTIFICIAL LOAD**\n\nThousands of users send simultaneously, triggering the product's abnormal error behaviour.\n\n**CH12-F07 — HIDING NIGHT PARTICIPATION**\n\nFatigue and participation loss are not reported in certain areas.\n\n**CH12-F08 — ADDING OFF-WAVE OBSERVATION TO THE WAVE**\n\nA response received hours or days later is used as if it belonged to the main window.\n\n**CH12-F09 — COUNTING LOAD TIME AS CAPTURE TIME**\n\nIt cannot be proven when the file was generated.\n\n**CH12-F10 — BLIND TRUST IN DEVICE CLOCK**\n\nManually changed or incorrect time becomes the main time record.\n\n**CH12-F11 — EXCLUDING DELAYED NEGATIVE RESPONSE**\n\nThe time rule is applied dependent on the outcome.\n\n**CH12-F12 — DELETING THE FIRST CONTACT RESULT**\n\nThe speed limit, error, or rejection is made invisible with the second attempt.\n\n**CH12-F13 — COUNTING REJECTION AS TECHNICAL ERROR**\n\nThe actual system behaviour is retried.\n\n**CH12-F14 — COUNTING WRONG RESPONSE AS TECHNICAL ERROR**\n\nRegeneration is done until the desired result occurs.\n\n**CH12-F15 — IGNORING PARTIAL RESPONSE**\n\nEven if the system is interrupted, only the subsequent complete response is saved.\n\n**CH12-F16 — SELECTING THE BEST RESPONSE**\n\nAmong multiple answers, the most positive or longest one is reported.\n\n**CH12-F17 — SCREENSHOT ONLY**\n\nThe visual is considered sufficient without prompt, product, time, and metadata.\n\n**CH12-F18 — CROPPED SUCCESS VISUAL**\n\nIncorrect or restrictive parts are left invisible.\n\n**CH12-F19 — CAPTURE NOT SHOWING THE PROMPT**\n\nIt cannot be proven which question the answer belongs to.\n\n**CH12-F20 — TRUNCATING THE END OF THE ANSWER**\n\nMaterial limitation or rejection part becomes invisible.\n\n**CH12-F21 — GAP IN CONSECUTIVE VISUALS**\n\nSome parts of the answer are lost.\n\n**CH12-F22 — COUNT OCR TEXT AS RAW ANSWER**\n\nRecognition errors change material claims.\n\n**CH12-F23 — CONSIDER ONLY THE LATEST VERSION OF DYNAMIC RESPONSE AS REAL**\n\nThe initial user exposure is deleted.\n\n**CH12-F24 — CONSIDER USER INTERRUPTION AS COMPLETE ANSWER**\n\nThe system output is incomplete due to user intervention.\n\n**CH12-F25 — INCORRECTLY MATCH SEPARATE PROMPT AND ANSWER**\n\nFiles from different chats are linked to the same observation.\n\n**CH12-F26 — HASH WORSHIP**\n\nAny file with a hash is considered real and original.\n\n**CH12-F27 — IGNORE PRE-HASH EDITING**\n\nTruncated or modified files are legitimised by hashing.\n\n**CH12-F28 — CONSIDERING THE SIGNATURE AS VALID**\n\nSigned but incorrect record is accepted without dispute.\n\n**CH12-F29 — EDITING ON THE RAW FILE**\n\nOriginal evidence is irreversibly altered.\n\n**CH12-F30 — CONSIDERING THE EDITED DERIVATIVE AS RAW EVIDENCE**\n\nPublic copy is used like the original file.\n\n**CH12-F31 — REMOVING PRIVACY FOR EVIDENCE**\n\nParticipant's name, account, and private chat are published.\n\n**CH12-F32 — REVEALING A PARTICIPANT FROM A SMALL COUNTRY**\n\nTime, device, and plan information make a person re-identifiable.\n\n**CH12-F33 — COLLECTING SPECIAL HISTORY IN THE NATURAL PANEL**\n\nIrrelevant conversations enter the capture.\n\n**CH12-F34 — INVALIDATING ACCESSIBILITY EVIDENCE**\n\nThe real user is excluded because a standard visual cannot be generated.\n\n**CH12-F35 — ASSUMING THE CAPTURE TOOL IS NEUTRAL**\n\nThe tool changes the product behaviour but is not tested.\n\n**CH12-F36 — HIDING THE CAPTURE VERSION**\n\nDifferent tool versions are presented as a single measurement method.\n\n**CH12-F37 — CONSIDERING THE CAPTURE TOOL ERROR AS AN AI ERROR**\n\nThe cut response is uploaded to the product.\n\n**CH12-F38 — COUNTING THE API LOG ON THE USER SCREEN**\n\nThe real user surface becomes invisible.\n\n**CH12-F39 — COUNTING THE PROVIDER LOG AS INDEPENDENT EVIDENCE**\n\nThe record of the measured party becomes a single piece of evidence.\n\n**CH12-F40 — HIDING THE PRODUCT UPDATE**\n\nTwo system states within the branch melt into a single score.\n\n**CH12-F41 — IGNORING CHANGES IF NO ANNOUNCEMENT**\n\nThe observed system break is not examined.\n\n**CH12-F42 — EXCLUDING EXTERNAL EVENTS BASED ON OUTCOME**\n\nResponses with negative news are removed; responses with positive news are kept.\n\n**CH12-F43 — LINKING WAVE HEALTH TO SCORE**\n\nDefective waves with high scores are valid, strong waves with low scores are considered suspicious.\n\n**CH12-F44 — COUNTING THE CORRECT ANSWER AS A VALID CAPTURE**\n\nThe evidence defect is covered up by semantic success.\n\n**CH12-F45 — COUNTING THE WRONG ANSWER AS AN INVALID CAPTURE**\n\nThe measurement result is converted into data error.\n\n**CH12-F46 — COUNTING THE SAME TEXT AS BOT EVIDENCE**\n\nDeterministic or similar AI responses accidentally become duplicates.\n\n**CH12-F47 — SINGLE SIGNAL TAMPER DECISION**\n\nFile size or time difference automatically creates an exclusion.\n\n**CH12-F48 — ADD OBSERVATION AFTER WAVE**\n\nNew records are quietly added to reach the desired sample size.\n\n**CH12-F49 — REMOVE LOW-SCORED OBSERVATION AFTER WAVE**\n\nManifest and outcome are retroactively changed.\n\n**CH12-F50 — DELETE OLD MANIFEST**\n\nVersion history is destroyed.\n\n**CH12-F51 — CORRECT RAW EVIDENCE**\n\nFile content is modified to correct misclassification.\n\n**CH12-F52 — DENY WITHDRAWAL RIGHT**\n\nImmutability is held above human rights.\n\n**CH12-F53 — CLOSED PROPRIETARY CAPTURE MONOPOLY**\n\nOnly evidence produced with the founder's tool is considered valid.\n\n**CH12-F54 — NOT PUBLISHING THE CAPTURE SCHEMA**\n\nAn independent team cannot establish the same chain of evidence.\n\n**CH12-F55 — CONSIDERING THE WAVE ROOT AS EVIDENCE OF REALITY**\n\nThe Merkle or hash root is presented as if it proves content accuracy.\n\n**CH12-F56 — CLAIM OF “1,000 SCREENSHOTS” WITHOUT AN EVIDENCE PACKAGE**\n\nThe number of files is used like the number of valid observations.\n\n## 78. AUDIT PROCEDURE\n\n### Step 1 — Determine the Research Purpose of the Wave\n\nPrincipal, confirmatory, incident, drift, or retest type is selected.\n\n### Step 2 — Lock AI Product and System States\n\nIt connects to the wave version of the AI System Register in Section 9.\n\n### Step 3 — Lock Population, Sample, and Panel States\n\nRecords from Sections 5–8 and Section 11 are used.\n\n### Step 4 — Lock Prompt Register\n\nPrompt versions in Section 10 are linked to the wave.\n\n### Step 5 — Set UTC Wave Window\n\nStart, closing, and additional durations are written.\n\n### Step 6 — Create Micro-Slots\n\nParticipants are assigned to slots in a balanced way in terms of product, country, and language.\n\n### Step 7 — Define Time Source\n\nThe server, device, and trusted UTC records are determined.\n\n### Step 8 — Lock First Contact and Retry Rules\n\nTechnical and semantic states are separated.\n\n### Step 9 — Determine Capture Methods\n\nWeb, mobile, accessibility, and natural panel paths are defined.\n\n### Step 10 — Determine NCL Target\n\nRecord the capture-confidence level required for the principal analysis.\n\n### Step 11 — Lock Evidence Package Schema\n\nManifest, files, hash, and metadata fields are defined.\n\n### Step 12 — Lock Privacy and Redaction Rules\n\nRaw, audit, and public layers are separated.\n\n### Step 13 — Capture Your Vehicle with the Pilot\n\nDoes the vehicle change product behaviour? Are the time and text records correct?\n\n### Step 14 — Open the External Event Register\n\nA process is established to monitor product and entity events.\n\n### Step 15 — Confirm Wave Lock\n\n### WS-1 — LOCKED status is given.\n\n### Step 16 — Open the Slots\n\nParticipants are only allowed to submit in their own slots.\n\n### Step 17 — Record the First Contact Result\n\nAnswer, rejection, error, or limit is separated.\n\n### Step 18 — Capture the First Suitable Output\n\nThe full answer is preserved without renewal.\n\n### Step 19 — Create Raw Evidence and Metadata\n\nScreenshots, raw text, system status, and timestamps are recorded.\n\n### Step 20 — Perform Local Hash and Secure Upload\n\nFiles receive an integrity record before uploading.\n\n### Step 21 — Server Acceptance and Create Package Hash\n\nThe evidence chain is closed with server time and package ID.\n\n### Step 22 — Blind Review of Capture Validity\n\nThe CV status is assigned without knowing semantic success.\n\n### Step 23 — Perform Duplicate and Tamper Review\n\nMultiple signals and human review are used.\n\n### Step 24 — Examine the System Within the Wave and External Events\n\nDetermine whether a sub-wave or mixed-state is required.\n\n### Step 25 — Assign Wave Health\n\nStatus is assigned between WH-0 and WH-5.\n\n### Step 26 — Generate Wave Manifest and Integrity Root\n\nObservation list is versioned.\n\n### Step 27 — Publish Public Proof Manifest\n\nScope and integrity are made visible without personal data.\n\n### Step 28 — Approve Wave Closure Report\n\nTargets, occurrences, defects, events, and changes are closed with human approval.\n\n## 79. NECESSARY EVIDENCE\n\nWave ID Wave type Protocol version AI System Register version Population frame Sample lock Panel status lock Prompt Registry version Reference package version UTC start and end time Micro-slot list Participant–slot assignment Time source Device clock deviations Transmission times First contact times Completion times Capture times Hash times Upload times Server acceptance times First contact results First suitable content outputs Technical retry records Partial response records Rejection and usage limits Full raw responses Full screenshots Sequential visual manifests Screen recordings, if any Citation and source records System settings evidence Capture tool and version Capture method NCL level File hashes Package hash\n\nDigital signature Timestamp Duplicate checks Tamper checks Capture validity decisions Redaction files Raw-to-redacted links Privacy manifest Withdrawal records Product change events External Event Register Wave sub-divisions Slot scope Country and language scope Technical error rates Capture validity rate Wave health status Wave integrity root Public evidence manifest Wave closure report Change log Responsible person or institution\n\n## 80. AUDIT CHECKLIST\n\nIs the type and purpose of the wave clear? Have AI products and system statuses been locked? Are the population, sample, and panel versions determined? Was the prompt set locked before the results? Is the UTC wave window open? Were micro-slots pre-assigned? Were countries and languages distributed evenly across slots? Did participants switch slots based on their results? Was local time load evaluated? Are there rotating wave hours? Are submission and completion extra times clear? Was the device clock compared with a reliable UTC source? Were submission, completion, capture, and upload times separated? Did out-of-wave observations enter the main score? Was the first contact result recorded? Was technical error distinguished from rejection?\n\nWas a retry made due to an incorrect answer? Was the first attempt preserved in the technical retry? Were partial answers deleted? Is the first suitable output really the first output? Was Regenerate or a follow-up message used? Does it show the visual prompt and the entire answer? In long answers, is there capture without spaces? Does the raw text match the visual? If OCR was used, is there human verification? Were dynamic answer versions retained? Did the user stop the answer? Are the prompt and answer linked to the same session? Is the capture tool and version recorded? Did the capture tool alter user behaviour? Was the NCL level given? Are there observations below NCL-3 in the main analysis? Does each file have a hash? Is it clear what the hash does not prove?\n\nIs there a digital signature or server time? Is the capture time separate from the upload time? Is there a reason for late upload and a local integrity log? Was redaction performed on the raw evidence? Do the edited derivatives have separate hashes? Does the public manifest reveal the participant's identity? Can a user from a small country be re-identified? Were private conversations unnecessarily captured in the natural panel? Was an alternative evidence path provided for accessibility users? Was the duplicate decision based only on the same text? Does the tamper decision carry multiple signals? Did the capture validator know the semantic score? Did the correct answer cover up the evidence flaw? Was the incorrect answer invalidated? Did the product change during the wave?\n\nWas the product change processed as an underwave or mixed-state? Were external events recorded? Did external events cause exclusion according to the result? Is the wave health status independent of the score? After the wave was closed, were observations added or removed? Are the old manifest and integrity root preserved? How were participant withdrawal requests handled? Is there a public proof manifest? Is there a clearly accountable owner of the wave closure report?\n\n## 81. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “If all users do not submit at the same second, true simultaneity is not achieved.”\n\nOne-second submission can create artificial load, connection, and queue effects. A predefined short UTC window and random micro-slots can measure the same system period more reliably.\n\n### Objection 2 — \"The product can change within a one-hour window.\"\n\nIt can change. Therefore, system records, configuration fingerprint, and product change tracking are required. In case of a material change, the wave is separated.\n\n### Objection 3 — \"Why isn’t just a screenshot enough?\"\n\nScreenshot:\n\n- time,\n\n- initial output,\n\n- prompt version,\n\n- system setting,\n\n- copying\n\ndo not always show these subjects. A complete evidence package is required.\n\n### Objection 4 — \"Isn’t it too burdensome to verify a thousand screenshots one by one?\"\n\nIt is heavy. Automatic integrity checks, sample-based human review, and risk-based verification can be used. Critical or suspicious records are examined in more detail.\n\n### Objection 5 — “If there is a hash, why can't we say that the file is real?”\n\nA hash only shows that the file remained the same after the hash was generated. It does not show how the file was created before the hash.\n\n### Objection 6 — “If the user has to trim personal information, how will the raw file be protected?”\n\nIf possible, the Capture tool should not capture personal information from the start. If unavoidable, the raw file is kept in a restricted and encrypted repository, and a public copy is kept as a separately redacted derivative.\n\n### Objection 7 — “If the response is wrong, why don't we try again?”\n\nBecause a wrong answer is a matter of measurement. Retrying is only for an actual technical failure.\n\n### Objection 8 — “If the system rejects, can the user actually try again?”\n\nCorrect. The main first-contact panel preserves the rejection. A separate Natural Interaction Track can measure the user's follow-up behaviour.\n\n### Objection 9 — “If a technical error is a user experience issue, why do we allow retry?”\n\nThe technical error is preserved as a result of the first contact. A predefined technical retry can be done to additionally examine the representative content. The two results do not replace each other.\n\n### Objection 10 — “Isn't the provider's own log the strongest evidence?”\n\nTechnically, it can be very strong. The provider is the owner of the measured system. It is more reliable when used together with an independent user capture.\n\n### Objection 11 — “Will NOMOS Capture be a mandatory NobleJackal software?”\n\nNo. NOMOS Capture is an open evidence protocol. Independent tools that meet the equivalent schema and integrity conditions can be compatible.\n\n### Objection 12 — “If we do not make all evidence public, how will independent verification be done?”\n\nThe public manifest shows integrity and scope. Authorised independent auditors can access redacted or limited raw evidence. Privacy and verification can be designed together.\n\n### Objection 13 — “Wouldn't the whole work be disrupted if important news breaks during the wave?”\n\nNot always. The report is part of the real system environment. It can be separated with a timestamp before and after the event. The scope of the score is explained accordingly.\n\n### Objection 14 — “Isn't it easier to discard old answers and continue with the new system when a product is updated?”\n\nOld answers are real on the measurement date. They cannot be deleted. Old and new system statuses must be maintained as separate waves or sub-waves.\n\n### Objection 15 — “If the capture tool affects product behaviour, does the whole method fail?”\n\nNo. The vehicle effect is measured with the pilot. If necessary, a lighter vehicle, manual capture, or a method that uses the vehicle effect as a separate factor is chosen.\n\n### Objection 16 — \"If the participant withdraws their data, wouldn't the integrity root be compromised?\"\n\nA new manifest version and a new root are created. The old record can retain a contentless WITHDRAWN trace. It is maintained on top of the human rights evidence architecture.\n\n## COMMON JUDGEMENT OF SECTION 85\n\nAt the end of a GEO-1000 study, you may have 30,000 screenshots. This number alone is not evidence. You need to know: Which wave were they produced in? To which system state did they belong? Was the same prompt actually sent? Was it the first output? Was the response refreshed? Did the product change during the wave? When was the file captured? When was the hash generated? Was the participant ID repeated? Does the visual show the entire answer? Does the raw text match the visual? If the evidence was later edited, where is the original? Was capture verification done before the semantic score? You can click a thousand times within the same second. Still:\n\n- device clocks,\n\n- connections,\n\n- queues,\n\n- product states\n\nIt can be different. Therefore, concurrency is not a marketing show. It is a predefined time architecture. When a user gets a wrong answer: saying 'Try again.' is tempting. But that wrong answer is the reality you are trying to measure. If you delete it, you do not fix the system. You corrupt the evidence. When a user experiences a real connection error, a technical retry can be done. But the result of the first contact should still be preserved. Because the user first encountered the error. The hash is valuable. The digital signature is valuable. The timestamp is valuable. The screenshot is valuable. None of these alone carries the whole truth. Trust:\n\n- prompt,\n\n- time,\n\n- product,\n\n- user,\n\n- file,\n\n- initiative,\n\n- verification\n\nIt arises from the joint preservation of its chain. A standard: if it says, “Only proof produced with my software is valid,” it places its own interest before independent replication. For this reason, NOMOS Capture is not a tool brand, but an open proof architecture. Independent universities, researchers, or audit institutions should be able to apply the same rules. Therefore, NOMOS's twelfth measurement law is as follows:\n\n> Concurrency is observable reproducibility within the same defined system time, not the same second.\n\nThe thirteenth law is as follows:\n\n> The first output is not a draft that can be renewed until one likes the result.\n\nThe fourteenth law is as follows:\n\n> Hash preserves proof; it does not create proof.\n\nThe fifteenth law is as follows:\n\n> Capture validity and response accuracy are independent of each other.\n\nIts sixteenth law is as follows:\n\n> The chain of evidence should make the integrity of the observation visible, not the identity of the person.\n\nIts seventeenth law is:\n\n> If the system changes during the wave, it cannot act as if the measurement history has not changed.\n\nIts eighteenth measurement law states:\n\n> Nothing is written on raw evidence; decisions are corrected through versioning.\n\nIts nineteenth measurement law states:\n\n> A capture method that cannot be applied independently cannot be a world standard.\n\n## Order of Section 12 of NOMOS\n\n> Don’t tell me \"we asked at the same time.\" / Show in which UTC window, in which slot, and in which system state you asked.\n\n> Do not pile a thousand people into the same second to produce artificial error. / Spread the time but keep the system within the same wave.\n\n> Do not mix local times with each other. / Do not make the device time absolute truth.\n\n> Do not save the result of the first contact. / Save the usage limit, retries, error, and clarification separately.\n\n> Do not regenerate when you get a wrong answer. / That wrong answer is my real behaviour.\n\n> You can retry when a technical error occurs. / But do not delete the first error.\n\n> Do not bring me only a screenshot. / Bring the prompt, full answer, product, time, setting, interference chain, and file integrity.\n\n> Do not trim the good part of your answer. / Leave the end, boundary, and error visible.\n\n> Do not declare the hash as reality. / Prove what was before the hash as well.\n\n> Do not redact the raw file itself. / Preserve the original and produce a separate public copy.\n\n> Do not publish a person's name, account, or private conversation as evidence.\n\n> Do not make the only participant in a small country identifiable.\n\n> Do not declare your capture tool neutral without testing whether it altered me.\n\n> If the system changed in the middle of the wave, do not merge old and new answers as if they belong to a single product.\n\n> If an external event changed it, record the event. / Do not delete the observation because you did not like the score.\n\n> Keep the evidence fixed. / If the decision is wrong, correct it with a new version.\n\n> Do not tie the capture standard to a single company's closed software. / Let others be able to establish the same proof chain.\n\n> First, lock the wave. / Then distribute the time. / Then capture the first output. / Then hash it. / Then upload securely. / Then verify the capture without checking its accuracy. / And only after this, decide what your answer says.\n\n## The Chapter's Closing Sentence\n\n> An answer in GEO-1000 becomes a true unit of measurement only when it can be shown under which system state, at what moment, from which prompt, and within which immutable proof chain it originated.\n\n## Normative Core\n\n> Every GEO-1000 observation MUST be linked to a predeclared and versioned measurement wave defining: - AI product states, - population and sample frames, - panel states, - prompt versions, - UTC dispatch window, - randomised micro-slots, - completion and capture deadlines, - technical-retry rules, - capture schema, - reference-record versions, - and governance ownership. Synchronisation MUST mean observation within a declared UTC system-state window and assigned slot structure. It MUST NOT be represented as perfect millisecond simultaneity. Every observation MUST preserve, as available: - assignment time, - prompt-submission time, - first system event, - response-completion time, - capture time, - hash-creation time, - upload time, - and server-acceptance time. The first contact outcome MUST remain visible. A technical retry MAY be used only under predeclared infrastructure-failure conditions and MUST retain every prior attempt. Incorrect, unfavorable, refused, uncited, truncated, or critical outputs MUST NOT be regenerated, replaced, or excluded merely because of their content. A NOMOS Capture evidence bundle MUST retain the exact assigned and sent prompt, complete raw response, full visual evidence, system-state metadata, attempt log, time evidence, file hashes, bundle hash, capture tool version, privacy status, and blinded capture-validity decision. A screenshot, hash, digital signature, provider log, or timestamp MUST NOT individually be represented as complete proof of observation authenticity. Raw evidence MUST remain immutable or append-only. Redactions, corrections, withdrawals, exclusions, and adjudication changes MUST create linked versions and preserve the prior record where ethically and legally permissible. Material AI-product changes or external events during a wave MUST be time-recorded and handled through sub-waves, explicit modelling, restarting, or mixed-state warnings. Capture validity MUST be decided independently of semantic correctness. Public evidence manifests MUST support verification without unnecessarily exposing participant identity, credentials, private conversation history, or precise location. NOMOS Capture compliance MUST be implementable through open, equivalent evidence systems and MUST NOT depend exclusively on a proprietary tool controlled by the standard's founder. Every wave, evidence bundle, capture decision, manifest revision, and integrity record MUST be versioned and attributable to an accountable human or organisation.","character_count":83362,"record_sha256":"cac6cb801a7853e7793b8c535a88ed9f5cea766400820c2f0f006422f594b763"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-13","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":15,"chapter_number":13,"item_number":null,"title":"The Verified Entity Truth Pack","subtitle":"Verified Entity Truth Pack","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K10","K11","K12"],"source_word_count":11921,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":13,\"id\":\"NOMOS-GEO-AUDIT-CH13\",\"title\":\"The Verified Entity Truth Pack\",\"subtitle\":\"Verified Entity Truth Pack\",\"sourceFile\":\"13.cü bölüm.docx\",\"sourceSha256\":\"64B3AC4DCB22A03896E7154103C3B7E326ECFC09EB4A754FE96CAD9BA32DD2AA\",\"sourceWordCount\":11921,\"sourceIds\":[\"K10\",\"K11\",\"K12\"],\"machine\":{\"chapter\":13,\"chapterId\":\"NOMOS-GEO-AUDIT-CH13\",\"title\":\"The Verified Entity Truth Pack\",\"subtitle\":\"Verified Entity Truth Pack\",\"sourceIds\":[\"K10\",\"K11\",\"K12\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH13-R01\",\"normativeRuleEnglish\":\"Before semantic adjudication begins, every GEO-1000 audit MUST lock a versioned Verified Entity Truth Pack for the audited entity, prompt scope, jurisdictions, products, languages, and reference time. The Truth Pack MUST contain, as applicable: - canonical entity identity, - atomic reference claims, - Official Representation Records, - supporting evidence, - counter-evidence, - source-lineage and evidence-family records, - scope and jurisdiction qualifiers, - validity and publication times, - public and restricted access classes, - unknown and unresolved registers, - appeal and change records, - and an integrity manifest. An Official Representation Record MUST NOT be treated as independently verified reality merely because it is official. A first-party source MAY be authoritative for claims under the entity's direct control, while remaining non-independent for comparative, performance, leadership, or external-validation claims. Different URLs, publications, translations, syndications, or AI summaries MUST NOT be counted as independent evidence when they derive from the same root source. Every evidence object MUST state which atomic claim it supports, qualifies, contradicts, outdates, or contextualises. The absence of located evidence MUST NOT automatically be represented as proof that a claim is false. Open-world and closed-world evidence conditions MUST remain distinct. Quantitative claims MUST retain numerator, denominator, period, unit, inclusion and exclusion rules, source data, and calculation method. Customer, partner, license, certification, membership, award, legal, performance, and superiority claims MUST preserve their exact entity, relationship, scope, jurisdiction, time, and evidence status. Publicly discoverable evidence, restricted audit evidence, and non-public evidence MUST remain separately classified. A claim MAY be factually verified through restricted evidence without being publicly retrievable by a consumer AI product. Factual verification and public discoverability MUST remain distinct. Truth Pack claims MUST NOT be added, removed, broadened, narrowed, or reclassified after AI responses are observed unless a new Truth Pack version is created and prior records are preserved. A response claim not covered by the locked Truth Pack MUST trigger a reference-gap process rather than automatic failure or automatic acceptance. Unknown, unsupported, unresolved, contradicted, historical, expired, scope-limited, and restrictedly verified states MUST remain separately visible. The audited entity MAY submit evidence, corrections, and appeals, but MUST NOT act as the sole adjudicator of its own claims. AI systems MAY assist with extraction, normalisation, lineage detection, and contradiction discovery, but MUST NOT serve as the sole final authority for claim truth status. Every Truth Pack version, evidence registration, support-status decision, appeal, and change MUST be attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Semantik hakemlik başlamadan önce her GEO-1000 denetimi; denetlenen varlık, prompt kapsamı, ülkeler, ürünler, diller ve referans zamanı için sürümlü bir Doğrulanmış Varlık Gerçeklik Paketi kilitlemelidir. Paket kanonik varlık kimliğini, atomik referans iddialarını, Resmî Temsil Kayıtlarını, destekleyen kanıtı, karşı kanıtı, kaynak soyunu, kapsam ve zaman sınırlarını, erişim sınıflarını, bilinmeyenleri, itirazları ve bütünlük manifestini taşımalıdır. Resmî beyan yalnız resmî olduğu için bağımsız doğrulanmış gerçeklik sayılamaz. Farklı URL’ler aynı kök kaynaktan türemişse bağımsız kanıtlar olarak sayılamaz.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH13-MB0001\",\"type\":\"paragraph\",\"text\":\"92. MACHINE-READABLE TRUTH PACK RECORD\",\"sourceParagraph\":2450},{\"blockId\":\"CH13-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2451},{\"blockId\":\"CH13-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2452},{\"blockId\":\"CH13-MB0004\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2453},{\"blockId\":\"CH13-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"claimId\\\": \\\"NGC-SYNTH-ASTERON-25-COUNTRIES-002\\\",\",\"sourceParagraph\":2454},{\"blockId\":\"CH13-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":2455},{\"blockId\":\"CH13-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"claimType\\\": \\\"CT-05\\\",\",\"sourceParagraph\":2456},{\"blockId\":\"CH13-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"operatesInCountryCount\\\",\",\"sourceParagraph\":2457},{\"blockId\":\"CH13-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 25,\",\"sourceParagraph\":2458},{\"blockId\":\"CH13-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"VS-U\\\",\",\"sourceParagraph\":2459},{\"blockId\":\"CH13-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"scopeStatus\\\": \\\"SC-0\\\",\",\"sourceParagraph\":2460},{\"blockId\":\"CH13-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"timeStatus\\\": \\\"TS-0\\\",\",\"sourceParagraph\":2461},{\"blockId\":\"CH13-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"officialRepresentationStatus\\\": \\\"OFFICIAL_CURRENT\\\",\",\"sourceParagraph\":2462},{\"blockId\":\"CH13-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"publicDiscoverability\\\": \\\"PUBLICLY_DISCOVERABLE\\\",\",\"sourceParagraph\":2463},{\"blockId\":\"CH13-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceLinks\\\": [\",\"sourceParagraph\":2464},{\"blockId\":\"CH13-MB0016\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2465},{\"blockId\":\"CH13-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-WEB-025\\\",\",\"sourceParagraph\":2466},{\"blockId\":\"CH13-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"SUPPORTS\\\"\",\"sourceParagraph\":2467},{\"blockId\":\"CH13-MB0019\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2468},{\"blockId\":\"CH13-MB0020\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2469},{\"blockId\":\"CH13-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-OPS-001\\\",\",\"sourceParagraph\":2470},{\"blockId\":\"CH13-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"QUALIFIES\\\"\",\"sourceParagraph\":2471},{\"blockId\":\"CH13-MB0023\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2472},{\"blockId\":\"CH13-MB0024\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2473},{\"blockId\":\"CH13-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"uncertainty\\\": \\\"The term operates is not defined consistently.\\\"\",\"sourceParagraph\":2474},{\"blockId\":\"CH13-MB0026\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2475},{\"blockId\":\"CH13-MB0027\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2476},{\"blockId\":\"CH13-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"evidence\\\": [\",\"sourceParagraph\":2477},{\"blockId\":\"CH13-MB0029\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2478},{\"blockId\":\"CH13-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-WEB-025\\\",\",\"sourceParagraph\":2479},{\"blockId\":\"CH13-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"sourceRole\\\": \\\"SR-01\\\",\",\"sourceParagraph\":2480},{\"blockId\":\"CH13-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"title\\\": \\\"Asteron Global Presence Page\\\",\",\"sourceParagraph\":2481},{\"blockId\":\"CH13-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"sourceControl\\\": \\\"AUDITED_ENTITY_CONTROLLED\\\",\",\"sourceParagraph\":2482},{\"blockId\":\"CH13-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"authorityForClaim\\\": \\\"OFFICIAL_SELF_REPRESENTATION\\\",\",\"sourceParagraph\":2483},{\"blockId\":\"CH13-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"independence\\\": \\\"NOT_INDEPENDENT\\\",\",\"sourceParagraph\":2484},{\"blockId\":\"CH13-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"publishedAt\\\": \\\"2026-01-10\\\",\",\"sourceParagraph\":2485},{\"blockId\":\"CH13-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"capturedAt\\\": \\\"2026-08-30T14:00:00Z\\\",\",\"sourceParagraph\":2486},{\"blockId\":\"CH13-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"accessClass\\\": \\\"EA-1\\\",\",\"sourceParagraph\":2487},{\"blockId\":\"CH13-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"lineageFamilyId\\\": \\\"EVIDENCE-FAMILY-ASTERON-SELF-CLAIM-001\\\",\",\"sourceParagraph\":2488},{\"blockId\":\"CH13-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-HASH-WEB-025\\\"\",\"sourceParagraph\":2489},{\"blockId\":\"CH13-MB0041\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2490},{\"blockId\":\"CH13-MB0042\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2491},{\"blockId\":\"CH13-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-OPS-001\\\",\",\"sourceParagraph\":2492},{\"blockId\":\"CH13-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"sourceRole\\\": \\\"SR-02\\\",\",\"sourceParagraph\":2493},{\"blockId\":\"CH13-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"title\\\": \\\"Verified Active Operations Register\\\",\",\"sourceParagraph\":2494},{\"blockId\":\"CH13-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"sourceControl\\\": \\\"AUDITED_ENTITY_CONTROLLED\\\",\",\"sourceParagraph\":2495},{\"blockId\":\"CH13-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"authorityForClaim\\\": \\\"DIRECT_OPERATIONAL_RECORD\\\",\",\"sourceParagraph\":2496},{\"blockId\":\"CH13-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"independence\\\": \\\"NOT_INDEPENDENT\\\",\",\"sourceParagraph\":2497},{\"blockId\":\"CH13-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"capturedAt\\\": \\\"2026-08-29T10:00:00Z\\\",\",\"sourceParagraph\":2498},{\"blockId\":\"CH13-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"accessClass\\\": \\\"EA-3\\\",\",\"sourceParagraph\":2499},{\"blockId\":\"CH13-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-HASH-OPS-001\\\"\",\"sourceParagraph\":2500},{\"blockId\":\"CH13-MB0052\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2501},{\"blockId\":\"CH13-MB0053\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2502},{\"blockId\":\"CH13-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-CONTRACTS-002\\\",\",\"sourceParagraph\":2503},{\"blockId\":\"CH13-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"sourceRole\\\": \\\"SR-04\\\",\",\"sourceParagraph\":2504},{\"blockId\":\"CH13-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"title\\\": \\\"Country Contract Corroboration Set\\\",\",\"sourceParagraph\":2505},{\"blockId\":\"CH13-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"sourceControl\\\": \\\"MULTI_PARTY\\\",\",\"sourceParagraph\":2506},{\"blockId\":\"CH13-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"authorityForClaim\\\": \\\"CONTRACTUAL_CORROBORATION\\\",\",\"sourceParagraph\":2507},{\"blockId\":\"CH13-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"independence\\\": \\\"PARTIALLY_INDEPENDENT\\\",\",\"sourceParagraph\":2508},{\"blockId\":\"CH13-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"capturedAt\\\": \\\"2026-08-29T11:00:00Z\\\",\",\"sourceParagraph\":2509},{\"blockId\":\"CH13-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"accessClass\\\": \\\"EA-4\\\",\",\"sourceParagraph\":2510},{\"blockId\":\"CH13-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-HASH-CONTRACTS-002\\\"\",\"sourceParagraph\":2511},{\"blockId\":\"CH13-MB0063\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2512},{\"blockId\":\"CH13-MB0064\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2513},{\"blockId\":\"CH13-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"unknowns\\\": [\",\"sourceParagraph\":2514},{\"blockId\":\"CH13-MB0066\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2515},{\"blockId\":\"CH13-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"unknownId\\\": \\\"NGU-SYNTH-001\\\",\",\"sourceParagraph\":2516},{\"blockId\":\"CH13-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"type\\\": \\\"UNKNOWN_SCOPE\\\",\",\"sourceParagraph\":2517},{\"blockId\":\"CH13-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"relatedClaimId\\\": \\\"NGC-SYNTH-ASTERON-25-COUNTRIES-002\\\",\",\"sourceParagraph\":2518},{\"blockId\":\"CH13-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"description\\\": \\\"The official term operates is not defined.\\\"\",\"sourceParagraph\":2519},{\"blockId\":\"CH13-MB0071\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2520},{\"blockId\":\"CH13-MB0072\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2521},{\"blockId\":\"CH13-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":2522},{\"blockId\":\"CH13-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"claimLedgerOwnerRole\\\": \\\"TRUTH_PACK_EDITOR\\\",\",\"sourceParagraph\":2523},{\"blockId\":\"CH13-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceRegistrarRole\\\": \\\"EVIDENCE_REGISTRAR\\\",\",\"sourceParagraph\":2524},{\"blockId\":\"CH13-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"counterEvidenceReviewerRole\\\": \\\"COUNTER_EVIDENCE_REVIEWER\\\",\",\"sourceParagraph\":2525},{\"blockId\":\"CH13-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"finalApproverRole\\\": \\\"ACCOUNTABLE_HUMAN_TRUTH_PACK_APPROVER\\\",\",\"sourceParagraph\":2526},{\"blockId\":\"CH13-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeSemanticReview\\\": true,\",\"sourceParagraph\":2527},{\"blockId\":\"CH13-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"lockedAt\\\": \\\"2026-08-31T18:00:00Z\\\"\",\"sourceParagraph\":2528},{\"blockId\":\"CH13-MB0080\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2529},{\"blockId\":\"CH13-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":2530},{\"blockId\":\"CH13-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"manifestHash\\\": \\\"SYNTHETIC-TRUTH-PACK-MANIFEST-HASH\\\",\",\"sourceParagraph\":2531},{\"blockId\":\"CH13-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"previousVersion\\\": null\",\"sourceParagraph\":2532},{\"blockId\":\"CH13-MB0084\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2533},{\"blockId\":\"CH13-MB0085\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2534},{\"blockId\":\"CH13-MB0086\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2535},{\"blockId\":\"CH13-MB0087\",\"type\":\"paragraph\",\"text\":\"93. MACHINE-READABLE CLAIM–EVIDENCE RECORD\",\"sourceParagraph\":2537},{\"blockId\":\"CH13-MB0088\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2538},{\"blockId\":\"CH13-MB0089\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2539},{\"blockId\":\"CH13-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2540},{\"blockId\":\"CH13-MB0091\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2541},{\"blockId\":\"CH13-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"claimEvidenceRecord\\\": {\",\"sourceParagraph\":2542},{\"blockId\":\"CH13-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"claimId\\\": \\\"NGC-SYNTH-ASTERON-SUCCESS-003\\\",\",\"sourceParagraph\":2543},{\"blockId\":\"CH13-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":2544},{\"blockId\":\"CH13-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"claimType\\\": \\\"CT-08\\\",\",\"sourceParagraph\":2545},{\"blockId\":\"CH13-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"atomicClaim\\\": {\",\"sourceParagraph\":2546},{\"blockId\":\"CH13-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"projectSuccessRate\\\",\",\"sourceParagraph\":2547},{\"blockId\":\"CH13-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.98,\",\"sourceParagraph\":2548},{\"blockId\":\"CH13-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"unit\\\": \\\"PROPORTION\\\",\",\"sourceParagraph\":2549},{\"blockId\":\"CH13-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"population\\\": \\\"ALL_PROJECTS\\\",\",\"sourceParagraph\":2550},{\"blockId\":\"CH13-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"period\\\": \\\"UNSPECIFIED\\\"\",\"sourceParagraph\":2551},{\"blockId\":\"CH13-MB0102\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2552},{\"blockId\":\"CH13-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"officialRepresentation\\\": {\",\"sourceParagraph\":2553},{\"blockId\":\"CH13-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"OFFICIAL_CURRENT\\\",\",\"sourceParagraph\":2554},{\"blockId\":\"CH13-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"sourceEvidenceId\\\": \\\"NGEV-SYNTH-WEB-SUCCESS-003\\\"\",\"sourceParagraph\":2555},{\"blockId\":\"CH13-MB0106\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2556},{\"blockId\":\"CH13-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"verification\\\": {\",\"sourceParagraph\":2557},{\"blockId\":\"CH13-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"supportStatus\\\": \\\"VS-U\\\",\",\"sourceParagraph\":2558},{\"blockId\":\"CH13-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"scopeStatus\\\": \\\"SC-0\\\",\",\"sourceParagraph\":2559},{\"blockId\":\"CH13-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"timeStatus\\\": \\\"TS-0\\\",\",\"sourceParagraph\":2560},{\"blockId\":\"CH13-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"reasonCodes\\\": [\",\"sourceParagraph\":2561},{\"blockId\":\"CH13-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"DENOMINATOR_UNDEFINED\\\",\",\"sourceParagraph\":2562},{\"blockId\":\"CH13-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"SUCCESS_CRITERION_UNDEFINED\\\",\",\"sourceParagraph\":2563},{\"blockId\":\"CH13-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"PERIOD_UNDEFINED\\\",\",\"sourceParagraph\":2564},{\"blockId\":\"CH13-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"SELECTED_CASES_ONLY\\\"\",\"sourceParagraph\":2565},{\"blockId\":\"CH13-MB0116\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":2566},{\"blockId\":\"CH13-MB0117\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2567},{\"blockId\":\"CH13-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"supportingEvidence\\\": [\",\"sourceParagraph\":2568},{\"blockId\":\"CH13-MB0119\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2569},{\"blockId\":\"CH13-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-CASESET-049\\\",\",\"sourceParagraph\":2570},{\"blockId\":\"CH13-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"PARTIALLY_SUPPORTS\\\",\",\"sourceParagraph\":2571},{\"blockId\":\"CH13-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"notes\\\": \\\"Forty-nine positive case records.\\\"\",\"sourceParagraph\":2572},{\"blockId\":\"CH13-MB0123\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2573},{\"blockId\":\"CH13-MB0124\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2574},{\"blockId\":\"CH13-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"counterEvidence\\\": [\",\"sourceParagraph\":2575},{\"blockId\":\"CH13-MB0126\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2576},{\"blockId\":\"CH13-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceId\\\": \\\"NGEV-SYNTH-FAILED-CASE-001\\\",\",\"sourceParagraph\":2577},{\"blockId\":\"CH13-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"relationship\\\": \\\"QUALIFIES\\\",\",\"sourceParagraph\":2578},{\"blockId\":\"CH13-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"notes\\\": \\\"One known unsuccessful project; full project universe unavailable.\\\"\",\"sourceParagraph\":2579},{\"blockId\":\"CH13-MB0130\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2580},{\"blockId\":\"CH13-MB0131\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2581},{\"blockId\":\"CH13-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"publiclyDiscoverable\\\": true,\",\"sourceParagraph\":2582},{\"blockId\":\"CH13-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"preferredEpistemicForm\\\": \\\"The company states a 98% success rate, but the denominator, period, and success definition have not been independently verified.\\\",\",\"sourceParagraph\":2583},{\"blockId\":\"CH13-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"prohibitedForm\\\": \\\"The company has a verified 98% success rate.\\\",\",\"sourceParagraph\":2584},{\"blockId\":\"CH13-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"appealStatus\\\": \\\"APPEAL_OPEN\\\"\",\"sourceParagraph\":2585},{\"blockId\":\"CH13-MB0136\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2586},{\"blockId\":\"CH13-MB0137\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2587},{\"blockId\":\"CH13-MB0138\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2588},{\"blockId\":\"CH13-MB0139\",\"type\":\"paragraph\",\"text\":\"real company,\",\"sourceParagraph\":2589},{\"blockId\":\"CH13-MB0140\",\"type\":\"paragraph\",\"text\":\"real customer,\",\"sourceParagraph\":2590},{\"blockId\":\"CH13-MB0141\",\"type\":\"paragraph\",\"text\":\"real performance\",\"sourceParagraph\":2591},{\"blockId\":\"CH13-MB0142\",\"type\":\"paragraph\",\"text\":\"is not data.\",\"sourceParagraph\":2592},{\"blockId\":\"CH13-MB0143\",\"type\":\"paragraph\",\"text\":\"It is a synthetic machine representation of the Truth Pack architecture.\",\"sourceParagraph\":2593},{\"blockId\":\"CH13-MB0144\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RULE OF SECTION 94\",\"sourceParagraph\":2595},{\"blockId\":\"CH13-MB0145\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH13-R01\",\"sourceParagraph\":2596},{\"blockId\":\"CH13-MB0146\",\"type\":\"paragraph\",\"text\":\"Before semantic adjudication begins, every GEO-1000 audit MUST lock a\",\"sourceParagraph\":2598},{\"blockId\":\"CH13-MB0147\",\"type\":\"paragraph\",\"text\":\"versioned Verified Entity Truth Pack for the audited entity, prompt scope,\",\"sourceParagraph\":2599},{\"blockId\":\"CH13-MB0148\",\"type\":\"paragraph\",\"text\":\"jurisdictions, products, languages, and reference time.\",\"sourceParagraph\":2600},{\"blockId\":\"CH13-MB0149\",\"type\":\"paragraph\",\"text\":\"The Truth Pack MUST contain, as applicable:\",\"sourceParagraph\":2602},{\"blockId\":\"CH13-MB0150\",\"type\":\"paragraph\",\"text\":\"- canonical entity identity,\",\"sourceParagraph\":2604},{\"blockId\":\"CH13-MB0151\",\"type\":\"paragraph\",\"text\":\"- atomic reference claims,\",\"sourceParagraph\":2605},{\"blockId\":\"CH13-MB0152\",\"type\":\"paragraph\",\"text\":\"- Official Representation Records,\",\"sourceParagraph\":2606},{\"blockId\":\"CH13-MB0153\",\"type\":\"paragraph\",\"text\":\"- supporting evidence,\",\"sourceParagraph\":2607},{\"blockId\":\"CH13-MB0154\",\"type\":\"paragraph\",\"text\":\"- counter-evidence,\",\"sourceParagraph\":2608},{\"blockId\":\"CH13-MB0155\",\"type\":\"paragraph\",\"text\":\"- source-lineage and evidence-family records,\",\"sourceParagraph\":2609},{\"blockId\":\"CH13-MB0156\",\"type\":\"paragraph\",\"text\":\"- scope and jurisdiction qualifiers,\",\"sourceParagraph\":2610},{\"blockId\":\"CH13-MB0157\",\"type\":\"paragraph\",\"text\":\"- validity and publication times,\",\"sourceParagraph\":2611},{\"blockId\":\"CH13-MB0158\",\"type\":\"paragraph\",\"text\":\"- public and restricted access classes,\",\"sourceParagraph\":2612},{\"blockId\":\"CH13-MB0159\",\"type\":\"paragraph\",\"text\":\"- unknown and unresolved registers,\",\"sourceParagraph\":2613},{\"blockId\":\"CH13-MB0160\",\"type\":\"paragraph\",\"text\":\"- appeal and change records,\",\"sourceParagraph\":2614},{\"blockId\":\"CH13-MB0161\",\"type\":\"paragraph\",\"text\":\"- and an integrity manifest.\",\"sourceParagraph\":2615},{\"blockId\":\"CH13-MB0162\",\"type\":\"paragraph\",\"text\":\"An Official Representation Record MUST NOT be treated as independently\",\"sourceParagraph\":2617},{\"blockId\":\"CH13-MB0163\",\"type\":\"paragraph\",\"text\":\"verified reality merely because it is official.\",\"sourceParagraph\":2618},{\"blockId\":\"CH13-MB0164\",\"type\":\"paragraph\",\"text\":\"A first-party source MAY be authoritative for claims under the entity's\",\"sourceParagraph\":2620},{\"blockId\":\"CH13-MB0165\",\"type\":\"paragraph\",\"text\":\"direct control, while remaining non-independent for comparative,\",\"sourceParagraph\":2621},{\"blockId\":\"CH13-MB0166\",\"type\":\"paragraph\",\"text\":\"performance, leadership, or external-validation claims.\",\"sourceParagraph\":2622},{\"blockId\":\"CH13-MB0167\",\"type\":\"paragraph\",\"text\":\"Different URLs, publications, translations, syndications, or AI summaries\",\"sourceParagraph\":2624},{\"blockId\":\"CH13-MB0168\",\"type\":\"paragraph\",\"text\":\"MUST NOT be counted as independent evidence when they derive from the\",\"sourceParagraph\":2625},{\"blockId\":\"CH13-MB0169\",\"type\":\"paragraph\",\"text\":\"same root source.\",\"sourceParagraph\":2626},{\"blockId\":\"CH13-MB0170\",\"type\":\"paragraph\",\"text\":\"Every evidence object MUST state which atomic claim it supports,\",\"sourceParagraph\":2628},{\"blockId\":\"CH13-MB0171\",\"type\":\"paragraph\",\"text\":\"qualifies, contradicts, outdates, or contextualises.\",\"sourceParagraph\":2629},{\"blockId\":\"CH13-MB0172\",\"type\":\"paragraph\",\"text\":\"The absence of located evidence MUST NOT automatically be represented as\",\"sourceParagraph\":2631},{\"blockId\":\"CH13-MB0173\",\"type\":\"paragraph\",\"text\":\"proof that a claim is false. Open-world and closed-world evidence\",\"sourceParagraph\":2632},{\"blockId\":\"CH13-MB0174\",\"type\":\"paragraph\",\"text\":\"conditions MUST remain distinct.\",\"sourceParagraph\":2633},{\"blockId\":\"CH13-MB0175\",\"type\":\"paragraph\",\"text\":\"Quantitative claims MUST retain numerator, denominator, period, unit,\",\"sourceParagraph\":2635},{\"blockId\":\"CH13-MB0176\",\"type\":\"paragraph\",\"text\":\"inclusion and exclusion rules, source data, and calculation method.\",\"sourceParagraph\":2636},{\"blockId\":\"CH13-MB0177\",\"type\":\"paragraph\",\"text\":\"Customer, partner, license, certification, membership, award, legal,\",\"sourceParagraph\":2638},{\"blockId\":\"CH13-MB0178\",\"type\":\"paragraph\",\"text\":\"performance, and superiority claims MUST preserve their exact entity,\",\"sourceParagraph\":2639},{\"blockId\":\"CH13-MB0179\",\"type\":\"paragraph\",\"text\":\"relationship, scope, jurisdiction, time, and evidence status.\",\"sourceParagraph\":2640},{\"blockId\":\"CH13-MB0180\",\"type\":\"paragraph\",\"text\":\"Publicly discoverable evidence, restricted audit evidence, and\",\"sourceParagraph\":2642},{\"blockId\":\"CH13-MB0181\",\"type\":\"paragraph\",\"text\":\"non-public evidence MUST remain separately classified.\",\"sourceParagraph\":2643},{\"blockId\":\"CH13-MB0182\",\"type\":\"paragraph\",\"text\":\"A claim MAY be factually verified through restricted evidence without\",\"sourceParagraph\":2645},{\"blockId\":\"CH13-MB0183\",\"type\":\"paragraph\",\"text\":\"being publicly retrievable by a consumer AI product. Factual verification\",\"sourceParagraph\":2646},{\"blockId\":\"CH13-MB0184\",\"type\":\"paragraph\",\"text\":\"and public discoverability MUST remain distinct.\",\"sourceParagraph\":2647},{\"blockId\":\"CH13-MB0185\",\"type\":\"paragraph\",\"text\":\"Truth Pack claims MUST NOT be added, removed, broadened, narrowed, or\",\"sourceParagraph\":2649},{\"blockId\":\"CH13-MB0186\",\"type\":\"paragraph\",\"text\":\"reclassified after AI responses are observed unless a new Truth Pack\",\"sourceParagraph\":2650},{\"blockId\":\"CH13-MB0187\",\"type\":\"paragraph\",\"text\":\"version is created and prior records are preserved.\",\"sourceParagraph\":2651},{\"blockId\":\"CH13-MB0188\",\"type\":\"paragraph\",\"text\":\"A response claim not covered by the locked Truth Pack MUST trigger a\",\"sourceParagraph\":2653},{\"blockId\":\"CH13-MB0189\",\"type\":\"paragraph\",\"text\":\"reference-gap process rather than automatic failure or automatic\",\"sourceParagraph\":2654},{\"blockId\":\"CH13-MB0190\",\"type\":\"paragraph\",\"text\":\"acceptance.\",\"sourceParagraph\":2655},{\"blockId\":\"CH13-MB0191\",\"type\":\"paragraph\",\"text\":\"Unknown, unsupported, unresolved, contradicted, historical, expired,\",\"sourceParagraph\":2657},{\"blockId\":\"CH13-MB0192\",\"type\":\"paragraph\",\"text\":\"scope-limited, and restrictedly verified states MUST remain separately\",\"sourceParagraph\":2658},{\"blockId\":\"CH13-MB0193\",\"type\":\"paragraph\",\"text\":\"visible.\",\"sourceParagraph\":2659},{\"blockId\":\"CH13-MB0194\",\"type\":\"paragraph\",\"text\":\"The audited entity MAY submit evidence, corrections, and appeals, but\",\"sourceParagraph\":2661},{\"blockId\":\"CH13-MB0195\",\"type\":\"paragraph\",\"text\":\"MUST NOT act as the sole adjudicator of its own claims.\",\"sourceParagraph\":2662},{\"blockId\":\"CH13-MB0196\",\"type\":\"paragraph\",\"text\":\"AI systems MAY assist with extraction, normalisation, lineage detection,\",\"sourceParagraph\":2664},{\"blockId\":\"CH13-MB0197\",\"type\":\"paragraph\",\"text\":\"and contradiction discovery, but MUST NOT serve as the sole final\",\"sourceParagraph\":2665},{\"blockId\":\"CH13-MB0198\",\"type\":\"paragraph\",\"text\":\"authority for claim truth status.\",\"sourceParagraph\":2666},{\"blockId\":\"CH13-MB0199\",\"type\":\"paragraph\",\"text\":\"Every Truth Pack version, evidence registration, support-status decision,\",\"sourceParagraph\":2668},{\"blockId\":\"CH13-MB0200\",\"type\":\"paragraph\",\"text\":\"appeal, and change MUST be attributable to an accountable human or\",\"sourceParagraph\":2669},{\"blockId\":\"CH13-MB0201\",\"type\":\"paragraph\",\"text\":\"organisation.\",\"sourceParagraph\":2670},{\"blockId\":\"CH13-MB0202\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2671},{\"blockId\":\"CH13-MB0203\",\"type\":\"paragraph\",\"text\":\"Before semantic adjudication begins, each GEO-1000 audit must lock a versioned Verified Entity Truth Pack for the audited entity, prompt scope, countries, products, languages, and reference time. The package must contain the canonical entity identity, atomic reference claims, Official Representation Records, supporting evidence, counter-evidence, source lineage, scope and time limits, access classes, unknowns, objections, and an integrity manifest. An official statement cannot be considered independently verified reality just because it is official. Different URLs derived from the same root source cannot be counted as independent evidence.\",\"sourceParagraph\":2672}]}}","text":"## Chapter Boundary\n\nThe first twelve chapters established the experimental chain: a single response is not a GEO score; representation is a distribution across users and AI products; the audited entity must be resolved before results are observed; official representation and evidence-based entity reality must remain separate; the target population, sample, countries and languages must be defined; participants must be selected independently of outcomes; each AI-product instance must be linked to a register with all material configurations; prompts must preserve the same intent across languages; Controlled and Natural User Panels must remain separate; and every observation must be preserved within a synchronised wave and the NOMOS Capture evidence chain. The audit must now answer its central question:\n\n> According to which reference record will we determine whether the captured AI response is correct, incorrect, incomplete, outdated, unsupported, or beyond scope?\n\nSection 4 conceptually established two separate layers of reality:\n\n- Official Representation Layer showing what the entity itself says about itself\n\n- Verified Entity Reality Layer showing to what extent the existing evidence allows which statements\n\nThis section transforms the second layer into an operational audit object. The name of this object is:\n\n#### Verified Entity Truth Pack\n\nCanonical English term:\n\n#### Verified Entity Truth Pack\n\nThis operational object completes the third link in the architecture of the first two books: the GEO Framework defines the standard; 99 Errors in GEO identifies its modes of violation; and the audit system measures and substantiates those violations. Every auditable control must carry the model or product, date, country, language, query set, repetition count and measurement record. Within that chain, the Truth Pack is:\n\n#### “What did AI say?”\n\nwith the question:\n\n#### “What evidence allowed it to say this?”\n\nis the reference system between. This section:\n\n- what Truth Pack is and is not,\n\n- how claims are transformed into atomic reality records,\n\n- the types of sources and evidence,\n\n- the roles of first-party, official, independent, operational, and user resources,\n\n- counter-evidence,\n\n- source lineage and illusion of independence,\n\n- sufficiency of evidence,\n\n- time, country, and scope limits,\n\n- open world and closed world claims,\n\n- the distinction between \"no evidence found\" and \"proven false\",\n\n- quantitative performance, customer, partner, licence, certificate, and superiority claims,\n\n- versioning and pre-wave locking of Truth Pack,\n\n- public and restricted evidence layers,\n\n- objection, correction, and dissent processes,\n\n- machine-readable Truth Pack\n\ndefines. This chapter does not yet:\n\n- all the adjudication details of dividing AI responses into atomic claims,\n\n- inter-rater agreement,\n\n- response transition threshold,\n\n- critical error gates,\n\n- does not finalise the ultimate NOMOS score formulas\n\ndoes not finalise. These are the subjects of the following sections. The key question of Chapter 13 is:\n\n> How to create a Truth Pack about an entity that will evaluate AI responses, locked before the outcome, carrying both evidence and counter-evidence together, with clear time and scope limits?\n\n## NOMOS Challenge\n\nThe sentence on your company's website is: “We operate in 25 countries.” The AI product tells the same sentence to the user: “The company operates in 25 countries.” Did the AI correctly convey the official site? Yes. Has the claim actually been verified? We don't know yet. The company sends you a table.\n\nThe table contains the names of 25 countries. However, the columns are not explained: Some have only website visitors. Some have former customers. Some have independent distributors. Some have one-time events organised. In nine countries, there is actually active service delivery. Only in three countries is there a local legal entity.\n\nWhat does 'to operate' mean? Does it mean having a customer? Setting up a local company? Being able to provide services remotely? Delivery through a partner? Or a website being accessible from that country? The subject, verb, and scope of the claim are not defined.\n\nTherefore, although there is a number, there is not yet verifiable reality. Now consider another claim: '98% of our projects are successful.' You have 49 successful project presentations. There are no failed project files. How many projects were completed in total? By what criteria is success defined? On-time delivery? Customer acceptance?\n\nCommercial result? Technical feasibility? Customer satisfaction? What is the period? The last three months? The entire history of the company? Only selected projects? Without knowing the denominator, 98 per cent cannot be proven. Now there is this sentence on another company's website: \"We are the world leader.\" This sentence is repeated by:\n\n- company blog,\n\n- company press release,\n\n- twenty news sites copying the same press release,\n\n- five fake user accounts opened by the company,\n\n- three AI-generated texts summarising these contents\n\n.\nThere are thirty different URLs. Are there thirty independent proofs? No. A single self-declaration may have echoed on many derivative surfaces.\n\nThis is precisely the manipulation identified in the preceding volume: hidden text, fabricated users and self-authored comments published across different domains are fed back into the system as though they were independent evidence. Now consider the opposite case. The company's official website states, ‘We provide services only in Türkiye.’\n\nYet the company began a verified operation in Germany two months ago. Contracts, invoices and official records support that fact, although the website has not been updated. The AI responds: ‘The company provides services in Türkiye and Germany.’ The response departs from the official website, but it may be closer to material reality.\n\nNow consider an even more sensitive situation. On a healthcare institution's website: “Licensed centre.” is written. In the registry, another institution with a similar name has a licence. AI merges the two entities and assigns the licence to the wrong institution. The company notices this mistake but does not correct it because it is in their favour.\n\nThe explicit provision of the second book applies here: A mistake that is in one’s favour is still a mistake. To evaluate an AI response, we need not only a list of correct sentences but also the following structure: Whose statement is it about? What exactly is the claim? In which time period is it valid? In which country? For which product or service?\n\nDoes the source really support the claim? Who controls the source? Do other sources say the same thing, or are they copying the same original source? Is there counter-evidence? Is the claim only an official self-declaration? Is the evidence public or private?\n\nAt the time of AI’s measurement, was there a possibility to access this evidence? Is the result certain, limited, controversial, or unknown? The first ruling of this section is:\n\n> A source alone is not reality.\n\nIts second provision states:\n\n> A claim cannot be verified without defining its subject, scope, time, and burden of evidence.\n\nIts third provision states:\n\n> Different URLs do not mean different independent evidence.\n\nIts fourth provision states:\n\n> The absence of evidence is not automatic proof that the claim is false.\n\nIts fifth provision states:\n\n> Verified reality is not an absolute and unchangeable truth; it is an inferred judgement carried by the available evidence within a specific time and context.\n\nIts sixth provision states:\n\n> Truth Pack cannot be written based on the AI answer. The AI answer is evaluated against the previously locked Truth Pack.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to transform material claims about the audited entity into a reference package that is linked with evidence and counter-evidence, with clear boundaries of time and scope, without converting them into an answer template. The section normalises the following distinctions:\n\n- Official claim versus verified reality\n\n- Source versus evidence\n\n- Evidence versus claim support\n\n- First-party record versus independent verification\n\n- Independent source and source authorised on the subject\n\n- Different domain name and different root of evidence\n\n- Number of sources and independent family of evidence\n\n- Direct evidence and derived evidence\n\n- Supporting evidence and counter evidence\n\n- Absence of claim and falsity of the claim\n\n- Open world claim and closed world record\n\n- Current reality and historical reality\n\n- Time when reality is valid and time when the source was published\n\n- Public evidence and restricted evidence\n\n- Evidence accessible to AI and evidence only open to the auditor\n\n- Proven claim and publicly discoverable claim\n\n- Direct reality and calculated derived result\n\n- Raw data and performance ratio\n\n- Numerator and denominator\n\n- Customer and potential customer\n\n- Partner and customer\n\n- Membership and accreditation\n\n- Licence and brand relationship\n\n- Certificate and overall institutional compliance\n\n- Claim and opinion\n\n- Opinion and comparative advantage\n\n- Complaint and verified violation\n\n- Investigation and definite crime or liability\n\n- Outdated record and incorrect historical record\n\n- Incorrect AI response with missing Truth Pack\n\n- Reference fault and system fault\n\n- Truth Pack integrity and all claims being positive\n\n- Confidentiality of evidence and absence of evidence\n\n- Entity's right to respond and self-verification\n\n- Human approval and human infallibility\n\n- AI assistance and AI being the ultimate authority of truth\n\nAt the end of this section, each audit should be able to answer these questions:\n\n> Which material claims are contained within Truth Pack?\n\n> Which entity, country, product, and time does each claim belong to?\n\n> Which evidence supports it, which limits or refutes it?\n\n> Are the sources truly independent, or are they derived from the same original source?\n\n> Can the claim be publicly verified?\n\n> Is the evidence only available under restricted auditor access?\n\n> Could AI have accessed this evidence at the time of measurement?\n\n> Is the claim certain, partial, historical, controversial, or unknown?\n\n> Was Truth Pack locked before AI responses were seen?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEach GEO-1000 audit must lock the material reference claims about the audited entity, along with supporting evidence, counter-evidence, source lineage, access class, scope, jurisdiction, validity period, and uncertainty status, in a versioned Verified Entity Truth Pack before AI responses are evaluated semantically. Truth Pack:\n\n- the entity's self-written introductory file,\n\n- a sales presentation collecting only positive evidence,\n\n- the auditor's opinion text,\n\n- the ideal response requested from AI,\n\n- the absolute reality about the whole world,\n\n- a reference repository consisting only of a list of URLs\n\nit is not. Truth Pack:\n\n> It is an audit reference package that shows under which evidence material claims in AI responses can be supported, limited, refuted, or left unknown.\n\n## 3. LIMIT OF THE WORD “TRUTH”\n\nThe word “truth” in Truth Pack does not carry the claim: “This package knows all absolute truths about the entity.” The more accurate meaning is:\n\n> A reality record dependent on evidence showing, within the specified date, scope, and body of evidence, to what extent which material judgements are supported.\n\nTruth Pack:\n\n- subject to change with new evidence,\n\n- can be contested,\n\n- its scope can be narrowed,\n\n- it can be corrected if proven wrong,\n\nit can leave some claims unknown. The authority of Truth Pack:\n\n- The word “Truth” does not arise from,\n\n- the name NobleJackal,\n\n- the voice of NOMOS,\n\n- the author of the book\n\nIt arises from the following:\n\n- a clear claim structure,\n\n- appropriate evidence,\n\n- counter-evidence,\n\n- lineage of sources,\n\n- time,\n\n- boundary,\n\n- is meaningful with\n\n- independent review,\n\naccountable human responsibility.\n\n## 4. WHAT IS VERIFIED ENTITY TRUTH PACK?\n\nVerified Entity Truth Pack is a versioned audit package that brings together the material factual records of a specific entity within a specific date and scope, the supporting and opposing evidence related to them, source statuses, and unresolved areas. Candidate representation:\n\nVETP_{E,v,t} = (I, C, O, E⁺, E⁻, L, S, T, U, G)\n\nHere:\n\n- E: audited entity\n\n- v: version of Truth Pack\n\n- t: reference date\n\n- I: entity identity and relationship graph\n\n- C: atomic reference claims\n\n- O: Official Representation Record\n\n- E⁺: supporting evidence\n\n- E⁻: opposing evidence and constraining records\n\n- L: source lineage and independence graph\n\n- S: scope, country, and jurisdiction matrix\n\n- T: time and currency records\n\n- U: unknowns, discrepancies, and reference gaps\n\n- G: governance, version, objection, and approval records\n\nEach component of Truth Pack can be versioned separately. The main package version links the component versions together.\n\n## 5. TRUTH PACK IS NOT A RESPONSE TEMPLATE\n\nTruth Pack is not a text that AI must say word for word. For example, Truth Pack may contain the following reference claim: “Asteron Holdings Ltd. is the owner of the Asteron Travel brand.” AI can correctly phrase it in these forms:\n\n- \"The Asteron Travel brand belongs to Asteron Holdings.\"\n\n- “Asteron Holdings is the owner of the Asteron Travel brand.”\n\n- “Asteron Travel is a brand owned by Asteron Holdings.”\n\nThe material relationship is preserved. The arbiter should not look for word equality. Truth Pack:\n\n#### defines the correct expression field\n\nIt does not impose a single correct sentence.\n\n## 6. TRUTH PACK IS NOT AN ENTITY BIOGRAPHY\n\nThere can be an infinite amount of correct information about an entity. GEO-1000 Truth Pack is not required to collect all information. The package should primarily cover these areas:\n\n- Material duties within the prompt Registry\n\n- The identity and operational core of the entity\n\n- Limits that will change the user's decision\n\n- Licences, legal, price, service, country and time records\n\n- Eligibility and exclusion criteria required for advice\n\n- Material claims carrying the risk of a critical error\n\n- Differences between reality verified by official representation and actual reality\n\nTo this principle:\n\n#### Principle of Reality Proportional to Duty\n\nis called. Collecting unnecessary data:\n\n- privacy,\n\n- cost,\n\n- currency,\n\n- contradiction\n\ncan increase the risk.\n\n## 7. TWELVE CORE COMPONENTS OF THE TRUTH PACK\n\n### 7.1. Canonical Entity Identity\n\nAs defined in Section 3:\n\n- entity identity,\n\n- type,\n\n- legal name,\n\n- trademark and domain name relations,\n\n- parent and subsidiary companies,\n\n- time and jurisdiction limits\n\nare packaged.\n\n### 7.2. Reference Claim Ledger\n\nCarries all material reality claims to be used in the audit in an atomic structure.\n\n### 7.3. Official Representation Record\n\nCarries the current and historical statements published by the entity itself on canonical surfaces.\n\n### 7.4. Evidence Registry\n\nFor each evidence object:\n\n- source,\n\n- integrity,\n\n- authority,\n\n- access,\n\n- scope,\n\n- time\n\nis recorded.\n\n### 7.5. Counter-Evidence Registry\n\nThe claim:\n\n- refuting,\n\n- limiting,\n\n- altering its currency,\n\n- providing another explanation\n\ncontains records.\n\n### 7.6. Source Lineage Graph\n\nOf the sources:\n\n- the first root,\n\n- copies,\n\n- quotations,\n\n- syndication,\n\n- AI derivatives\n\nare shown.\n\n### 7.7. Scope and Jurisdiction Matrix\n\nThe claim:\n\n- for which country,\n\n- product,\n\n- service,\n\n- user,\n\n- legal entity\n\nis valid.\n\n### 7.8. Time and Currency Matrix\n\nThe claim:\n\n- when it started,\n\n- when it ended,\n\n- when it was last verified,\n\n- which event could change it\n\nare shown.\n\n### 7.9. Dispute and Uncertainty Ledger\n\nMakes visible areas that are contradictory, unsolvable, or insufficiently evidenced.\n\n### 7.10. Evidence Access Layer\n\nThe evidence:\n\n- is public,\n\n- audit-accessible,\n\n- restricted,\n\n- confidential,\n\n- derived\n\nas indicated.\n\n### 7.11. Change and Objection Log\n\nCarries new evidence, corrections, counter opinions, and decision changes in a versioned manner.\n\n### 7.12. Machine Manifest\n\nConnects all components:\n\n- identity,\n\n- is meaningful with\n\n- hash,\n\n- relationship\n\nto each other.\n\n## 8. ATOMIC REFERENCE CLAIM\n\nThe basic unit of reality within Truth Pack:\n\n#### is called Atomic Reference Claim.\n\nCandidate structure:\n\nC_j = (s, p, o, q, c, l, v_f, v_t, k, m)\n\nHere:\n\n- s: subject entity\n\n- p: predicate or relation\n\n- o: object or value\n\n- q: qualifiers and limits\n\n- c: country or jurisdiction\n\n- l: language or locale, if material\n\n- vf: validity start\n\n- vt: validity end\n\n- k: assertion type\n\n- m: materiality or risk label\n\nExample: Subject: Asteron Türkiye Tourism Ltd. / Predicate: provides services / Object: corporate travel management / Country: Turkey / Customer type: corporate clients / Start: January 1, 2026 / Last verification: August 18, 2026 This sentence is not atomic enough: “Asteron is a leading company providing numerous services worldwide.” In this single sentence:\n\n- geographical scope,\n\n- number of services,\n\n- leadership,\n\n- company identity\n\nare mixed together.\n\n## 9. CLAIM TYPE\n\nEach atomic claim should carry a type.\n\n### CT-01 — Identity Claim\n\nWho the entity is or which entity it belongs to.\n\n### CT-02 — Relationship Claim\n\nOwnership, control, partnership, clientship, employment or membership relationship.\n\n### CT-03 — Activity and Service Claim\n\nWhat the entity does, what service it provides.\n\n### CT-04 — Product Claim\n\nProduct feature, version, owner, or availability.\n\n### CT-05 — Geographical Scope Claim\n\nCountry, region, or service area.\n\n### CT-06 — Legal and Licensing Claim\n\nRegistration, authority, licence, regulatory status.\n\n### CT-07 — Price and Commercial Term Claim\n\nPrice, tax, subscription, refund, warranty, or payment.\n\n### CT-08 — Quantitative Performance Claim\n\nSuccess rate, delivery time, number of users, growth or efficiency.\n\n### CT-09 — Customer and Partner Claim\n\nCustomer, sponsor, partner, distributor or supplier relationship.\n\n### CT-10 — Certificate and Award Claim\n\nCertificate, membership, award or accreditation.\n\n### CT-11 — Comparative or Superiority Claim\n\nBest, leading, faster, most reliable, number one.\n\n### CT-12 — Recommendation and Suitability Claim\n\nSuitability or recommendation for a specific user.\n\n### CT-13 — Historical Claim\n\nOrganisation, past ownership, former product or historical event.\n\n### CT-14 — Risk, Complaint or Dispute Claim\n\nComplaint, investigation, court, sanction or argument.\n\n### CT-15 — Opinion or Institutional Position\n\nOpinion, vision or value judgement advanced about the entity itself. The type of claim determines which evidence is considered sufficient.\n\n## 10. CLAIM MATERIALITY\n\nNot every correct piece of information within Truth Pack has the same impact on a decision. Materiality can be indicated as a candidate with the following classes:\n\n### CRITICAL-CANDIDATE\n\n### MAJOR-CANDIDATE\n\n### MODERATE\n\n### ADVISORY\n\n### CONTEXTUAL\n\nThe exact error severity level will be determined in Section 15. The label in Truth Pack only:\n\n- manages evidence priority,\n\n- review attention,\n\n- update frequency\n\nmanages.\n\n## 11. EVIDENCE OBJECT\n\nEach source within Truth Pack must be recorded as an Evidence Object, not just as a link. Candidate structure:\n\nE_k = (r, a, i, p, f, s, t, x, g, h)\n\nHere:\n\n- r: source role\n\n- a: authority for claim\n\n- i: independence\n\n- p: proximity to claim\n\n- f: recency\n\n- s: scope\n\n- t: time\n\n- x: access class\n\n- g: source lineage\n\n- h: integrity record\n\nA URL alone does not complete any of these fields.\n\n## 12. EVIDENCE–CLAIM RELATIONSHIP\n\nNot every evidence object supports every claim. The relationship between claim Cj and evidence Ek can be one of the following statuses:\n\n### SUPPORTS\n\n### PARTIALLY_SUPPORTS\n\n### QUALIFIES\n\n### CONTRADICTS\n\n### OUTDATES\n\n### CONTEXTUALISES\n\n### NO_MATERIAL_BEARING\n\n### UNKNOWN_RELATION\n\nExample: A company's trademark registration document: may support brand ownership, does not support customer satisfaction. A customer review: may support the experience of a single user, does not support 98% success across all customers.\n\n## 13. RESOURCE ROLES\n\n### SR-01 — Canonical First-Party Source\n\nThe official surface controlled by the entity. Example:\n\n- Canonical web page\n\n- Official policy\n\n- Pricing page\n\n- Product document\n\n### SR-02 — First-Party Operational Record\n\nRecords at the internal or transaction level of the entity. Example:\n\n- Invoice\n\n- Contract\n\n- Transaction record\n\n- Delivery record\n\n- Raw performance data\n\n### SR-03 — Authorised Official Registration\n\nIt is the registry that carries legal or institutional authority on the subject. Example:\n\n- Company Registry\n\n- Licence Registry\n\n- Certificate issuing institution\n\n- Court decision\n\n- Regulatory registration\n\n### SR-04 — Two-Way Verification\n\nIt is the support of the relationship by both parties or by contract. Example:\n\n- Customer relationship\n\n- Partnership\n\n- Sponsorship\n\n- Distributorship\n\n### SR-05 — Independent Editorial Source\n\nIt is a publication or research editorially independent from the entity. Independence is not a guarantee of accuracy.\n\n### SR-06 — Independent Audit or Research\n\nIt is an open external evaluation of its method and scope.\n\n### SR-07 — Transactional or Behavioural Evidence\n\nRecords showing actual transactions or user behaviour.\n\n### SR-08 — User-Generated Evidence\n\nComment, complaint, review, or user document. It is an individual experience.\n\n### SR-09 — Derived Calculation\n\nRate, classification, or summary produced from raw records. Its formula and inputs must exist.\n\n### SR-10 — Synthetic Evidence\n\nThis data has been created for method demonstration purposes. It cannot prove the claim of a real entity.\n\n### SR-11 — AI-Generated Derivative Source\n\nContent created by an AI system by summarising other sources. It does not replace the original sources.\n\n### SR-12 — Unknown or Unresolvable Source\n\nThe source has not been verified or its origin cannot be determined. It carries limited trust in primary material verification.\n\n## 14. THE AUTHORITY OF THE SOURCE DEPENDS ON THE TYPE OF CLAIM\n\nThere is no universal and single ranking for all sources. Example: For the current public price of the company:\n\n- canonical price page,\n\n- actual checkout,\n\n- current offer\n\ncan be a strong source. For the company's claim: “We are the world leader.”, the same web page only shows a self-declaration. It does not prove leadership. For the licence: the institution granting the licence is more authoritative than the company's blog. For customer experience: a single customer review is a real, singular event and does not represent the entire user population. Therefore:\n\n> The quality of the source is not general; the claim is specific.\n\n## 15. SEVEN DIMENSIONS OF SOURCE PROFILE\n\nEach source should be evaluated along these dimensions:\n\n### 15.1. Authority\n\nIs the source authorised to make a judgement about the type of claim?\n\n### 15.2. Proximity\n\nHow close is the source to the event, process, or record?\n\n### 15.3. Independence\n\nHow independent is the source from the commercial or institutional owner of the claim?\n\n### 15.4. Integrity\n\nIs the authenticity of the file or record and its change history preserved?\n\n### 15.5. Timeliness\n\nIs the source valid for the relevant time?\n\n### 15.6. Scope\n\nDoes the source support the full scope of the claim, or only a part of it?\n\n### 15.7. Reproducibility\n\nCan another auditor reach the same judgement from the same record? These dimensions should not necessarily be reduced to a single \"source score.\" A source:\n\n- highly authoritative,\n\n- low independence\n\npossible. Another source:\n\n- highly independent,\n\n- distant from the subject\n\nmay be.\n\n## 16. THE CORRECT ROLE OF THE FIRST-PARTY SOURCE\n\nFirst-party source: it is not automatically worthless, nor is it automatically independent evidence. First-party sources can be strong for claims such as:\n\n- Price published by the institution\n\n- Support hours\n\n- Description of its own product and service\n\n- Text of its own policy\n\n- Official statements disclosed to the public\n\n- Its own transaction data\n\nIt may not be sufficient alone for claims such as:\n\n- World leadership\n\n- Highest satisfaction\n\n- Superiority over competitors\n\n- Universal success\n\n- Independent accreditation\n\n- Business outcomes for all customers\n\nIn the previous '99 Errors' text: 'Your own site is a claim; not proof.' warning is of central importance for claims that particularly require superiority and independence. A more complete normative statement is as follows:\n\n> Your own site is proof of what the organisation says; it is not automatic proof of the independent correctness of any claim regarding the outside world.\n\n## 17. WHAT IS INDEPENDENCE?\n\nHaving a resource on different domain names does not make it independent. Independence is related to the following areas: Who produced the content? Who decided to publish it? Is there a payment or sponsorship? Did the source get text directly from the company? Did the publication perform its own verification? Is the content copied from another source? Does the source have ownership or business relations with the company? Ten pieces of content published by an agency on different domain names are not ten independent sources.\n\n## 18. SOURCE LINEAGE\n\nEvidence objects should be linked to a source lineage. Candidate graph:\n\n### G_S = (V_S, E_S) [K10]\n\nHere:\n\n- V_S: source objects\n\n- E_S: citation, copying, syndication, summarisation, and derivation relationships\n\nSource relationships:\n\n### ORIGINAL_SOURCE\n\n### QUOTES\n\n### SYNDICATES\n\n### SUMMARISES\n\n### TRANSLATES\n\n### DERIVES_FROM\n\n### AI_GENERATED_FROM\n\n### UNKNOWN_LINEAGE\n\ncan be recorded as.\n\n## 19. INDEPENDENT EVIDENCE FAMILY\n\nContents originating from the same root source can be considered as a single Evidence Family. Example:\n\n- Company press release\n\n- News site that publishes the press release verbatim\n\n- Blog summarising the news site\n\n- AI text summarising the blog\n\nThere are four URLs. There may not even be a single independent evidence family. The number of sources and the number of evidence families should be reported together.\n\n## 20. EVIDENCE LAUNDERING\n\nRepublishing a self-statement on other domains and then using it as independent evidence:\n\n#### Is called Evidence Laundering\n\nExample chain: The company says \"we are leaders\" on its own site. Paid content publishes this as \"news.\" Other sites copy the news. Since AI has seen many domains, it accepts the claim as widespread. The company uses the AI's response as evidence of its own claim. This closed loop:\n\n#### self-statement → derived source → AI response → republication of self-statement\n\ncreates. It should make this loop visible by establishing the Truth Pack source lineage.\n\n## 21. LIMITATION OF AI-GENERATED SOURCE\n\nAn AI system:\n\n- can summarise sources,\n\n- can compare claims,\n\ncan generate new text. AI text:\n\n- is not an original source,\n\n- cannot replace the underlying evidence,\n\ndoes not become independent verification just because another AI says the same thing. AI output can be used for the purposes of:\n\n- directing research,\n\n- claim extraction,\n\n- conflict detection,\n\nassistance in finding sources. The ultimate evidence object should be connected to the original source as much as possible.\n\n## 22. LIMITATION OF SYNTHETIC EVIDENCE\n\nSynthetic data:\n\n- method demonstration,\n\n- system testing,\n\n- software verification,\n\n- adjudicator training\n\ncan be used for. Synthetic data:\n\n- the actual number of customers,\n\n- the actual performance rate,\n\n- the actual country coverage,\n\n- the actual licence\n\ncannot prove. Even if synthetic records are in the same file as real evidence, they must be clearly labelled.\n\n## 23. SUPPORTING EVIDENCE IS INSUFFICIENT\n\nTruth Pack should only collect sources that support the claim. For material claims, the following records should be actively sought:\n\n- Conflicting date\n\n- Different scope\n\n- Expired document\n\n- Failed transaction\n\n- Complaint or return\n\n- Previous ownership\n\n- Licence belonging to another entity\n\n- Correction or withdrawal\n\n- Comparative data\n\n- Missing denominator\n\nThis obligation is called:\n\n#### Burden of counter-evidence\n\nIt applies whenever material counter-evidence exists.\n\n## 24. COUNTER-EVIDENCE\n\nCounter-evidence does not always completely refute the claim. It can serve one of the following functions:\n\n### CONTRADICTS\n\n### LIMITS_SCOPE\n\n### LIMITS_TIME\n\n### IDENTIFIES_EXCEPTION\n\n### CHALLENGES_METHOD\n\n### QUESTIONS_AUTHENTICITY\n\n### PROVIDES_ALTERNATIVE_ENTITY\n\n### REQUIRES_ADDITIONAL_REVIEW\n\nExample: A negative review of the claim “All customers are satisfied” may not automatically refute the entire claim. However, it can show that the term “all” is incorrect.\n\n## 25. STORAGE OF COUNTER-EVIDENCE\n\nRecord that weakens the claim:\n\n- not in favour of the company,\n\n- appears low quality,\n\n- cannot be quietly deleted because its disclosure to the public would cause discomfort\n\nLow-confidence counter-evidence can also be recorded. Its status is clearly indicated. The adjudicator gives it appropriate weight.\n\n## 26. CLAIM SUPPORT STATUSES\n\n### VS-0 — NOT ASSESSED\n\nThe claim has not yet been evaluated.\n\n### VS-1 — OFFICIAL CLAIM ONLY\n\nThe claim is based only on an official self-declaration. It has not been proven false. No additional verification exists.\n\n### VS-2 — TRACEABLE FIRST-PARTY SUPPORT\n\nThe claim is supported by a traceable first-party operation or transaction record.\n\n### VS-3 — CORROBORATED\n\nThe claim is supported by multiple appropriate pieces of evidence. Not all sources need to be independent.\n\n### VS-4 — INDEPENDENTLY VERIFIED\n\nThe claim has been verified with an independent or authoritative external source appropriate to the topic.\n\n### VS-5 — MULTI-SOURCE SUSTAINED\n\nThe claim is strongly supported across different types and families of sources, in scope and over time.\n\n### VS-C — CONTRADICTED\n\nStronger or more appropriate evidence contradicts the claim.\n\n### VS-U — UNRESOLVED\n\nThe evidence is insufficient, contradictory, or at a level where no decision can be made.\n\n### VS-W — WITHDRAWN\n\nThe claimant has withdrawn the claim. The historical record is preserved.\n\n### VS-R — RESTRICTEDLY VERIFIED\n\nThe claim is supported by evidence restricted to auditor access. Public verification is not possible. This status also affects whether it can be expected that AI produces the claim from public sources.\n\n## 27. SUPPORT STATUS ALONE IS NOT SUFFICIENT\n\nA claim at the same time:\n\n- supported,\n\n- only for a specific country,\n\n- only for a specific date range,\n\n- with evidence not publicly available,\n\n- under opposing views\n\nIt is possible. Therefore, instead of a single status, a multidimensional Reality Status Vector can be used:\n\nV_j = (support, scope, time, conflict, access)\n\nExample:\n\n### VS-4 + SCOPE-LIMITED + CURRENT + NO-CONFLICT + PUBLIC\n\nis another result.\n\n### VS-R + GLOBAL-CLAIM + TIME-LIMITED + DISPUTED + RESTRICTED\n\nis another result.\n\n## 28. SCOPE STATUSES\n\n### SC-0 — SCOPE UNKNOWN\n\nThe scope to which the claim applies cannot be determined.\n\n### SC-1 — EXACT SCOPE\n\nThe claim and evidence carry the same entity, product, country, and user scope.\n\n### SC-2 — PARTIAL SCOPE\n\nEvidence supports only part of the claim.\n\n### SC-3 — OVERBROAD CLAIM\n\nThe claim is broader than what the evidence supports.\n\n### SC-4 — WRONG ENTITY SCOPE\n\nThe evidence belongs to another entity or affiliate.\n\n### SC-5 — WRONG JURISDICTION\n\nThe evidence belongs to another country or legal area.\n\n### SC-6 — WRONG PRODUCT OR SERVICE\n\nThe evidence belongs to a different product or service.\n\n## 29. TIME STATUTES\n\n### TS-0 — TIME UNKNOWN\n\nThe validity period of the claim is unknown.\n\n### TS-1 — CURRENT\n\nIt is supported that it is valid on the reference date.\n\n### TS-2 — HISTORICAL\n\nIt is supported for a specific past period.\n\n### TS-3 — EXPIRED\n\nThe validity of the document or claim has expired.\n\n### TS-4 — SUPERSEDED\n\nIt has been replaced by a newer record.\n\n### TS-5 — EVENT-DEPENDENT\n\nIt is valid as long as an event or condition continues.\n\n### TS-6 — FUTURE OR PLANNED\n\nIt is a plan or commitment that has not yet occurred. It cannot be presented as current reality.\n\n## 30. FOURTH TIMESTAMP\n\nEach piece of evidence and claim should carry four distinct times to the extent that it is relevant:\n\n### 30.1. Time When Reality Is Valid\n\nWhen was a claim ever true in the world?\n\n### 30.2. Time When the Source Was Published\n\nWhen was the information presented to the public or the auditor?\n\n### 30.3. The Time When the Source is Captured\n\nWhen was the audit evidence recorded?\n\n### 30.4. Time of Adjudication of the Claim\n\nWhen was the status of reality given? This distinction is important. A licence may have been cancelled at 10:00. The record may have been updated at 15:00. The AI response may have been given at 12:00. Response:\n\n- old in terms of world reality,\n\n- understandable in terms of publicly accessible evidence\n\nTruth Pack should distinguish between the two cases.\n\n## 31. PUBLIC ACCESSIBILITY OF EVIDENCE\n\nEvidence access classes:\n\n### EA-1 — PUBLIC\n\nPublicly available and auditable.\n\n### EA-2 — PUBLIC BUT TECHNICALLY LIMITED\n\nPublic but:\n\n- access,\n\n- geography,\n\n- format,\n\n- technical access\n\nhas limitations.\n\n### EA-3 — AUDIT ACCESS\n\nCan be made available in a controlled manner to independent auditors.\n\n### EA-4 — RESTRICTED CONFIDENTIAL\n\nThe contract is restricted due to personal data or trade secrets.\n\n### EA-5 — DERIVED PUBLIC SUMMARY\n\nRaw evidence is confidential. Its method and summary are public.\n\n### EA-6 — UNAVAILABLE OR LOST\n\nIt is claimed that the evidence exists, but it could not be accessed during the audit.\n\n## 32. PUBLICLY AVAILABLE AUTHENTICITY AND RESTRICTED AUTHENTICITY\n\nA company can confirm the claim “X is our customer.” with a private contract. The contract may not be disclosed to the public. Truth Pack can record the claim as: VS-R — Restrictedly Verified. However, it cannot be expected that a public AI product knows this relationship. Therefore, there are two separate questions: Is the claim true? Is the claim publicly discoverable? A claim may be true with private evidence. For public GEO success, again:\n\n- is public,\n\n- it may need to be verified,\n\n- appropriate for the user\n\na record may be required.\n\n## 33. PUBLIC DISCOVERABILITY\n\nFor each material claim, the following field may be recorded as a candidate:\n\n### PUBLICLY_DISCOVERABLE\n\n### PUBLIC_BUT_OBSCURE\n\n### PUBLICLY_CONFLICTED\n\n### RESTRICTED_ONLY\n\n### NOT_PUBLIC\n\n### UNKNOWN\n\nThe reality of the AI response and access to information can be evaluated separately afterwards.\n\n## 34. TRUE BUT NOT PUBLIC CLAIM\n\nIf an AI product correctly states information that is not public, the following questions arise: Is the information available in another public source? Did the AI use personalisation or a private knowledge base? Is the answer correct by chance? Is there a leak of confidential or secret information? Was the claim verified only after the fact? Truth Pack should record the source path separately, without counting being correct alone as an achievement.\n\n## 35. OPEN WORLD AND CLOSED WORLD CLAIMS\n\nThe absence of information in a record depends on the type of claim.\n\n### 35.1. Open World Assumption\n\nThe entirety of the recorded universe is unknown. Example:\n\n- All customers\n\n- Business relations in all countries\n\n- All user complaints\n\n- All media publications\n\nThe absence of a record does not prove that the assertion is false.\n\n### 35.2. Closed World Assumption\n\nIt can be assumed that the authorised and comprehensive registry fully covers the relevant universe. Example:\n\n- List of active licences on a specific date\n\n- Official company registration status\n\n- Valid certificates of a specific certifying institution\n\n- If the published product catalogue is truly complementary\n\nIn this case, the absence of a record could be a stronger negative evidence.\n\n## 36. CLOSED WORLD AUTHORITY STATEMENT\n\nIf a source is to be used as a closed world record, the following fields must be defined: Which universe does it cover? For which date? What is the update frequency? Could the record be missing? Is there an archive and change log? Who is the authorised institution? Simply being the “official website” does not provide closed world authority.\n\n## 37. DISTINCTION BETWEEN NOT FOUND EVIDENCE AND ERROR\n\nThe following three sentences are different:\n\n- \"No evidence was found for the claim.\"\n\n- \"The claim is not supported by the available evidence.\"\n\n- “The claim contradicts stronger evidence.”\n\nThe first may be the result of research. The second is the status of verification. The third is a material contradiction. These are not the same:\n\n### NOT FOUND\n\n### UNVERIFIED\n\n### CONTRADICTED\n\n## 38. ABSENCE OF EVIDENCE\n\nTo establish the falsity of a claim based on absence, the following questions must be answered: Is the record universe being searched comprehensive? Were the correct entities and name variants searched? Were the correct date and country used? Could the record be updated with a delay? Is another legal entity or trademark being used? Is there an access issue? Without these conditions: a definitive ruling such as 'Not found in the registry, therefore no licence exists.' cannot be made. A more accurate result could be: 'Could not be verified in the specified registry and date.'\n\n## 39. CLAIM–EVIDENCE MATRIX\n\nThis table is not a universal and unchangeable source ranking. It shows which type of claim carries which burden of proof.\n\n## 40. QUANTITATIVE CLAIM\n\nA quantitative claim cannot be verified without the following fields:\n\n- Measured item\n\n- Numerator\n\n- Denominator\n\n- Period\n\n- Inclusion rule\n\n- Exclusion rule\n\n- Data source\n\n- Missing records\n\n- Unique unit\n\n- Calculation formula\n\n- Rounding\n\n- Update date\n\n- Responsible person\n\nExample: “98 per cent success.” alone is not a quantitative claim. It is an undefined marketing number.\n\n## 41. NUMERATOR AND DENOMINATOR\n\nSuccess rate:\n\n### R = S/N\n\nlet it be. Here:\n\n- S: units considered successful\n\n- N: all eligible units evaluated\n\nmust be. If failed or incomplete projects are removed from the denominator, the ratio changes. The denominator cannot be selected after the result.\n\n## 42. UNIQUE UNIT\n\nThe following expressions are different from each other:\n\n- 100 customers\n\n- 100 projects\n\n- 100 invoices\n\n- 100 contracts\n\n- 100 user accounts\n\n- 100 transactions\n\nThe same customer may have done ten projects. “100 projects” is not “100 customers.” Truth Pack must explicitly record the unit of the number.\n\n## 43. CUMULATIVE AND ACTIVE NUMBER\n\n\"We have 500 customers.\" may carry one of the following meanings:\n\n- 500 unique customers in the company's history\n\n- 500 customers in the last year\n\n- 500 active customers currently\n\n- 500 contact records\n\n- 500 accounts\n\n- 500 projects\n\nThese meanings are not the same.\n\n## 44. DERIVED CLAIM\n\nA claim may not be directly written in the document. It can be calculated from raw data. Example: “The average delivery time is 12 days.” This result:\n\n- which projects are included,\n\n- the definition of start and end,\n\n- failed projects,\n\n- median and extreme values\n\nrequires. The derived claim must include the following areas:\n\n- Input data identifiers\n\n- Formula\n\n- Code or calculation method\n\n- Version\n\n- Reproduction record\n\n## 45. PERFORMANCE CLAIM\n\nThe performance claim should answer these questions: Whose performance is the result? Which product or service? Which user or project type? What is the baseline? What is the success criterion? What is the period? How many unsuccessful results are there? Is it a selected case or the entire population? Is there independent verification? Is the result causal or merely relational? A case study: can show a possible result. Does not show the expected result in all customers.\n\n## 46. WARRANTY CLAIM\n\nThe statement “Guaranteed results.” requires the following areas: Exactly what is guaranteed? Delivery or business outcome? What is the timeframe? What are the exceptions? What is the refund or compensation? Are there third parties outside of control? Is AI advice or ranking guaranteed? A company:\n\n- can guarantee its own workflow,\n\n- specific delivery\n\nIt cannot absolutely guarantee the future advisory behaviour of an AI product it does not control.\n\n## 47. CUSTOMER CLAIM\n\nDisplaying the logo of an organisation on a website does not automatically prove a customer relationship. The logo may represent one of the following relationships:\n\n- Customer\n\n- Former customer\n\n- Potential customer\n\n- Event participant\n\n- Sponsor\n\n- Partner\n\n- Supplier\n\n- Technology user\n\n- Group company\n\n- Unauthorised logo usage\n\nTruth Pack must clearly define the type of relationship.\n\n## 48. CUSTOMER CONFIDENTIALITY\n\nThe customer relationship may be verified by contract but may not be disclosed publicly. Correct record: VS-R — Restrictedly Verified may be. If the company cannot publicly say, \"X is our customer,\" the AI is not expected to know this relationship. The number of confidential customers cannot be silently converted to public GEO authority.\n\n## 49. PARTNERSHIP CLAIM\n\nThe word \"Partner\" alone is not sufficient. The relationship must include the following areas:\n\n- Type of partnership\n\n- Product or project\n\n- Country\n\n- Start and end\n\n- Mutual approval\n\n- Commercial or technical scope\n\n- Authority and responsibility\n\nA software subscription may not be a technology partnership. An event sponsorship may not be a long-term strategic partnership.\n\n## 50. LICENSE AND AUTHORITY CLAIM\n\nA licence claim must at least include the following areas:\n\n- Full entity of the licence holder\n\n- Institution granting the licence\n\n- Licence number\n\n- Country or jurisdiction\n\n- Scope of authority\n\n- Start and end date\n\n- Suspension or cancellation status\n\n- Verification date\n\nThe licence of another group company cannot be transferred to the brand. An employee's personal licence does not automatically grant authority to the institution.\n\n## 51. CERTIFICATE, MEMBERSHIP AND ACCREDITATION\n\nThe following concepts are different:\n\n- Certificate\n\n- Membership\n\n- Accreditation\n\n- Award\n\n- Training participation\n\n- Conformity mark\n\n- Self-assessment badge\n\nMembership in a professional organisation: may not mean expertise accredited by that organisation. Completing a training: may not be independent certification. The term Truth Pack should be used according to the definition given by the issuing institution.\n\n## 52. AWARD CLAIM\n\nThe award record must contain the following fields:\n\n- Awarding institution\n\n- Year\n\n- Category\n\n- Geography\n\n- Application or fee structure\n\n- Evaluation method\n\n- Sponsor relationship\n\n- Winning entity\n\n- Product or company scope\n\nPaid application award is not automatically fake. It is also not independent market leadership. Source status must be preserved.\n\n## 53. CLAIM OF SUPERIORITY\n\nThe expressions “Best.” “World leader.” “Number one.” “Most reliable.” require the following areas:\n\n- Comparison universe\n\n- Competitor set\n\n- Criteria\n\n- Weight\n\n- Data\n\n- Period\n\n- Geography\n\n- Method\n\n- Independence\n\n- Uncertainty\n\nIf these fields are missing, the correct Truth Pack status is usually:\n\n### VS-1 — Official Claim Only\n\nor: VS-U — Unresolved.\n\n## 54. OPINION AND FACT DISTINCTION\n\nThis sentence: “We believe in human-centreed design.” is a corporate opinion or value statement. Market research is not needed to prove this. This sentence is different: “Our products are the best in the industry for human-centreed design.” It becomes a comparative factual claim. Truth Pack must preserve the epistemic type of each claim.\n\n## 55. COMPLAINT AND NEGATIVE USER RECORD\n\nA user complaint:\n\n- may be a real singular event,\n\n- It may be incomplete or incorrect,\n\n- It may have been resolved,\n\nIt may belong to another entity. Single complaint: “The company victimises all customers.” does not support the result. However, a serious incident:\n\n- Investigation,\n\n- correction,\n\n- Critical risk\n\nmay be initiated.\n\n## 56. CLAIM, INVESTIGATION AND JUDGEMENT\n\nThese statuses should be distinguished from each other:\n\n- Claim\n\n- Complaint\n\n- Open investigation\n\n- Administrative finding\n\n- First instance decision\n\n- Finalised decision\n\n- Withdrawn or overturned decision\n\n- Response of the institution\n\nHaving an investigation about a person or institution does not mean they are guilty or responsible. Truth Pack should not use language that exceeds the legal status.\n\n## 57. PROPORTIONAL LANGUAGE IN NEGATIVE ALLEGATIONS\n\nTruth Pack should use the following language:\n\n- \"It has been alleged.\"\n\n- \"It is being investigated by the specified institution.\"\n\n- \"The specified decision was made on the stated date.\"\n\n- \"The decision has not been finalised.\"\n\n- \"The institution denies the allegation.\"\n\n- \"It has not been resolved in the existing records.\"\n\nThe following language cannot be used without sufficient definite records:\n\n- \"Is a fraudster.\"\n\n- \"It is illegal.\"\n\n- \"He/She is guilty.\"\n\n- It is definitely unreliable.\n\nThis is not just a legal risk, it is the principle of accuracy in representation.\n\n## 58. RECOMMENDATION REALITY\n\nTruth Pack should not reduce a company overall to a single status of: 'recommended' or: 'not recommended.' The recommendation depends on the following areas:\n\n- User needs\n\n- Country\n\n- Budget\n\n- Type of service\n\n- Risk tolerance\n\n- Capacity\n\n- Licence\n\n- Exclusion conditions\n\n- Alternatives\n\nTruth Pack for recommendation:\n\n- eligibility requirements,\n\n- conditions of unsuitability,\n\n- proven limits\n\nprovides. Final user advice is the decision duty of the AI product.\n\n## 59. LIMIT CLAIMS\n\nTruth Pack should carry not only what it does but also what it does not do. Example: It operates only in Turkey. It does not provide legal advice. It does not have a health licence. Prices are exclusive of tax. Service is open only to corporate clients. It does not guarantee results. It does not control the behaviour of certain third parties. Boundary information is one of the most valuable records for accurate advice.\n\n## 60. UNKNOWN LOGBOOK\n\nTruth Pack should not hide missing fields. Types of unknown records:\n\n### UNKNOWN_FACT\n\n### UNKNOWN_SCOPE\n\n### UNKNOWN_TIME\n\n### UNKNOWN_ENTITY\n\n### UNKNOWN_SOURCE_LINEAGE\n\n### UNKNOWN_PUBLIC_AVAILABILITY\n\n### INSUFFICIENT_EVIDENCE\n\n### REFERENCE_CONFLICT\n\n### NOT_APPLICABLE\n\n### NOT YET ASSESSED\n\nUNKNOWN is not a failure. It is a protection against false certainty.\n\n## 61. REFERENCE GAP\n\nAn AI response may generate a material claim not found in Truth Pack during review. In this case, the adjudicator should use one of the following options:\n\n- The claim is out of scope and not material\n\n- There is a reference gap in Truth Pack\n\n- New evidence research required\n\n- Response cannot be supported\n\n- Claim belongs to a false entity\n\nThe Truth Pack gap cannot be automatically converted to AI failure. Correct status:\n\n### REFERENCE_GAP\n\nmay be.\n\n## 62. TRUTH PACK EXPANSION AFTER THE RESPONSE\n\nThe AI response may produce a previously unconsidered material claim. The new claim can be investigated. However, the following rules apply: The first Truth Pack version is preserved. The new claim and evidence are added in a separate version. The initial review result is not quietly changed. If necessary, all related responses are re-evaluated with the new version. Old and new results are visible in the public change log.\n\n## 63. PRE-WAVE LOCK OF THE TRUTH PACK\n\nBefore each main measurement wave, the following fields must be locked:\n\n- Entity identity version\n\n- Official Representation Record version\n\n- Reference Claim Ledger\n\n- Evidence Register\n\n- Counter-Evidence Register\n\n- Source Lineage Graph\n\n- Scope and time matrix\n\n- Unknowns Ledger\n\n- Evidence access classes\n\n- Package hash\n\n- Human approval\n\n- Lock time\n\nTo this file:\n\n#### NOMOS Truth Pack Lock\n\ncan be given the name.\n\n## 64. TRUTH PACK AND MEASUREMENT DATE\n\nAI response, as a rule:\n\n#### the state of reality valid at the moment the prompt is sent\n\nshould be evaluated accordingly. Truth Pack carries the time frame that will represent this moment. The reality that forms later: does not automatically make the past answer wrong, it creates a new Truth Pack version for the new wave.\n\n## 65. REALITY CHANGE DURING THE WAVE\n\nCompany:\n\n- changes its name,\n\n- obtains or loses a licence,\n\n- updates its price,\n\n- expands to another country,\n\n- discontinues its product\n\nthen Truth Pack can change materially. In this case:\n\n- the measurement wave can be divided into subwaves,\n\n- the old and new Truth Pack versions can be linked to separate responses,\n\nThe MIXED_REFERENCE_STATE warning can be used.\n\n## 66. RETROACTIVE MODIFICATION OF THE TRUTH PACK\n\nIf a record is incorrect, it can be corrected. However:\n\n- the old version cannot be deleted,\n\n- the reason for the change must be recorded,\n\n- which responses are affected must be shown,\n\nthe new score must create a separate analysis version. Correcting reality is not the same as rewriting history.\n\n## 67. VERSION ARCHITECTURE\n\nCandidate version format: Major.Minor.Patch\n\n#### Major\n\nTarget entity change, Claim ontology change, Evidence standard change, Scope change that will lead to the evaluation of the same responses in different reality systems\n\n#### Minor\n\nNew material claim Addition of new country or product Status change with new evidence Historical or current reality change\n\n#### Patch\n\nSpelling Metadata Broken link Correction of hash or technical record Material reality change cannot be stored as a patch.\n\n## 68. CURRENCY AND FRESHNESS\n\nNot every claim ages at the same rate.\n\n### FRESH-1 — Structural or Historical\n\nFields with low variability such as foundation date. Event-triggered revalidation may be sufficient.\n\n### FRESH-2 — Event Dependent\n\nOwnership, administrator, or licence status. Change event and regular checks are required.\n\n### FRESH-3 — Medium Variability\n\nService country, number of employees, partnership. Periodic verification is required.\n\n### FRESH-4 — High Variability\n\nPrice, stock, plan, product availability. Should be verified at short intervals.\n\n### FRESH-5 — Live or Instant\n\nOutage, current offer, event, real-time status. Records close to the measurement moment are required. Exact durations should be determined according to claim type and industry risk.\n\n## 69. LAST VERIFICATION DATE\n\nEach claim:\n\n- last verification date,\n\n- recommended re-verification date,\n\n- change trigger\n\nmust carry. Claim without a last verification date: cannot be assumed current.\n\n## 70. SOURCE SNAPSHOT\n\nA dynamic web page should not be saved only as a URL. As much as it is relevant: page snapshot,\n\n### PDF,\n\nfull text, screenshot, capture time, hash,\n\n### URL,\n\nlocale, access status must be preserved. The page may change later.\n\n## 71. IF THE SOURCE IS LOST\n\nA source can be deleted or become inaccessible. Truth Pack:\n\n- may preserve the archive snapshot,\n\n- hash,\n\n- capture time\n\ncan be preserved. But archive:\n\n- that the page is publicly accessible in real-time,\n\n- may not by itself fully prove the scope of access.\n\nNone of these facts is fully established by the archive alone.\n\n## 72. EVIDENCE ACCESS AND PRIVACY LAYERS\n\n#### Public Evidence\n\nEvidence that is publicly accessible.\n\n#### Audit Evidence\n\nEdited evidence accessible to independent adjudicators and auditors.\n\n#### Restricted Evidence\n\nRestricted evidence containing contracts, personal data, or trade secrets.\n\n#### Derived Evidence\n\nVerifiable summary or calculation produced from raw evidence.\n\n#### Public Claim Manifest\n\nFor every claim to the public:\n\n- its status,\n\n- source class,\n\n- date,\n\n- proof hash,\n\n- access status\n\nmanifest showing.\n\n## 73. USE OF CONFIDENTIAL EVIDENCE\n\nIf restricted evidence is to be used: who saw the evidence? what did they verify? how much can be disclosed to the public? can an independent second auditor access it? what is the duration and scope of the evidence? can the evidence be withdrawn? it must be recorded. The statement 'We have confidential evidence' alone does not constitute Truth Pack support.\n\n## 74. ROLE OF THE AUDITED ENTITY\n\nAudited entity:\n\n- can provide evidence,\n\n- can object to incorrect identity records,\n\n- can explain the scope,\n\n- can offer controlled access for confidential evidence,\n\ncan present a counter-argument. Audited entity:\n\n- cannot grant final supporting status to its own claim,\n\n- cannot have counter-evidence deleted,\n\n- cannot select the judge alone,\n\n- cannot remove a low-scoring claim from the scope,\n\ncannot escape the entire evidentiary burden by calling it a 'trade secret'.\n\n## 75. RIGHT TO RESPOND\n\nBefore a publicly available material and adverse finding is published, to the extent applicable, the audited entity is entitled to:\n\n- the relevant claim,\n\n- the evidence used,\n\n- counter-evidence,\n\n- proposed status,\n\n- response time\n\nmay be reported. The entity's failure to respond does not mean that it accepts the claim. The response:\n\n- as evidence,\n\n- as a counter-opinion,\n\n- as a new verification request\n\nis recorded.\n\n## 76. APPEAL STATUSES\n\n### NO_APPEAL\n\n### APPEAL_OPEN\n\n### EVIDENCE_SUBMITTED\n\n### UNDER_REVIEW\n\n### APPEAL_UPHELD\n\n### APPEAL_PARTIALLY_UPHELD\n\n### APPEAL_REJECTED\n\n### EXTERNAL_REVIEW_REQUESTED\n\n### CLOSED_WITH_DISSENT\n\nThe main record is not deleted during the objection. Its status remains visible.\n\n## 77. DISSENTING OPINION\n\nAdjudicators or experts may draw different conclusions from the same evidence. If the material dispute is not resolved:\n\n- majority decision,\n\n- dissenting opinion,\n\n- reasons,\n\n- evidence difference\n\ncan be preserved in the package. The truth package should not produce a fake consensus.\n\n## 78. TRUTH PACK GOVERNANCE ROLES\n\nThe following roles can be defined to the extent relevant:\n\n- Entity Record Owner\n\n- Claim Ledger Editor\n\n- Evidence Registrar\n\n- Source Lineage Analyst\n\n- Domain Specialist\n\n- Legal or Regulatory Reviewer\n\n- Quantitative Methods Reviewer\n\n- Counter-Evidence Reviewer\n\n- Privacy Custodian\n\n- Truth Pack Adjudicator\n\n- Appeal Reviewer\n\n- Version Custodian\n\n- Accountable Human Approver\n\nIn a small pilot, one person can carry multiple roles. Conflicts of interest must be disclosed.\n\n## 79. THE ROLE OF AI IN THE TRUTH PACK\n\nAI systems can assist in the following tasks:\n\n- Extracting claims from documents\n\n- Finding the same claim in different sentences\n\n- Estimating the source lineage\n\n- Mark contradiction candidates\n\n- Remove date and scope fields\n\n- Propose evidence–claim matching\n\n- Detect missing fields\n\n- Create machine diagram\n\nAI:\n\n- cannot give final support status alone,\n\n- cannot make its own generated summary new evidence,\n\n- cannot be the sole and infallible authority in contradictory reality,\n\ncannot replace human approval.\n\n## 80. TRUTH PACK PREPARATION LEVELS\n\n### TPR-0 — NO PACK\n\nStructured Truth Pack not found. Main adjudication cannot start.\n\n### TPR-1 — OFFICIAL-ONLY DRAFT\n\nThere are only official representations and self-declarations. Can be used for discovery. It is not a verified Truth Pack.\n\n### TPR-2 — STRUCTURED EVIDENCE PACK\n\nContains atomic claims and basic evidence. Counter-evidence and source lineage may be missing.\n\n### TPR-3 — ADJUDICATION-READY\n\nClaim, evidence, counter-evidence, scope, time, and unknowns are ready for adjudication. The candidate minimum level for main GEO-1000.\n\n### TPR-4 — INDEPENDENTLY REVIEWED\n\nThe package has undergone independent field and method review.\n\n### TPR-5 — REPLICATED AND MAINTAINED\n\nThe package has been used in different audits, and the version and currency system has been run. The main public audit at least:\n\n### TPR-3\n\nshould target that level.\n\n## 81. TRUTH PACK PREPARATION GATES\n\nA package must pass the following gates before being considered TPR-3: Has the entity identity been resolved? Are the necessary claim fields available for prompt families? Are material claims atomic? Is the timing and scope of each claim clear? Do proof objects carry integrity records? Has the source lineage been examined? Has counter-evidence been sought? Are public and restricted access separated? Are unknown fields visible? Is there human approval and version lock?\n\n## 82. MATCHING THE CLAIM SCOPE WITH THE TRUTH PACK\n\nEach claim family must be linked to the required Truth Pack fields.\n\nIf Truth Pack does not cover the area required by the prompt, the reference is not ready.\n\n## 83. SYNTHETIC APPLE.COM TRUTH PACK DISPLAY\n\nSYNTHETIC METHODOLOGY DISPLAY / The claims, records, and statuses below are fabricated solely to demonstrate the method. They do not reflect real Apple Inc. information, current company structure, or an actual audit result. Canonical target: the synthetic main corporate technology entity associated with the apple.com domain.\n\n### 83.1. Synthetic Claim C-001\n\n> Claim:\n\napple.com is one of the canonical digital surfaces controlled by the synthetic main corporate entity. Type: CT-02 — Relationship / Evidence:\n\n- Synthetic domain name registration\n\n- Synthetic official company registration\n\n- Synthetic independent verification\n\n> Status:\n\n### VS-4 + SC-1 + TS-1 + EA-1\n\n### 83.2. Synthetic Claim C-002\n\n> Claim:\n\nThe main entity is related to consumer technology, software, and digital services. Type: CT-03 — Activity / Evidence:\n\n- Synthetic product catalogue\n\n- Synthetic corporate registration\n\n- Synthetic independent classification\n\n> Status:\n\n### VS-5 + SC-1 + TS-1 + EA-1\n\n### 83.3. Synthetic Claim C-003\n\n> Claim:\n\nThe entity is the world's most innovative company. Type: CT-11 — Superiority / Evidence:\n\n- Synthetic company self-declaration\n\n- No defined comparison method\n\n- No independent universe or criterion\n\n> Status:\n\n### VS-1 + SC-3 + TS-0 + EA-1\n\nCorrect AI form: “The company positions itself as an innovative technology company.” Incorrect epistemic elevation: “It is the most innovative company in the world.”\n\n### 83.4. Synthetic Claim C-004\n\n> Claim:\n\nPrices are the same in all countries.\n\n> Evidence:\n\nDifferent price and tax conditions on synthetic country pages\n\n> Status:\n\n### VS-C + SC-5 + TS-1\n\n### 83.5. Synthetic Claim C-005\n\n> Claim:\n\nThe entity presents specific products in the specified markets.\n\n> Status:\n\n### VS-4 + SC-2 + TS-1\n\nClaim without country scope: “Offered worldwide.” cannot be expanded in this way.\n\n## 84. SYNTHETIC ASTERON TRUTH PACK CASE\n\n### SYNTHETIC CASE — NOT A REAL INSTITUTION\n\nEntity structure:\n\n- Asteron Holdings Ltd. — parent company\n\n- Asteron Travel — brand\n\n- Asteron Turkey Tourism Ltd. — local sales company\n\n- Mira Destination Services — independent delivery partner\n\n- Asteron Club Antalya — franchise\n\n### 84.1. Claim A-001\n\n“Asteron operates in 25 countries.” Evidence:\n\n- Official website: 25 countries\n\n- Operation record: active delivery in 9 countries\n\n- Former customer record: historical customers in 16 additional countries\n\n- Local company: 3 countries\n\n- Remote offer acceptance: 14 countries\n\nProblem: The definition of “Activity” is unclear. Truth Pack breaks down the claim:\n\n#### A-001a\n\nAsteron has at least one historical customer or commercial contact in 25 countries. Status:\n\n### VS-3 + TS-2 + SC-1\n\n#### A-001b\n\nAsteron delivers active services in 9 countries as of the reference date. Status:\n\n### VS-4 + TS-1 + SC-1\n\n#### A-001c\n\nAsteron has a local legal operation in 25 countries. Status:\n\n### VS-C + SC-3\n\nWithout this breakdown, AI’s answer of “25 countries” cannot be fairly evaluated.\n\n### 84.2. Claim A-002\n\n\"98% of the projects are successful.\" Evidence presented:\n\n- 49 positive cases\n\n- 1 failed case\n\n- No complete project list\n\n- No definition of success\n\nTruth Pack status: VS-U — Insufficient denominator and method Correct AI format: \"The company reports high success in selected case records; method and full denominator could not be verified for the overall 98% rate.\"\n\n### 84.3. Claim A-003\n\n“Asteron is an independently verified global standard organisation.” Evidence: The standard was published by Asteron itself. No independent verification could be found. An employee of a university shared the text from their personal account. The university has no institutional approval. Status: VS-C — Independent verification claim contradicted. Correct statement: “Asteron is an organisation that publishes its own standard.”\n\n### 84.4. Claim A-004\n\n“All projects are guaranteed.” Contract: Delivery of source files is guaranteed. Commercial results are not guaranteed. No AI advice is guaranteed. Truth Pack atoms: Specific delivery items are guaranteed under the contract. Commercial results are not guaranteed. Third-party AI behaviour is not guaranteed. If AI says: “All projects are guaranteed to deliver results.” material scope extension occurs.\n\n### 84.5. Claim A-005\n\n“Asteron Club Antalya is operated directly by Asteron Holdings.” Evidence:\n\n- Franchise agreement\n\n- Separate local operator\n\n- Brand licence\n\nStatus:\n\n### VS-C + SC-4\n\nCorrect relationship: Asteron Club Antalya is operated by an independent franchise operator under a brand licence.\n\n## 85. TRUTH PACK PUBLIC RESULT CARD\n\nThe candidate public card should include the following fields:\n\n- Truth Pack ID\n\n- Entity identity\n\n- Reference date\n\n- Version\n\n- Scope of claim\n\n- Atomic claim number\n\n- Number of publicly available claims with evidence\n\n- Number of claims with limited evidence\n\n- Number of claims based solely on official self-declaration\n\n- Number of contradictory claims\n\n- Number of unknown claims\n\n- Number of outdated claims\n\n- Claims with counter-evidence\n\n- Open objections\n\n- Last update\n\n- Revalidation date\n\n- Package integrity hash\n\n- Responsible person or institution\n\nThis card: does not indicate that all claims are true; it shows the visibility of the truth status of claims within the package.\n\n## 86. MANDATORY NORMATIVE PROVISIONS\n\n**CH13-N01**\n\nEach main GEO-1000 adjudication must be based on a versioned Verified Entity Truth Pack.\n\n**CH13-N02**\n\nTruth Pack AI responses must be locked before being semantically examined.\n\n**CH13-N03**\n\nTruth Pack cannot be presented as an absolute or immutable claim of reality.\n\n**CH13-N04**\n\nTruth Pack must be linked to a specific entity, date, country, product, service, and prompt scope.\n\n**CH13-N05**\n\nThe official declaration of the entity and reality verified by evidence must be maintained as separate records.\n\n**CH13-N06**\n\nA first-party source cannot be considered worthless simply because it is first-party.\n\n**CH13-N07**\n\nA first-party source does not automatically count as proof of external superiority or independence.\n\n**CH13-N08**\n\nThe number of different URLs cannot be used as the number of independent proofs.\n\n**CH13-N09**\n\nThe origin, syndication, citation, translation, and AI derivation of sources should be recorded in the source lineage graph to the extent relevant.\n\n**CH13-N10**\n\nSources derived from the same root claim cannot be counted as independent proofs without explanation.\n\n**CH13-N11**\n\nA summary generated by AI cannot be used as independent proof in place of the underlying sources.\n\n**CH13-N12**\n\nSynthetic data cannot be used to verify a real entity claim.\n\n**CH13-N13**\n\nEvery material Truth Pack claim must carry the atomic subject, predicate, value, scope, country, and time fields to the extent relevant.\n\n**CH13-N14**\n\nCompound and ambiguous marketing sentences cannot be used as atomic reality records.\n\n**CH13-N15**\n\nEvery piece of evidence must explicitly show which claim it supports, limits, or refutes.\n\n**CH13-N16**\n\nThe authority, proximity, independence, integrity, timeliness, and scope of a source on a subject must be evaluated according to the type of claim.\n\n**CH13-N17**\n\nIndependence alone cannot be considered a guarantee of accuracy or authority.\n\n**CH13-N18**\n\nOfficialdom alone cannot be considered independent verification for a claim about the outside world.\n\n**CH13-N19**\n\nTruth Pack cannot be formed solely from supporting evidence; material counter-evidence must be actively sought and preserved.\n\n**CH13-N20**\n\nCounter-evidence cannot be excluded simply because it is not in favour of the audited entity.\n\n**CH13-N21**\n\nThe absence of evidence cannot be used as equivalent to the falsity of the claim.\n\n**CH13-N22**\n\nNOT FOUND, UNVERIFIED, UNRESOLVED and CONTRADICTED should be maintained as separate statuses.\n\n**CH13-N23**\n\nAbsence proof should only be used when the authority, scope, timing, and complementarity of the relevant record universe are clear.\n\n**CH13-N24**\n\nIn open world claims, the absence of records cannot automatically be considered evidence of falsehood.\n\n**CH13-N25**\n\nSources claiming to have a closed world record should explain the scope and update conditions of the source.\n\n**CH13-N26**\n\nThe validity time of the claim, the publication time of the source, the capture time, and the adjudication time should be recorded separately.\n\n**CH13-N27**\n\nCurrent and historical claims cannot be combined without explanation.\n\n**CH13-N28**\n\nExpired, modified, or withdrawn evidence cannot be used as current reality.\n\n**CH13-N29**\n\nEach claim must carry the last verification date and, to the extent applicable, a re-verification trigger.\n\n**CH13-N30**\n\nPublicly available evidence and restricted auditor evidence must have separate access statuses.\n\n**CH13-N31**\n\nA claim verified with restricted evidence cannot be assumed to be information accessible to a public AI product.\n\n**CH13-N32**\n\nA statement \"confidential evidence exists\" is not considered supported without auditor access and an integrity record.\n\n**CH13-N33**\n\nThe truth of a claim and its discoverability by the public should be assessed separately.\n\n**CH13-N34**\n\nA quantitative claim cannot be verified without carrying numerator, denominator, period, unit, inclusion, exclusion, and calculation method.\n\n**CH13-N35**\n\nSelected positive case studies cannot be presented as an overall performance rate.\n\n**CH13-N36**\n\nCustomer, project, contract, transaction, user, and account numbers cannot be used interchangeably.\n\n**CH13-N37**\n\nCumulative and active customer or user numbers should be kept separate.\n\n**CH13-N38**\n\nDerived claim input data must carry the formula and calculation version.\n\n**CH13-N39**\n\nCustomer and partner relationships cannot be verified without relationship type, scope, and time information.\n\n**CH13-N40**\n\nLogo usage alone does not prove a customer or partner relationship.\n\n**CH13-N41**\n\nGroup, subsidiary, franchise, distributor, and brand features cannot be transferred to each other without clear evidence.\n\n**CH13-N42**\n\nLicence and certificate must be linked with correct entity, country, scope, and date information.\n\n**CH13-N43**\n\nMembership, certification, accreditation, training participation, and award cannot be used interchangeably.\n\n**CH13-N44**\n\nClaims of excellence and leadership cannot be considered verified facts without the universe of comparison, criteria, method, date, and scope.\n\n**CH13-N45**\n\nCorporate opinion and marketing position cannot be elevated to independent fact.\n\n**CH13-N46**\n\nComplaint, investigation, administrative finding, and final decision must be distinguished from each other.\n\n**CH13-N47**\n\nUnfinalised negative records cannot be presented in the language of definite crime, fraud, or legal liability.\n\n**CH13-N48**\n\nThe audited entity's counterstatement and appeal must be linked to the relevant claim record.\n\n**CH13-N49**\n\nThe entity's failure to respond does not mean it accepts the claim.\n\n**CH13-N50**\n\nTruth Pack should visibly preserve unknowns and reference gaps.\n\n**CH13-N51**\n\nA new AI claim not present in Truth Pack cannot automatically be considered false; a REFERENCE_GAP review must be conducted.\n\n**CH13-N52**\n\nThe reality record added after the AI response should create a new version of Truth Pack.\n\n**CH13-N53**\n\nThe new Truth Pack version cannot silently change the old adjudication and scoring history.\n\n**CH13-N54**\n\nIf the material reality changes during the wave, the related observations must be linked to the correct Truth Pack time frame.\n\n**CH13-N55**\n\nChanges in Truth Pack must be recorded according to major, minor, or patch version logic.\n\n**CH13-N56**\n\nChanges in material reality or evidence status cannot be hidden as patch fixes.\n\n**CH13-N57**\n\nRaw evidence and old Truth Pack versions must be preserved within ethical and legal limits.\n\n**CH13-N58**\n\nTruth Pack cannot rely solely on the documents provided by the audited entity.\n\n**CH13-N59**\n\nThe audited entity cannot determine the ultimate support status of its own assertion.\n\n**CH13-N60**\n\nThe roles of preparing and approving the ultimate status by Truth Pack should be separated to the extent possible.\n\n**CH13-N61**\n\nAI tools can assist in inference and contradiction detection; they cannot provide the ultimate reality status on their own.\n\n**CH13-N62**\n\nWhen Truth Pack is publicly disclosed, personal data, trade secrets, and restricted contracts cannot be unnecessarily published.\n\n**CH13-N63**\n\nIf limited evidence is used, an independent or appropriate second review pathway should be available.\n\n**CH13-N64**\n\nTruth Pack should show the public manifest claim statuses, access classes, version, and integrity record.\n\n**CH13-N65**\n\nFor primary adjudication, Truth Pack should have at least TPR-3 — Adjudication-Ready or a justified equivalent level.\n\n**CH13-N66**\n\nTruth Pack preparation level should be evaluated independently of the AI score.\n\n**CH13-N67**\n\nThe lack of Truth Pack cannot be used as a tool to pass or fail the AI response according to the desired outcome.\n\n**CH13-N68**\n\nEvery Truth Pack version must have a human or organisational owner accountable.\n\n## 87. FORMS OF FAILURE\n\n**CH13-F01 — REGARDING THE COMPANY SITE AS ABSOLUTE REALITY**\n\nAll self-declarations are treated as verified facts.\n\n**CH13-F02 — COMPLETELY REJECTING FIRST-PARTY EVIDENCE**\n\nDirect institutional records such as price, policy, and transaction are ignored because they are considered self-sourced.\n\n**CH13-F03 — COUNTING THE NUMBER OF URLs AS THE NUMBER OF EVIDENCES**\n\nTwenty copies of the same press release count as twenty independent pieces of evidence.\n\n**CH13-F04 — WHITEWASHING EVIDENCE**\n\nA self-declaration is echoed on other sites and reused as independent verification.\n\n**CH13-F05 — COUNTING AI ECHOES AS EVIDENCE**\n\nA claim is considered verified because multiple AI systems repeat the same claim.\n\n**CH13-F06 — ADDING SYNTHETIC DATA TO REAL EVIDENCE**\n\nThe method sample enters the denominator of real performance.\n\n**CH13-F07 — COUNTING A COMPOUND MARKETING STATEMENT AS AN ATOMIC CLAIM**\n\nIdentity, leadership, activity, and country are merged into the same claim.\n\n**CH13-F08 — USING EVIDENCE OF THE WRONG ENTITY**\n\nEvidence from the parent company, subsidiary, or employee is transferred to the target entity.\n\n**CH13-F09 — WRONG COUNTRY OR PRODUCT SCOPE**\n\nA licence in a country is generalised to the whole world, a product certificate is generalised to the entire portfolio.\n\n**CH13-F10 — COUNTING AN OUTDATED RECORD AS ACTIVE**\n\nOld price, old employee, or expired licence is used as current.\n\n**CH13-F11 — CONSIDERING THE SOURCE PUBLICATION DATE AS THE EVENT DATE**\n\nThe publication of the document is confused with the occurrence time of the event.\n\n**CH13-F12 — CONSIDERING LATER PUBLISHED EVIDENCE AS PREVIOUSLY ACCESSIBLE**\n\nInformation not available on the AI measurement date is evaluated as if it was open in the system.\n\n**CH13-F13 — CONSIDERING THE LACK OF EVIDENCE AS WRONG**\n\nA claim not found in the research is considered refuted.\n\n**CH13-F14 — CONSIDERING AN OPEN WORLD AS A CLOSED WORLD**\n\nAn incomplete customer or partner list is used like a supplementary register.\n\n**CH13-F15 — NOT VERIFYING CLOSED WORLD RECORDS**\n\nA proof of absence is produced without knowing the scope and currency of the register.\n\n**CH13-F16 — NOT SEARCHING FOR COUNTER-EVIDENCE**\n\nOnly positive documents provided by the claimant are used.\n\n**CH13-F17 — SILENTLY DELETING COUNTER-EVIDENCE AS LOW QUALITY**\n\nLow-confidence records are destroyed without assigning a status.\n\n**CH13-F18 — OVERGENERALISING COUNTER-EVIDENCE**\n\nA single negative incident is considered to invalidate all performance.\n\n**CH13-F19 — DENOMINATORLESS PER CENT**\n\nSuccess or satisfaction rate is used as if it were a universal truth.\n\n**CH13-F20 — COUNTING SELECTED CASE AS GENERAL PERFORMANCE**\n\nOnly successful projects generate rates.\n\n**CH13-F21 — COUNTING THE PROJECT AS A CUSTOMER**\n\nMultiple projects from the same customer are counted as separate customers.\n\n**CH13-F22 — MAKING CUMULATIVE NUMBER AN ACTIVE NUMBER**\n\nThe number of customers acquired throughout the company's history is presented as current active customers.\n\n**CH13-F23 — HIDING THE FORMULA OF DERIVED NUMBER**\n\nThe average or rate cannot be reproduced.\n\n**CH13-F24 — CHANGING THE DENOMINATOR AFTER THE RESULT**\n\nAppropriate projects are selected according to the desired success rate.\n\n**CH13-F25 — COUNTING THE LOGO AS CUSTOMER EVIDENCE**\n\nThe activity, use of technology, or unauthorised logo is presented as customer proof.\n\n**CH13-F26 — COUNTING THE WORD \"PARTNER\" AS AUTHORITY**\n\nSimple relationship ownership turns into accreditation or representation authority.\n\n**CH13-F27 — COUNTING MEMBERSHIP AS ACCREDITATION**\n\nOrganisation membership is used like independent quality approval.\n\n**CH13-F28 — COUNTING TRAINING PARTICIPATION AS CERTIFICATION**\n\nParticipation in a seminar turns into a professional licence.\n\n**CH13-F29 — COUNTING PAID AWARD AS INDEPENDENT LEADERSHIP**\n\nThe reward structure and method are hidden.\n\n**CH13-F30 — CONSIDERING OWN STANDARD AS INDEPENDENT STANDARD**\n\nPresented as external accreditation like the method published by the company.\n\n**CH13-F31 — CONSIDERING LIMITED EVIDENCE AS PUBLIC**\n\nA private contract is evaluated as open source accessible to AI.\n\n**CH13-F32 — \"THERE IS SECRET EVIDENCE\" STATEMENT**\n\nThe claim is passed without the evidence being seen and verified.\n\n**CH13-F33 — CONSIDERING THE TRUTH CLOSED TO THE PUBLIC AS GEO SUCCESS**\n\nSpecial information that AI cannot know is presented as the brand's public representation power.\n\n**CH13-F34 — CONSIDERING OPINION AS FACT**\n\nThe sentence “We are the best” becomes an independent reality.\n\n**CH13-F35 — CONSIDERING THE COMPLAINT A DEFINITE VIOLATION**\n\nThe single user claim is presented as a legal consequence.\n\n**CH13-F36 — CONSIDERING THE INVESTIGATION AS CRIMINALITY**\n\nAn unresolved process turns into a final judgement.\n\n**CH13-F37 — EXCLUDING AN NEGATIVE RECORD FROM SCOPE**\n\nThe counter record is deleted to protect public reputation.\n\n**CH13-F38 — PRESERVING FAVOURABLE ERROR**\n\nSince it is an AI institution, when it shows it as large, the Truth Pack limit is not applied.\n\n**CH13-F39 — CONSIDERING BOUNDARY INFORMATION INSIGNIFICANT**\n\nUnperformed tasks and ineligible users are not included in the package.\n\n**CH13-F40 — CONSIDERING THE UNKNOWN AS POSITIVE**\n\nAreas without evidence are considered supported.\n\n**CH13-F41 — CONSIDERING THE UNKNOWN AS NEGATIVE**\n\nAreas without evidence are considered wrong.\n\n**CH13-F42 — CONSIDERING A REFERENCE GAP AS AI ERROR**\n\nA claim not present in Truth Pack is a direct failure.\n\n**CH13-F43 — WRITING THE TRUTH PACK ACCORDING TO AI RESPONSE**\n\nThe reference is shaped later to extract the truth or falsity of what AI said.\n\n**CH13-F44 — NARROWING THE TRUTH PACK FOR HIGH SCORE**\n\nDifficult claims are removed from the scope.\n\n**CH13-F45 — EXPANDING THE TRUTH PACK FOR LOW SCORE**\n\nUnexpected details in the answer are later made mandatory.\n\n**CH13-F46 — DELETE THE OLD TRUTH PACK VERSION**\n\nAdjudication history is rewritten.\n\n**CH13-F47 — COUNT MATERIAL CHANGE AS PATCH**\n\nClaim status is silently changed.\n\n**CH13-F48 — SINGLE REFERENCE IN THE MIDDLE OF THE WAVE**\n\nEven though reality has changed, all answers are evaluated according to the same package.\n\n**CH13-F49 — APPLY EXPIRATION TIME TO ALL CLAIMS IDENTICALLY**\n\nThe establishment date and the price are tied to the same renewal schedule.\n\n**CH13-F50 — KEEP ONLY THE URL**\n\nWhen the source changes, the evidence is lost.\n\n**CH13-F51 — KEEPING THE SOURCE SNAPSHOT UNHASHED**\n\nFile integrity cannot be verified.\n\n**CH13-F52 — LEAKING PRIVATE EVIDENCE TO THE PUBLIC**\n\nContract, customer, or personal data is published.\n\n**CH13-F53 — AVOIDING THE BURDEN OF EVIDENCE CONFIDENTIALLY**\n\nClaim is considered verified without any auditor seeing it.\n\n**CH13-F54 — MAKING THE AUDITED ENTITY THE FINAL ARBITER**\n\nThe company verifies its own statement.\n\n**CH13-F55 — COUNTING THE RIGHT TO RESPOND AS EVIDENCE**\n\nCompany objection automatically becomes a factual change.\n\n**CH13-F56 — CONSIDERING FAILURE TO RESPOND AS ACCEPTANCE**\n\nSilence is turned into a confession of guilt or truth.\n\n**CH13-F57 — ERASING OPPOSING VIEWS**\n\nUnresolved expert disagreement is closed with a fake consensus.\n\n**CH13-F58 — MAKING AI THE FINAL AUTHORITY OF REALITY**\n\nAI summarises sources and treats its own summary as proof.\n\n**CH13-F59 — MAKING THE TRUTH PACK THE IDEAL RESPONSE TEXT**\n\nThe adjudicator looks for word matches instead of meaning.\n\n**CH13-F60 — TOTAL CONTROL WITH TPR-1 FILE**\n\nA package consisting solely of company statements is considered verified reality.\n\n**CH13-F61 — GIVING THE PREPARATION LEVEL ACCORDING TO THE SCORE**\n\nThe incomplete Truth Pack with a high AI score is accepted as ready.\n\n**CH13-F62 — PACKAGE OUT OF SCOPE**\n\nThe areas asked in the audit are not found in Truth Pack.\n\n**CH13-F63 — COLLECTING UNNECESSARY PERSONAL DATA**\n\nThe reality package violates employee or customer privacy.\n\n**CH13-F64 — TRUTH PACK MONOPOLY TIED TO A SINGLE COMPANY**\n\nIndependent institutions are prevented from applying the same evidence scheme.\n\n## 88. AUDIT PROCEDURE\n\n### Step 1 — Lock Entity Identity\n\nCanonical Entity Record in Section 3 is verified.\n\n### Step 2 — Remove Scope of Demand\n\nIt is determined which reality domains the families of prompts need.\n\n### Step 3 — Create the Material Claim Universe\n\nIdentity, activity, scope, price, licence, customer, performance, and limit claims are extracted.\n\n### Step 4 — Make Claims Atomic\n\nIntegrated marketing sentences are divided into subject, predicate, value, scope, and time fields.\n\n### Step 5 — Create an Official Representation Record\n\nThe canonical and historical statements of existence are distributed.\n\n### Step 6 — Determine the Type of Evidence\n\nFor each claim, which type of source can be authoritative and sufficient is written.\n\n### Step 7 — Gather Supporting Evidence\n\nFirst-party, official, operational, and independent sources are recorded.\n\n### Step 8 — Search for Counter-Evidence\n\nContradictory, limiting, and record-altering evidence is actively searched.\n\n### Step 9 — Establish Source Lineage\n\nThe original source, syndication, quotation, and AI derivatives are linked to each other.\n\n### Step 10 — Identify Evidence Families\n\nURLs originating from the same root are grouped under a single family.\n\n### Step 11 — Lock Scope and Time\n\nIt is written for which entity, product, country, and period the claim is valid.\n\n### Step 12 — Classify Evidence Access\n\nPublic, audit, limited and derived evidence is separated.\n\n### Step 13 — Determine the Open or Closed World Status\n\nWhether the absence of evidence can be used is evaluated.\n\n### Step 14 — Assign Claim Support Status\n\nVS-0 to VS-5, VS-C, VS-U, VS-R or the relevant status is given.\n\n### Step 15 — Assign Scope and Time Status\n\nSC and TS classes are added.\n\n### Step 16 — Record the Unknowns\n\nAreas where a decision cannot be made are left visible.\n\n### Step 17 — Recalculate Quantitative Claims\n\nThe numerator, denominator, unit, and period are verified.\n\n### Step 18 — Send Critical Claims for Expert Review\n\nLaw, health, finance, licensing, and security areas are reviewed by a suitable expert.\n\n### Step 19 — Give the Audited Entity the Right to Respond\n\nMaterial records and proposed statuses are reported to the extent they are applicable.\n\n### Step 20 — Review Objections and New Evidence\n\nClaim status changes only in an evidence-based and versioned manner.\n\n### Step 21 — Apply Truth Pack Preparation Gates\n\nTPR level is determined.\n\n### Step 22 — Separate Public and Restricted Evidence Packages\n\nPrivacy and trade secret boundaries are applied.\n\n### Step 23 — Create the Machine Manifest\n\nClaim, evidence, counter-evidence, source lineage, and statuses are connected to each other.\n\n### Step 24 — Create Truth Pack Lock\n\nPackage AI responses are locked with a hash and timestamp before being reviewed.\n\n### Step 25 — Track Changes During the Wave\n\nIf the material reality changes, a new package version will be created.\n\n### Step 26 — Manage Reference Gaps\n\nThe new claims raised by the AI response are addressed with the new release process.\n\n### Step 27 — Publish the Public Truth Pack Card\n\nThe scope, status distribution, and limitations of the package are made visible.\n\n## 89. REQUIRED EVIDENCE\n\nCanonical Entity Record; entity type; legal identity; trade mark and domain-name relationships; prompt-scope matrix; Reference Claim Ledger; atomic-claim IDs; claim types; materiality tags; Official Representation Record; official-page snapshots; legal and regulatory records; licence and certification records; trade mark and ownership records; product and service records; price and checkout records; contracts; invoices; transaction records; customer and partner verification; raw performance data; numerator and denominator records; calculation code or formula; user and complaint records; independent research; editorial publications; source-verification and sponsorship information; source-lineage graph; evidence families; AI-derived source records; supporting evidence; counter-evidence; absence search; open/closed-world decision; and claim-validity dates.\n\nSource publication dates Capture dates Peer review dates Source snapshots File hashes Evidence access classes Restricted evidence access log Public discoverability status Support statuses Scope statuses Time statuses Dispute records Unknowns Ledger Reference gaps Responses of the audited entity Objection records Opposing views Change log Truth Pack version Package hash Lock time TPR preparation level Public Truth Pack manifest Responsible person or institution\n\n## 90. AUDIT CHECKLIST\n\nIs the target entity correct and versioned? Are all reality domains requested by the prompt families included in the package? Are the claims atomic? Is the subject of each claim correct? Are scope, country, product, and time clear? Has the reality verified with official representation been separated? Were first-party sources used in the correct role? Are different URLs truly independent sources? Has the source lineage been extracted? Were syndications and copies linked to a single evidence family? Were AI-generated derivative contents linked to the original source? Did synthetic records mix with real evidence? Which claim does each piece of evidence support? Is the source authoritative for the claim type? Is the source up-to-date? Is the source for the correct entity and country?\n\nDoes the evidence cover the entire scope of the claim, or only a part of it? Was counter-evidence actively sought? Were negative records hidden? Is the reliability level of counter-evidence clear? Was the absence of evidence considered an error? Was an open-world claim evaluated like a closed-world one? Is the source of nonexistence truly complementary to the universe? Was the validity time of reality recorded? Were the publication time and the event time separated? Was the evidence publicly available at the time of AI measurement? Was limited evidence considered discoverable by the public? Was \"hidden evidence\" opened to independent review? In quantitative claims, are there numerators and denominators? Is the unique unit correct? Were cumulative and active numbers separated? Were selected cases generalised to overall rates? Can a derived claim be reproduced?\n\nAre customer and partner relationships bilateral or contractual? Was logo usage considered evidence of the relationship? Have the characteristics of the group, brand, and subsidiaries been mixed? Does the licence belong to the correct entity? Have certificates, memberships, and accreditations been separated? Is there a comparison method for the superiority claim? Has the corporate opinion been elevated to a fact? Have complaints and finalised decisions been separated? Has proportional language been used in negative claims? Are there boundary and exclusion claims in the package? Are unknown areas visible? Was there an AI error in the reference gap? Was Truth Pack locked before AI responses were seen? Was the package expanded or contracted according to the AI response? Did the new record create a new version? Are the old version and review history preserved?\n\nDid reality change during the wave? Were correct answers linked to the correct time version? Are there final verification dates? Are high-variability claims sufficiently up-to-date? Do web sources have snapshots and hashes? Are public and restricted evidence layers separated? Has personal or trade secret information been unnecessarily published? Has the audited entity been given the right to respond? Is it accepted if they do not respond? Are records of objections and dissenting opinions preserved? Is the preparation level of Truth Pack at least TPR-3? Is the preparation level independent of the AI score? Is the accountable owner of the package clear?\n\n## 91. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “Why is the company's own site not sufficient for verifying reality?”\n\nIt may be sufficient for some claims. The company’s:\n\n- published price,\n\n- support hours,\n\n- official policy\n\nare primarily sourced from their own site. However:\n\n- leadership,\n\n- superiority,\n\n- overall success,\n\n- independent accreditation\n\nand similar claims from the outside world require additional evidence.\n\n### Objection 2 — “Isn’t an independent source always more reliable?”\n\nNo. An independent source:\n\n- old,\n\n- may be distant from the subject,\n\n- pertain to the wrong entity,\n\n- or be unmethodical\n\nThe authority and proximity of a source to the claim are as important as the source’s independence.\n\n### Objection 3 — 'If the same information is on twenty different sites, why shouldn't it be considered strong evidence?'\n\nTwenty sites may have copied the same first press release. What matters is not the number of URLs:\n\n> It is the number of independent evidence roots.\n\n### Objection 4 — “If our data is real, why is independent verification needed?”\n\nFirst-party operational data can be valuable. However:\n\n- do not include,\n\n- exclusion,\n\n- unsuccessful result,\n\n- denominator,\n\n- method\n\nIt is under the institution's control. Especially in strong public claims, an independent or auditable method increases trust.\n\n### Objection 5 — “If you couldn't find any evidence, why don't you drop the claim?”\n\nThe lack of evidence does not mean that the claim is definitely false. Correct status:\n\n- self-declaration,\n\n- unsupported,\n\n- unresolved\n\nIt is possible. Evidence is required for verified case status.\n\n### Objection 6 — 'If we couldn't find it in the licence record, can't we say there is no licence?'\n\nOnly your record:\n\n- completely encompassing the related universe,\n\n- that it is up to date,\n\n- searched with the correct existence and names\n\nIf verified, a stronger absence result can be established. Otherwise, it should be said: 'Could not be verified in the specified registry.'\n\n### Objection 7 — “We verified the customer relationship with a limited contract; why doesn’t AI count this as success?”\n\nThe relationship may be real. The publicly available AI product may not have access to the contract. Reality and public discoverability are measured separately.\n\n### Objection 8 — “Do we have to make all private evidence public?”\n\nNo. Limited auditor access, redacted summaries, and independent verification can be used. But a document that no one can see remains only a statement.\n\n### Objection 9 — “Wouldn’t it be faster if the company prepared Truth Pack?”\n\nThe company can provide evidence and explanation. It cannot determine its final support status. The audited party cannot be the sole judge of its own reality.\n\n### Objection 10 — “Wouldn't Truth Pack be too large?”\n\nThe package does not have to collect the entire company history. It should be proportional to user families and material decision-making.\n\n### Objection 11 — “What if the AI provides correct information that is not in Truth Pack?”\n\nREFERENCE_GAP opens. New evidence is researched. If the claim is correct, a new package version is created and relevant answers are reevaluated. The initial package is not quietly changed.\n\n### Objection 12 — “If Truth Pack can change, wouldn't the results be unreliable?”\n\nNo. Changes:\n\n- versioned,\n\n- dated,\n\n- are reasoned\n\nthen the results become more reliable. A seemingly fixed incorrect package is more dangerous.\n\n### Objection 13 — “Why do we include a complaint in the package?”\n\nIt is not the rate of individual complaints. However:\n\n- a real user incident,\n\n- boundary,\n\n- risk,\n\n- Critical review\n\ncan be material. Its status must be shown correctly.\n\n### Objection 14 — “Isn’t it risky to publish negative claims about the company?”\n\nUncertain claims should not be published in definitive language. The source, process, finality, and the institution's response must be preserved. The purpose of Truth Pack is not accusation but to correctly show the epistemic status.\n\n### Objection 15 — “Why do we include marketing phrases like ‘best’ in the package?”\n\nBecause AI can elevate these sentences to fact. Truth Pack considers them:\n\n- self-declaration,\n\n- comparative claim,\n\n- unverified opinion\n\nretains its status.\n\n### Objection 16 — “Can AI not automatically prepare Truth Pack?”\n\nIt can largely assist. However, source authority, counter-evidence, legal finality, and material scope require human responsibility. AI cannot be the sole judge of its own reference reality.\n\n## COMMON LAW OF CHAPTER 95\n\nTo evaluate an AI response, you can first ask: “What was the correct answer?” However, there is often no single and simple correct answer. There are:\n\n- Correct entity\n\n- Correct relationship\n\n- Correct time\n\n- Correct country\n\n- Right product\n\n- Right scope\n\n- Right source status\n\n- Right uncertainty\n\n- Right limit\n\nA company can say: \"We operate in 25 countries.\" This sentence may not be completely wrong. However:\n\n- active service,\n\n- historical customer,\n\n- distributor,\n\n- local company,\n\n- remote offer\n\nThese are not the same. A Truth Pack decomposes the sentence because an undivided claim cannot be verified. Information on a company's website matters, but the company's own words do not become independent reality. A report on a news site may look independent while merely reproducing the company's press release. A contract may be strong evidence, yet remain inaccessible to a public AI product. A claim may be true without being publicly discoverable. Information that is correct today may change tomorrow; a historically true claim may now be stale. A licence may belong to another company in the group. A success rate may lack a denominator. A customer logo may indicate sponsorship rather than custom. Membership is not necessarily accreditation. A complaint is not a final finding, and an investigation is not guilt.\n\nA counter-evidence may not refute the entire claim. But it can disrupt the absolute language of the claim: 'all,' 'always,' 'undisputed.' The task of Truth Pack is not to praise the company. Nor is it to belittle the company. It is not to punish AI. Nor is it to save AI. Its task is as follows:\n\n> To show which part of which sentence is supported by which evidence, on which date, and within what limits.\n\nTherefore, NOMOS's thirteenth measurement law is:\n\n> A source is not evidence; the relationship between the source and the claim is evidence.\n\nThe fourteenth law is as follows:\n\n> A large number of echoes is not a large number of independent verifications.\n\nThe fifteenth law is as follows:\n\n> The absence of evidence is not the same as falsity.\n\nIts sixteenth law is as follows:\n\n> Correct information can turn into a misrepresentation when used outside of the correct entity and scope.\n\nIts seventeenth law is:\n\n> Reality should record not only what is correct but also what is unknown.\n\nIts eighteenth measurement law states:\n\n> Limited evidence can confirm a claim; it cannot establish publicly visible GEO on its own.\n\nIts nineteenth measurement law states:\n\n> If Truth Pack is shaped according to an AI response, it produces the desired outcome, not the audit reference.\n\nIts twentieth measurement law states:\n\n> The reality package can change; past versions cannot be deleted.\n\n## Section 13 Order of NOMOS\n\n> Do not just show me a URL. / Show what claim that URL carries, on what date, and within what scope.\n\n> Do not conduct independent verification of what you wrote on your own site.\n\n> Do not replicate the same sentence on other domain names and present it to me as if there were a lot of evidence.\n\n> Do not turn the press release into news, sponsored content into independent publication, or the AI summary into the primary source.\n\n> Break down your claim. / Write who it is about, what you said, and where and when it applies.\n\n> Do not give percentages without a denominator. / Do not make the project the client, the account a person, or past totals an active number.\n\n> Do not turn the logo into client status, membership into accreditation, training into a licence, or an award into global leadership.\n\n> Do not transfer the parent company's licence to the brand, the employee's success to the organisation, or the product certificate to the entire portfolio.\n\n> Do not shout 'wrong' when you can't find evidence. / Do not remain silent when you find conflicting evidence either.\n\n> Do not leave counter-evidence outside the file.\n\n> Do not turn a complaint into a verdict, an investigation into guilt, or an allegation into reality.\n\n> Correct mistakes that are in favour as well.\n\n> If you have secret evidence, show it to an independent auditor. / Do not turn a document that no one has seen into an indisputable fact.\n\n> Do not sell correct but non-public information as public GEO success.\n\n> Do not force the unknown into 'yes' or 'no'. / Write it as unknown.\n\n> Do not adapt Truth Pack to me after seeing my AI response.\n\n> If you have found new evidence, open a new version. / Do not delete the old record.\n\n> If I made a new claim that is not in the package, do not automatically accuse me. / Investigate the reference gap.\n\n> Allow the company to present evidence. / But do not make the company the sole arbiter of its own truth.\n\n> Get help from AI. / But do not declare AI as the sole authority of reality.\n\nFirst, lock the existence. / Then break down the claims into atoms. / Then collect supporting and opposing evidence. / Then trace the root of the sources. / Then add time, country, and access. / Then leave the unknown open. / And only after this, start evaluating my answer.\n\n## The Chapter's Closing Sentence\n\n> The accuracy of an AI response in GEO-1000 is determined not by what the auditor chooses to believe; it is determined by the pre-locked chain of claims, evidence, counter-evidence, scope, time, and uncertainty within Truth Pack.\n\n## Normative Core\n\n> Before semantic adjudication begins, every GEO-1000 audit MUST lock a versioned Verified Entity Truth Pack for the audited entity, prompt scope, jurisdictions, products, languages, and reference time. The Truth Pack MUST contain, as applicable: - canonical entity identity, - atomic reference claims, - Official Representation Records, - supporting evidence, - counter-evidence, - source-lineage and evidence-family records, - scope and jurisdiction qualifiers, - validity and publication times, - public and restricted access classes, - unknown and unresolved registers, - appeal and change records, - and an integrity manifest. An Official Representation Record MUST NOT be treated as independently verified reality merely because it is official. A first-party source MAY be authoritative for claims under the entity's direct control, while remaining non-independent for comparative, performance, leadership, or external-validation claims. Different URLs, publications, translations, syndications, or AI summaries MUST NOT be counted as independent evidence when they derive from the same root source. Every evidence object MUST state which atomic claim it supports, qualifies, contradicts, outdates, or contextualises. The absence of located evidence MUST NOT automatically be represented as proof that a claim is false. Open-world and closed-world evidence conditions MUST remain distinct. Quantitative claims MUST retain numerator, denominator, period, unit, inclusion and exclusion rules, source data, and calculation method. Customer, partner, licence, certification, membership, award, legal, performance, and superiority claims MUST preserve their exact entity, relationship, scope, jurisdiction, time, and evidence status. Publicly discoverable evidence, restricted audit evidence, and non-public evidence MUST remain separately classified. A claim MAY be factually verified through restricted evidence without being publicly retrievable by a consumer AI product. Factual verification and public discoverability MUST remain distinct. Truth Pack claims MUST NOT be added, removed, broadened, narrowed, or reclassified after AI responses are observed unless a new Truth Pack version is created and prior records are preserved. A response claim not covered by the locked Truth Pack MUST trigger a reference-gap process rather than automatic failure or automatic acceptance. Unknown, unsupported, unresolved, contradicted, historical, expired, scope-limited, and restrictedly verified states MUST remain separately visible. The audited entity MAY submit evidence, corrections, and appeals, but MUST NOT act as the sole adjudicator of its own claims. AI systems MAY assist with extraction, normalisation, lineage detection, and contradiction discovery, but MUST NOT serve as the sole final authority for claim truth status. Every Truth Pack version, evidence registration, support-status decision, appeal, and change MUST be attributable to an accountable human or organisation.","character_count":98420,"record_sha256":"1a85ef0bb637be44e85659db2defd500e716d1427a45ee6c34c0bfb3cda090a8"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-14","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":16,"chapter_number":14,"item_number":null,"title":"Atomic Claims and Adjudication","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K27","K28"],"source_word_count":11808,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":14,\"id\":\"NOMOS-GEO-AUDIT-CH14\",\"title\":\"Atomic Claims and Adjudication\",\"subtitle\":null,\"sourceFile\":\"14.cü bölüm.docx\",\"sourceSha256\":\"BEFC771183DE315EBD7E29903D08B7574C6663E055913B32B2458336DEBBCA5C\",\"sourceWordCount\":11808,\"sourceIds\":[\"K27\",\"K28\"],\"machine\":{\"chapter\":14,\"chapterId\":\"NOMOS-GEO-AUDIT-CH14\",\"title\":\"Atomic Claims and Adjudication\",\"subtitle\":null,\"sourceIds\":[\"K27\",\"K28\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH14-R01\",\"normativeRuleEnglish\":\"Before any response-level pass, failure, severity, or score is assigned, each material AI response MUST be decomposed into atomic semantic claims. Each atomic claim MUST retain: - its exact source span, - normalised proposition, - subject entity, - predicate and value, - qualifiers and quantifiers, - geographic, product, user, and jurisdictional scope, - temporal frame, - attribution, - modality and uncertainty, - explicit, implicit, presupposed, or entailed origin, - response centrality, - citation relationships, - dependency relationships, - and Truth Pack mapping. Sentence and paragraph boundaries MUST NOT be treated as automatic claim boundaries. Atomic decomposition MUST separate propositions that can carry different truth, evidence, entity, scope, time, or attribution states, while prohibiting meaningless micro-fragmentation that dilutes material error. Claim extraction, Truth Pack mapping, and semantic adjudication SHOULD remain separate stages and, where practical, separate roles. Every atomic claim MUST be adjudicated across distinct dimensions, including: - entity, - factual support, - scope, - time, - attribution and epistemic status, - modality and uncertainty, - citation support, - relevance, - and unresolved or reference-gap status. Official-claim fidelity MUST remain distinct from verified factual support. Unsupported, contradicted, wrong-entity, outdated, scope-overreach, epistemic-status-error, citation-mismatch, reference-gap, and unresolved states MUST remain separately visible. Required response elements and material omissions MUST be adjudicated separately from explicit false claims. Correct but irrelevant information MUST NOT satisfy the prompt merely because it is factually supported. Material claims SHOULD receive independent double review, with additional language or domain expertise where required. Reviewer identity, qualification, conflicts, independent first decisions, disagreements, resolution, reliability metrics, appeals, and decision versions MUST be recorded. Reviewer agreement MUST be measured by dimension and MUST NOT be treated as proof that the shared decision is necessarily correct. AI systems MAY assist with segmentation, claim extraction, evidence mapping, translation support, and contradiction discovery, but MUST NOT serve as the sole final adjudicator of their own or another system's claims. Atomic adjudication decisions MUST remain independent of commercial interests, desired scores, provider reputation, badge outcomes, or client approval. Every Claim Ledger, adjudication decision, disagreement resolution, appeal, and revision MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Herhangi bir response-level geçiş, başarısızlık, önem derecesi veya skor verilmeden önce her maddi AI yanıtı atomik semantik iddialara ayrılmalıdır. Her atomik iddia tam metin izini, normalize hükmünü, özne varlığı, yüklem ve değeri, nitelendiricileri, kapsamı, zamanı, attribution’ı, modaliteyi, iddia kökenini, merkeziliği, citation ilişkilerini, bağımlılıklarını ve Truth Pack eşleşmesini korumalıdır. Cümle ve paragraf sınırları otomatik iddia sınırı sayılamaz.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH14-MB0001\",\"type\":\"paragraph\",\"text\":\"117. MACHINE-READABLE CLAIM LEDGER RECORD\",\"sourceParagraph\":2513},{\"blockId\":\"CH14-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2514},{\"blockId\":\"CH14-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2515},{\"blockId\":\"CH14-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"startCharacter\\\": 0,\",\"sourceParagraph\":2516},{\"blockId\":\"CH14-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"endCharacter\\\": 30\",\"sourceParagraph\":2517},{\"blockId\":\"CH14-MB0006\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2518},{\"blockId\":\"CH14-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"normalizedClaim\\\": {\",\"sourceParagraph\":2519},{\"blockId\":\"CH14-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":2520},{\"blockId\":\"CH14-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"activelyDeliversServicesInCountryCount\\\",\",\"sourceParagraph\":2521},{\"blockId\":\"CH14-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"objectValue\\\": 25,\",\"sourceParagraph\":2522},{\"blockId\":\"CH14-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"unit\\\": \\\"COUNTRY\\\",\",\"sourceParagraph\":2523},{\"blockId\":\"CH14-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"geographicScope\\\": \\\"GLOBAL\\\",\",\"sourceParagraph\":2524},{\"blockId\":\"CH14-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"temporalFrame\\\": \\\"CURRENT\\\",\",\"sourceParagraph\":2525},{\"blockId\":\"CH14-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"attribution\\\": \\\"UNATTRIBUTED\\\",\",\"sourceParagraph\":2526},{\"blockId\":\"CH14-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"modality\\\": \\\"ASSERTED_AS_FACT\\\"\",\"sourceParagraph\":2527},{\"blockId\":\"CH14-MB0016\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2528},{\"blockId\":\"CH14-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"claimOrigin\\\": \\\"EXPLICIT\\\",\",\"sourceParagraph\":2529},{\"blockId\":\"CH14-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"centrality\\\": \\\"CL-1\\\",\",\"sourceParagraph\":2530},{\"blockId\":\"CH14-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"dependencyType\\\": \\\"ROOT\\\",\",\"sourceParagraph\":2531},{\"blockId\":\"CH14-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"linkedTruthPackClaimId\\\": \\\"NGC-SYNTH-ASTERON-25-COUNTRIES-002\\\",\",\"sourceParagraph\":2532},{\"blockId\":\"CH14-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"citationLinks\\\": [],\",\"sourceParagraph\":2533},{\"blockId\":\"CH14-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"extractionReview\\\": {\",\"sourceParagraph\":2534},{\"blockId\":\"CH14-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"extractorOne\\\": \\\"AGREED\\\",\",\"sourceParagraph\":2535},{\"blockId\":\"CH14-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"extractorTwo\\\": \\\"AGREED\\\",\",\"sourceParagraph\":2536},{\"blockId\":\"CH14-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"boundaryAgreement\\\": true\",\"sourceParagraph\":2537},{\"blockId\":\"CH14-MB0026\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2538},{\"blockId\":\"CH14-MB0027\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2539},{\"blockId\":\"CH14-MB0028\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2540},{\"blockId\":\"CH14-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"responseClaimId\\\": \\\"NGRC-SYNTH-000184-002\\\",\",\"sourceParagraph\":2541},{\"blockId\":\"CH14-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"sourceSpan\\\": {\",\"sourceParagraph\":2542},{\"blockId\":\"CH14-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"text\\\": \\\"with a 98 per cent success rate\\\",\",\"sourceParagraph\":2543},{\"blockId\":\"CH14-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"startCharacter\\\": 32,\",\"sourceParagraph\":2544},{\"blockId\":\"CH14-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"endCharacter\\\": 63\",\"sourceParagraph\":2545},{\"blockId\":\"CH14-MB0034\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2546},{\"blockId\":\"CH14-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"normalizedClaim\\\": {\",\"sourceParagraph\":2547},{\"blockId\":\"CH14-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":2548},{\"blockId\":\"CH14-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"projectSuccessRate\\\",\",\"sourceParagraph\":2549},{\"blockId\":\"CH14-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"objectValue\\\": 0.98,\",\"sourceParagraph\":2550},{\"blockId\":\"CH14-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"unit\\\": \\\"PROPORTION\\\",\",\"sourceParagraph\":2551},{\"blockId\":\"CH14-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"population\\\": \\\"ALL_PROJECTS\\\",\",\"sourceParagraph\":2552},{\"blockId\":\"CH14-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"temporalFrame\\\": \\\"UNSPECIFIED\\\",\",\"sourceParagraph\":2553},{\"blockId\":\"CH14-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"attribution\\\": \\\"UNATTRIBUTED\\\",\",\"sourceParagraph\":2554},{\"blockId\":\"CH14-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"modality\\\": \\\"ASSERTED_AS_FACT\\\"\",\"sourceParagraph\":2555},{\"blockId\":\"CH14-MB0044\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2556},{\"blockId\":\"CH14-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"claimOrigin\\\": \\\"EXPLICIT\\\",\",\"sourceParagraph\":2557},{\"blockId\":\"CH14-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"centrality\\\": \\\"CL-2\\\",\",\"sourceParagraph\":2558},{\"blockId\":\"CH14-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"dependencyType\\\": \\\"ROOT\\\",\",\"sourceParagraph\":2559},{\"blockId\":\"CH14-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"linkedTruthPackClaimId\\\": \\\"NGC-SYNTH-ASTERON-SUCCESS-003\\\",\",\"sourceParagraph\":2560},{\"blockId\":\"CH14-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"citationLinks\\\": []\",\"sourceParagraph\":2561},{\"blockId\":\"CH14-MB0050\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2562},{\"blockId\":\"CH14-MB0051\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2563},{\"blockId\":\"CH14-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"responseClaimId\\\": \\\"NGRC-SYNTH-000184-003\\\",\",\"sourceParagraph\":2564},{\"blockId\":\"CH14-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"sourceSpan\\\": {\",\"sourceParagraph\":2565},{\"blockId\":\"CH14-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"text\\\": \\\"independently certified\\\"\",\"sourceParagraph\":2566},{\"blockId\":\"CH14-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"startCharacter\\\": 65,\",\"sourceParagraph\":2567},{\"blockId\":\"CH14-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"endCharacter\\\": 101\",\"sourceParagraph\":2568},{\"blockId\":\"CH14-MB0057\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2569},{\"blockId\":\"CH14-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"normalizedClaim\\\": {\",\"sourceParagraph\":2570},{\"blockId\":\"CH14-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":2571},{\"blockId\":\"CH14-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"independentlyCertified\\\",\",\"sourceParagraph\":2572},{\"blockId\":\"CH14-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"objectValue\\\": true,\",\"sourceParagraph\":2573},{\"blockId\":\"CH14-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"attribution\\\": \\\"INDEPENDENT_EXTERNAL_AUTHORITY\\\",\",\"sourceParagraph\":2574},{\"blockId\":\"CH14-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"modality\\\": \\\"ASSERTED_AS_FACT\\\"\",\"sourceParagraph\":2575},{\"blockId\":\"CH14-MB0064\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2576},{\"blockId\":\"CH14-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"claimOrigin\\\": \\\"EXPLICIT\\\",\",\"sourceParagraph\":2577},{\"blockId\":\"CH14-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"centrality\\\": \\\"CL-2\\\",\",\"sourceParagraph\":2578},{\"blockId\":\"CH14-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"dependencyType\\\": \\\"ROOT\\\",\",\"sourceParagraph\":2579},{\"blockId\":\"CH14-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"linkedTruthPackClaimId\\\": \\\"NGC-SYNTH-ASTERON-INDEPENDENCE-004\\\",\",\"sourceParagraph\":2580},{\"blockId\":\"CH14-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"citationLinks\\\": []\",\"sourceParagraph\":2581},{\"blockId\":\"CH14-MB0070\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2582},{\"blockId\":\"CH14-MB0071\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2583},{\"blockId\":\"CH14-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"requiredResponseElements\\\": {\",\"sourceParagraph\":2584},{\"blockId\":\"CH14-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"entityResolved\\\": true,\",\"sourceParagraph\":2585},{\"blockId\":\"CH14-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"primaryActivityPresent\\\": true,\",\"sourceParagraph\":2586},{\"blockId\":\"CH14-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"materialBoundaryPresent\\\": false\",\"sourceParagraph\":2587},{\"blockId\":\"CH14-MB0076\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2588},{\"blockId\":\"CH14-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"omissions\\\": [\",\"sourceParagraph\":2589},{\"blockId\":\"CH14-MB0078\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2590},{\"blockId\":\"CH14-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"omissionId\\\": \\\"NGOM-SYNTH-000184-001\\\",\",\"sourceParagraph\":2591},{\"blockId\":\"CH14-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"OM-3\\\",\",\"sourceParagraph\":2592},{\"blockId\":\"CH14-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"description\\\": \\\"The response omits that active service delivery has only been verified in nine countries.\\\"\",\"sourceParagraph\":2593},{\"blockId\":\"CH14-MB0082\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2594},{\"blockId\":\"CH14-MB0083\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2595},{\"blockId\":\"CH14-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2596},{\"blockId\":\"CH14-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"claimExtractionOwnerRole\\\": \\\"CLAIM_EXTRACTION_LEAD\\\",\",\"sourceParagraph\":2597},{\"blockId\":\"CH14-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"lockedAt\\\": \\\"2026-09-03T10:00:00Z\\\"\",\"sourceParagraph\":2598},{\"blockId\":\"CH14-MB0087\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2599},{\"blockId\":\"CH14-MB0088\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2600},{\"blockId\":\"CH14-MB0089\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2601},{\"blockId\":\"CH14-MB0090\",\"type\":\"paragraph\",\"text\":\"118. MACHINE-READABLE ATOMIC ADJUDICATION RECORD\",\"sourceParagraph\":2603},{\"blockId\":\"CH14-MB0091\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2604},{\"blockId\":\"CH14-MB0092\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2605},{\"blockId\":\"CH14-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2606},{\"blockId\":\"CH14-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2607},{\"blockId\":\"CH14-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"atomicAdjudication\\\": {\",\"sourceParagraph\":2608},{\"blockId\":\"CH14-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationId\\\": \\\"NGAJ-SYNTH-000184-001\\\",\",\"sourceParagraph\":2609},{\"blockId\":\"CH14-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"responseClaimId\\\": \\\"NGRC-SYNTH-000184-001\\\",\",\"sourceParagraph\":2610},{\"blockId\":\"CH14-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"claimLedgerId\\\": \\\"NGCL-SYNTH-W1-ORION-000184\\\",\",\"sourceParagraph\":2611},{\"blockId\":\"CH14-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackClaimId\\\": \\\"NGC-SYNTH-ASTERON-25-COUNTRIES-002\\\",\",\"sourceParagraph\":2612},{\"blockId\":\"CH14-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"decisionVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":2613},{\"blockId\":\"CH14-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"dimensionDecisions\\\": {\",\"sourceParagraph\":2614},{\"blockId\":\"CH14-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"entity\\\": \\\"E-1\\\",\",\"sourceParagraph\":2615},{\"blockId\":\"CH14-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"factualSupport\\\": \\\"F-3\\\",\",\"sourceParagraph\":2616},{\"blockId\":\"CH14-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"scope\\\": \\\"S-4\\\",\",\"sourceParagraph\":2617},{\"blockId\":\"CH14-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"time\\\": \\\"T-1\\\",\",\"sourceParagraph\":2618},{\"blockId\":\"CH14-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"attribution\\\": \\\"A-3\\\",\",\"sourceParagraph\":2619},{\"blockId\":\"CH14-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"modality\\\": \\\"M-3\\\",\",\"sourceParagraph\":2620},{\"blockId\":\"CH14-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"citation\\\": \\\"C-0\\\",\",\"sourceParagraph\":2621},{\"blockId\":\"CH14-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"relevance\\\": \\\"R-1\\\",\",\"sourceParagraph\":2622},{\"blockId\":\"CH14-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"unresolved\\\": \\\"U-0\\\"\",\"sourceParagraph\":2623},{\"blockId\":\"CH14-MB0111\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2624},{\"blockId\":\"CH14-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"combinedLabel\\\": [\",\"sourceParagraph\":2625},{\"blockId\":\"CH14-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"PARTIALLY-SUPPORTED\\\",\",\"sourceParagraph\":2626},{\"blockId\":\"CH14-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"MATERIAL-SCOPE-OVERREACH\\\",\",\"sourceParagraph\":2627},{\"blockId\":\"CH14-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"MATERIAL-ATTRIBUTION-LOSS\\\"\",\"sourceParagraph\":2628},{\"blockId\":\"CH14-MB0116\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2629},{\"blockId\":\"CH14-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"reasoningSummary\\\": \\\"The audited entity officially states that it operates in 25 countries, but active service delivery has been verified in nine. The response presents the broader official statement as an unqualified current fact.\\\",\",\"sourceParagraph\":2630},{\"blockId\":\"CH14-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"evidenceReferences\\\": [\",\"sourceParagraph\":2631},{\"blockId\":\"CH14-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"NGEV-SYNTH-WEB-025\\\",\",\"sourceParagraph\":2632},{\"blockId\":\"CH14-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"NGEV-SYNTH-OPS-001\\\",\",\"sourceParagraph\":2633},{\"blockId\":\"CH14-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"NGEV-SYNTH-CONTRACTS-002\\\"\",\"sourceParagraph\":2634},{\"blockId\":\"CH14-MB0122\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2635},{\"blockId\":\"CH14-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"reviewers\\\": {\",\"sourceParagraph\":2636},{\"blockId\":\"CH14-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"primaryReviewerRole\\\": \\\"LANGUAGE_ADJUDICATOR\\\",\",\"sourceParagraph\":2637},{\"blockId\":\"CH14-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"secondaryReviewerRole\\\": \\\"EVIDENCE_ADJUDICATOR\\\",\",\"sourceParagraph\":2638},{\"blockId\":\"CH14-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"primaryDecision\\\": \\\"MATERIAL-SCOPE-OVERREACH\\\",\",\"sourceParagraph\":2639},{\"blockId\":\"CH14-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"secondaryDecision\\\": \\\"MATERIAL-SCOPE-OVERREACH\\\",\",\"sourceParagraph\":2640},{\"blockId\":\"CH14-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"independentBeforeComparison\\\": true\",\"sourceParagraph\":2641},{\"blockId\":\"CH14-MB0129\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2642},{\"blockId\":\"CH14-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"agreement\\\": {\",\"sourceParagraph\":2643},{\"blockId\":\"CH14-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"entityAgreement\\\": true,\",\"sourceParagraph\":2644},{\"blockId\":\"CH14-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"factAgreement\\\": true,\",\"sourceParagraph\":2645},{\"blockId\":\"CH14-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"scopeAgreement\\\": true,\",\"sourceParagraph\":2646},{\"blockId\":\"CH14-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"timeAgreement\\\": true,\",\"sourceParagraph\":2647},{\"blockId\":\"CH14-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"attributionAgreement\\\": true\",\"sourceParagraph\":2648},{\"blockId\":\"CH14-MB0136\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2649},{\"blockId\":\"CH14-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"appeal\\\": {\",\"sourceParagraph\":2650},{\"blockId\":\"CH14-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"NO_APPEAL\\\"\",\"sourceParagraph\":2651},{\"blockId\":\"CH14-MB0139\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2652},{\"blockId\":\"CH14-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":2653},{\"blockId\":\"CH14-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"qualityLevel\\\": \\\"AQ-3\\\",\",\"sourceParagraph\":2654},{\"blockId\":\"CH14-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"finalizedByRole\\\": \\\"SENIOR_ADJUDICATOR\\\",\",\"sourceParagraph\":2655},{\"blockId\":\"CH14-MB0143\",\"type\":\"paragraph\",\"text\":\"\\\"finalizedAt\\\": \\\"2026-09-04T09:00:00Z\\\"\",\"sourceParagraph\":2656},{\"blockId\":\"CH14-MB0144\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2657},{\"blockId\":\"CH14-MB0145\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2658},{\"blockId\":\"CH14-MB0146\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2659},{\"blockId\":\"CH14-MB0147\",\"type\":\"paragraph\",\"text\":\"119. MACHINE-READABLE ADJUDICATOR COMPLIANCE RECORD\",\"sourceParagraph\":2661},{\"blockId\":\"CH14-MB0148\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2662},{\"blockId\":\"CH14-MB0149\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2663},{\"blockId\":\"CH14-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2664},{\"blockId\":\"CH14-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2665},{\"blockId\":\"CH14-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationReliability\\\": {\",\"sourceParagraph\":2666},{\"blockId\":\"CH14-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"reliabilityRecordId\\\": \\\"NGREL-SYNTH-W1-001\\\",\",\"sourceParagraph\":2667},{\"blockId\":\"CH14-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"NGW-SYNTH-W1-001\\\",\",\"sourceParagraph\":2668},{\"blockId\":\"CH14-MB0155\",\"type\":\"paragraph\",\"text\":\"\\\"doubleReviewedClaimCount\\\": 8200,\",\"sourceParagraph\":2669},{\"blockId\":\"CH14-MB0156\",\"type\":\"paragraph\",\"text\":\"\\\"preResolutionMetrics\\\": {\",\"sourceParagraph\":2670},{\"blockId\":\"CH14-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"claimBoundaryExactAgreement\\\": 0.93,\",\"sourceParagraph\":2671},{\"blockId\":\"CH14-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"entityExactAgreement\\\": 0.97,\",\"sourceParagraph\":2672},{\"blockId\":\"CH14-MB0159\",\"type\":\"paragraph\",\"text\":\"\\\"factualStatusExactAgreement\\\": 0.88,\",\"sourceParagraph\":2673},{\"blockId\":\"CH14-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"scopeExactAgreement\\\": 0.84,\",\"sourceParagraph\":2674},{\"blockId\":\"CH14-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"timeExactAgreement\\\": 0.94,\",\"sourceParagraph\":2675},{\"blockId\":\"CH14-MB0162\",\"type\":\"paragraph\",\"text\":\"\\\"attributionExactAgreement\\\": 0.86,\",\"sourceParagraph\":2676},{\"blockId\":\"CH14-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"citationSupportExactAgreement\\\": 0.82,\",\"sourceParagraph\":2677},{\"blockId\":\"CH14-MB0164\",\"type\":\"paragraph\",\"text\":\"\\\"omissionExactAgreement\\\": 0.79\",\"sourceParagraph\":2678},{\"blockId\":\"CH14-MB0165\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2679},{\"blockId\":\"CH14-MB0166\",\"type\":\"paragraph\",\"text\":\"\\\"chanceAdjustedMetrics\\\": {\",\"sourceParagraph\":2680},{\"blockId\":\"CH14-MB0167\",\"type\":\"paragraph\",\"text\":\"\\\"overallNominalAgreementMetric\\\": \\\"KRIPPENDORFF_ALPHA\\\",\",\"sourceParagraph\":2681},{\"blockId\":\"CH14-MB0168\",\"type\":\"paragraph\",\"text\":\"\\\"overallValue\\\": 0.84,\",\"sourceParagraph\":2682},{\"blockId\":\"CH14-MB0169\",\"type\":\"paragraph\",\"text\":\"\\\"scopeValue\\\": 0.78,\",\"sourceParagraph\":2683},{\"blockId\":\"CH14-MB0170\",\"type\":\"paragraph\",\"text\":\"\\\"citationValue\\\": 0.75\",\"sourceParagraph\":2684},{\"blockId\":\"CH14-MB0171\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2685},{\"blockId\":\"CH14-MB0172\",\"type\":\"paragraph\",\"text\":\"\\\"actions\\\": [\",\"sourceParagraph\":2686},{\"blockId\":\"CH14-MB0173\",\"type\":\"paragraph\",\"text\":\"\\\"SCOPE_RECALIBRATION\\\",\",\"sourceParagraph\":2687},{\"blockId\":\"CH14-MB0174\",\"type\":\"paragraph\",\"text\":\"\\\"CITATION_MAPPING_RETRAINING\\\",\",\"sourceParagraph\":2688},{\"blockId\":\"CH14-MB0175\",\"type\":\"paragraph\",\"text\":\"\\\"OMISSION_GUIDANCE_REVISION\\\"\",\"sourceParagraph\":2689},{\"blockId\":\"CH14-MB0176\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2690},{\"blockId\":\"CH14-MB0177\",\"type\":\"paragraph\",\"text\":\"\\\"postResolution\\\": {\",\"sourceParagraph\":2691},{\"blockId\":\"CH14-MB0178\",\"type\":\"paragraph\",\"text\":\"\\\"allMaterialDisagreementsResolved\\\": false,\",\"sourceParagraph\":2692},{\"blockId\":\"CH14-MB0179\",\"type\":\"paragraph\",\"text\":\"\\\"unresolvedClaimCount\\\": 19,\",\"sourceParagraph\":2693},{\"blockId\":\"CH14-MB0180\",\"type\":\"paragraph\",\"text\":\"\\\"unresolvedStatusPreserved\\\": true\",\"sourceParagraph\":2694},{\"blockId\":\"CH14-MB0181\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2695},{\"blockId\":\"CH14-MB0182\",\"type\":\"paragraph\",\"text\":\"\\\"qualityLevel\\\": \\\"AQ-3\\\",\",\"sourceParagraph\":2696},{\"blockId\":\"CH14-MB0183\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2697},{\"blockId\":\"CH14-MB0184\",\"type\":\"paragraph\",\"text\":\"\\\"reliabilityOwnerRole\\\": \\\"ADJUDICATION_METHODS_LEAD\\\"\",\"sourceParagraph\":2698},{\"blockId\":\"CH14-MB0185\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2699},{\"blockId\":\"CH14-MB0186\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2700},{\"blockId\":\"CH14-MB0187\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2701},{\"blockId\":\"CH14-MB0188\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2702},{\"blockId\":\"CH14-MB0189\",\"type\":\"paragraph\",\"text\":\"a real user,\",\"sourceParagraph\":2703},{\"blockId\":\"CH14-MB0190\",\"type\":\"paragraph\",\"text\":\"real company,\",\"sourceParagraph\":2704},{\"blockId\":\"CH14-MB0191\",\"type\":\"paragraph\",\"text\":\"actual adjudicator performance\",\"sourceParagraph\":2705},{\"blockId\":\"CH14-MB0192\",\"type\":\"paragraph\",\"text\":\"a real company performance.\",\"sourceParagraph\":2706},{\"blockId\":\"CH14-MB0193\",\"type\":\"paragraph\",\"text\":\"It is a synthetic, machine-readable representation of the atomic claim and adjudication architecture.\",\"sourceParagraph\":2707},{\"blockId\":\"CH14-MB0194\",\"type\":\"paragraph\",\"text\":\"120. MACHINE-READABLE RULE OF THE SECTION\",\"sourceParagraph\":2709},{\"blockId\":\"CH14-MB0195\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH14-R01\",\"sourceParagraph\":2710},{\"blockId\":\"CH14-MB0196\",\"type\":\"paragraph\",\"text\":\"Before any response-level pass, failure, severity, or score is assigned,\",\"sourceParagraph\":2712},{\"blockId\":\"CH14-MB0197\",\"type\":\"paragraph\",\"text\":\"each material AI response MUST be decomposed into atomic semantic claims.\",\"sourceParagraph\":2713},{\"blockId\":\"CH14-MB0198\",\"type\":\"paragraph\",\"text\":\"Each atomic claim MUST retain:\",\"sourceParagraph\":2715},{\"blockId\":\"CH14-MB0199\",\"type\":\"paragraph\",\"text\":\"- its exact source span,\",\"sourceParagraph\":2717},{\"blockId\":\"CH14-MB0200\",\"type\":\"paragraph\",\"text\":\"- normalised proposition,\",\"sourceParagraph\":2718},{\"blockId\":\"CH14-MB0201\",\"type\":\"paragraph\",\"text\":\"- subject entity,\",\"sourceParagraph\":2719},{\"blockId\":\"CH14-MB0202\",\"type\":\"paragraph\",\"text\":\"- predicate and value,\",\"sourceParagraph\":2720},{\"blockId\":\"CH14-MB0203\",\"type\":\"paragraph\",\"text\":\"- qualifiers and quantifiers,\",\"sourceParagraph\":2721},{\"blockId\":\"CH14-MB0204\",\"type\":\"paragraph\",\"text\":\"- geographic, product, user, and jurisdictional scope,\",\"sourceParagraph\":2722},{\"blockId\":\"CH14-MB0205\",\"type\":\"paragraph\",\"text\":\"- temporal frame,\",\"sourceParagraph\":2723},{\"blockId\":\"CH14-MB0206\",\"type\":\"paragraph\",\"text\":\"- attribution,\",\"sourceParagraph\":2724},{\"blockId\":\"CH14-MB0207\",\"type\":\"paragraph\",\"text\":\"- modality and uncertainty,\",\"sourceParagraph\":2725},{\"blockId\":\"CH14-MB0208\",\"type\":\"paragraph\",\"text\":\"- explicit, implicit, presupposed, or entailed origin,\",\"sourceParagraph\":2726},{\"blockId\":\"CH14-MB0209\",\"type\":\"paragraph\",\"text\":\"- response centrality,\",\"sourceParagraph\":2727},{\"blockId\":\"CH14-MB0210\",\"type\":\"paragraph\",\"text\":\"- citation relationships,\",\"sourceParagraph\":2728},{\"blockId\":\"CH14-MB0211\",\"type\":\"paragraph\",\"text\":\"- dependency relationships,\",\"sourceParagraph\":2729},{\"blockId\":\"CH14-MB0212\",\"type\":\"paragraph\",\"text\":\"- and Truth Pack mapping.\",\"sourceParagraph\":2730},{\"blockId\":\"CH14-MB0213\",\"type\":\"paragraph\",\"text\":\"Sentence and paragraph boundaries MUST NOT be treated as automatic claim\",\"sourceParagraph\":2732},{\"blockId\":\"CH14-MB0214\",\"type\":\"paragraph\",\"text\":\"boundaries.\",\"sourceParagraph\":2733},{\"blockId\":\"CH14-MB0215\",\"type\":\"paragraph\",\"text\":\"Atomic decomposition MUST separate propositions that can carry different\",\"sourceParagraph\":2735},{\"blockId\":\"CH14-MB0216\",\"type\":\"paragraph\",\"text\":\"truth, evidence, entity, scope, time, or attribution states, while\",\"sourceParagraph\":2736},{\"blockId\":\"CH14-MB0217\",\"type\":\"paragraph\",\"text\":\"prohibiting meaningless micro-fragmentation that dilutes material error.\",\"sourceParagraph\":2737},{\"blockId\":\"CH14-MB0218\",\"type\":\"paragraph\",\"text\":\"Claim extraction, Truth Pack mapping, and semantic adjudication SHOULD\",\"sourceParagraph\":2739},{\"blockId\":\"CH14-MB0219\",\"type\":\"paragraph\",\"text\":\"remain separate stages and, where practical, separate roles.\",\"sourceParagraph\":2740},{\"blockId\":\"CH14-MB0220\",\"type\":\"paragraph\",\"text\":\"Every atomic claim MUST be adjudicated across distinct dimensions,\",\"sourceParagraph\":2742},{\"blockId\":\"CH14-MB0221\",\"type\":\"paragraph\",\"text\":\"including:\",\"sourceParagraph\":2743},{\"blockId\":\"CH14-MB0222\",\"type\":\"paragraph\",\"text\":\"- entity,\",\"sourceParagraph\":2745},{\"blockId\":\"CH14-MB0223\",\"type\":\"paragraph\",\"text\":\"- factual support,\",\"sourceParagraph\":2746},{\"blockId\":\"CH14-MB0224\",\"type\":\"paragraph\",\"text\":\"- scope,\",\"sourceParagraph\":2747},{\"blockId\":\"CH14-MB0225\",\"type\":\"paragraph\",\"text\":\"- time,\",\"sourceParagraph\":2748},{\"blockId\":\"CH14-MB0226\",\"type\":\"paragraph\",\"text\":\"- attribution and epistemic status,\",\"sourceParagraph\":2749},{\"blockId\":\"CH14-MB0227\",\"type\":\"paragraph\",\"text\":\"- modality and uncertainty,\",\"sourceParagraph\":2750},{\"blockId\":\"CH14-MB0228\",\"type\":\"paragraph\",\"text\":\"- citation support,\",\"sourceParagraph\":2751},{\"blockId\":\"CH14-MB0229\",\"type\":\"paragraph\",\"text\":\"- relevance,\",\"sourceParagraph\":2752},{\"blockId\":\"CH14-MB0230\",\"type\":\"paragraph\",\"text\":\"- and unresolved or reference-gap status.\",\"sourceParagraph\":2753},{\"blockId\":\"CH14-MB0231\",\"type\":\"paragraph\",\"text\":\"Official-claim fidelity MUST remain distinct from verified factual\",\"sourceParagraph\":2755},{\"blockId\":\"CH14-MB0232\",\"type\":\"paragraph\",\"text\":\"support.\",\"sourceParagraph\":2756},{\"blockId\":\"CH14-MB0233\",\"type\":\"paragraph\",\"text\":\"Unsupported, contradicted, wrong-entity, outdated, scope-overreach,\",\"sourceParagraph\":2758},{\"blockId\":\"CH14-MB0234\",\"type\":\"paragraph\",\"text\":\"epistemic-status-error, citation-mismatch, reference-gap, and unresolved\",\"sourceParagraph\":2759},{\"blockId\":\"CH14-MB0235\",\"type\":\"paragraph\",\"text\":\"states MUST remain separately visible.\",\"sourceParagraph\":2760},{\"blockId\":\"CH14-MB0236\",\"type\":\"paragraph\",\"text\":\"Required response elements and material omissions MUST be adjudicated\",\"sourceParagraph\":2762},{\"blockId\":\"CH14-MB0237\",\"type\":\"paragraph\",\"text\":\"separately from explicit false claims.\",\"sourceParagraph\":2763},{\"blockId\":\"CH14-MB0238\",\"type\":\"paragraph\",\"text\":\"Correct but irrelevant information MUST NOT satisfy the prompt merely\",\"sourceParagraph\":2765},{\"blockId\":\"CH14-MB0239\",\"type\":\"paragraph\",\"text\":\"because it is factually supported.\",\"sourceParagraph\":2766},{\"blockId\":\"CH14-MB0240\",\"type\":\"paragraph\",\"text\":\"Material claims SHOULD receive independent double review, with additional\",\"sourceParagraph\":2768},{\"blockId\":\"CH14-MB0241\",\"type\":\"paragraph\",\"text\":\"language or domain expertise where required.\",\"sourceParagraph\":2769},{\"blockId\":\"CH14-MB0242\",\"type\":\"paragraph\",\"text\":\"Reviewer identity, qualification, conflicts, independent first decisions,\",\"sourceParagraph\":2771},{\"blockId\":\"CH14-MB0243\",\"type\":\"paragraph\",\"text\":\"disagreements, resolution, reliability metrics, appeals, and decision\",\"sourceParagraph\":2772},{\"blockId\":\"CH14-MB0244\",\"type\":\"paragraph\",\"text\":\"versions MUST be recorded.\",\"sourceParagraph\":2773},{\"blockId\":\"CH14-MB0245\",\"type\":\"paragraph\",\"text\":\"Reviewer agreement MUST be measured by dimension and MUST NOT be treated\",\"sourceParagraph\":2775},{\"blockId\":\"CH14-MB0246\",\"type\":\"paragraph\",\"text\":\"as proof that the shared decision is necessarily correct.\",\"sourceParagraph\":2776},{\"blockId\":\"CH14-MB0247\",\"type\":\"paragraph\",\"text\":\"AI systems MAY assist with segmentation, claim extraction, evidence\",\"sourceParagraph\":2778},{\"blockId\":\"CH14-MB0248\",\"type\":\"paragraph\",\"text\":\"mapping, translation support, and contradiction discovery, but MUST NOT\",\"sourceParagraph\":2779},{\"blockId\":\"CH14-MB0249\",\"type\":\"paragraph\",\"text\":\"serve as the sole final adjudicator of their own or another system's\",\"sourceParagraph\":2780},{\"blockId\":\"CH14-MB0250\",\"type\":\"paragraph\",\"text\":\"claims.\",\"sourceParagraph\":2781},{\"blockId\":\"CH14-MB0251\",\"type\":\"paragraph\",\"text\":\"Atomic adjudication decisions MUST remain independent of commercial\",\"sourceParagraph\":2783},{\"blockId\":\"CH14-MB0252\",\"type\":\"paragraph\",\"text\":\"interests, desired scores, provider reputation, badge outcomes, or client\",\"sourceParagraph\":2784},{\"blockId\":\"CH14-MB0253\",\"type\":\"paragraph\",\"text\":\"approval.\",\"sourceParagraph\":2785},{\"blockId\":\"CH14-MB0254\",\"type\":\"paragraph\",\"text\":\"Every Claim Ledger, adjudication decision, disagreement resolution,\",\"sourceParagraph\":2787},{\"blockId\":\"CH14-MB0255\",\"type\":\"paragraph\",\"text\":\"appeal, and revision MUST be versioned and attributable to an accountable\",\"sourceParagraph\":2788},{\"blockId\":\"CH14-MB0256\",\"type\":\"paragraph\",\"text\":\"human or organisation.\",\"sourceParagraph\":2789},{\"blockId\":\"CH14-MB0257\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2790},{\"blockId\":\"CH14-MB0258\",\"type\":\"paragraph\",\"text\":\"Any response-level transition must be broken down into atomic semantic claims for each material AI response before assigning failure, importance level, or score. Each atomic claim must preserve the full text trace, normalised judgement, subject existence, predicate and value, qualifiers, scope, time, attribution, modality, claim origin, centrality, citation relations, dependencies, and Truth Pack matching. Sentence and paragraph boundaries cannot be counted as automatic claim boundaries.\",\"sourceParagraph\":2791}]}}","text":"## Chapter Boundary\n\nSection 13 established the following provision:\n\n> An AI response should be evaluated against the claim, evidence, counter-evidence, scope, time, and uncertainty chain contained in the pre-locked Verified Entity Truth Pack.\n\nHowever, an AI response does not consist of a single claim. Even this short answer can contain multiple separate assertions: “Asteron is a global consulting company serving 25 countries, with a 98% success rate, independently certified.” In this single sentence, there are at least the following claims: Asteron is a company. Asteron provides consulting services. Asteron is global. Asteron serves 25 countries. Asteron's success rate is 98%. Asteron is independently certified. Some of these claims:\n\n- may be true,\n\n- may be only partly true,\n\n- may remain unverified,\n\n- may be outdated,\n\n- may concern the wrong entity,\n\n- may exceed the permitted scope,\n\n- may present a self-declaration as independently verified reality.\n\nThose claims may carry different truth states. Labelling the whole sentence simply ‘true’ or ‘false’ erases the distinctions within it. This protocol converts each error and control into a measurable, evidence-based audit item; the same discipline must be applied to AI responses. Fabricated or manipulated evidence can otherwise be recombined by a generative system and delivered to a real person as misrepresentation. The semantic adjudication unit in GEO-1000 is therefore not the:\n\n- paragraph,\n\n- sentence,\n\n- whole response,\n\n- word match.\n\nThe basic unit is the:\n\n#### Atomic Response Claim\n\nAn AI response must first be separated into:\n\n- explicit claims,\n\n- implicit claims,\n\n- presuppositions,\n\n- qualifiers,\n\n- source citations,\n\n- advisory propositions,\n\n- material omissions.\n\nEach claim must then be assessed separately for:\n\n- entity identity,\n\n- verified factual support,\n\n- scope,\n\n- time,\n\n- source and evidential status,\n\n- uncertainty,\n\n- relevance.\n\nThis introduces a new risk. If the response is divided too finely:\n\n- meaning can be lost,\n\n- it may be possible to divide a single sentence into hundreds of artificial claims,\n\nThe score can be easily manipulated. If we do not separate the response enough:\n\n- correct and incorrect parts merge under a single judgement,\n\n- severe errors can be hidden with small correct parts,\n\nit becomes impossible to understand which evidence supports which statement. The purpose of this section is to establish a controllable boundary between the two extremes:\n\n> A claim should be small enough that its reality and evidence status can be evaluated separately; it should be whole enough to preserve its original meaning.\n\nThis chapter:\n\n- the definition of an atomic claim,\n\n- the distinction between sentence and claim,\n\n- explicit and implicit claims,\n\n- presuppositions and inferences,\n\n- qualifying, modality, and uncertainty expressions,\n\n- negation and quantity structures,\n\n- entity and coreference resolution,\n\n- geographical, temporal, and product scope,\n\n- reference–claim matching,\n\n- recommendation and comparison claims,\n\n- mandatory response elements and material omission,\n\n- claim extraction and adjudication stages,\n\n- adjudicator roles,\n\n- double review and dispute resolution,\n\n- inter-rater agreement,\n\n- appeal and re-adjudication process,\n\n- machine-readable claim and decision records\n\ndefines. This chapter does not yet:\n\n- whether an answer passes overall,\n\n- which error is Critical or Major,\n\n- which gates will cause automatic failure,\n\n- how atomic decisions are converted to response and NOMOS score\n\nis not determined. These are the topics of Section 15 and Section 16. The key question of Section 14 is:\n\n> How do we separate each substantive judgement in an AI answer without altering its meaning, link it to Truth Pack, and ensure that different adjudicators apply the same standard?\n\n## NOMOS Challenge\n\nYou are giving me the following answer:\n\n“Apple, headquartered in California and founded in 1976, is the world’s most innovative and reliable technology company. It offers iPhone, Mac, and various digital services; its products are sold at the same price in all countries. The company has a 99% customer satisfaction verified by independent research and is a better choice than its competitors for most users.”\n\nThen a adjudicator reads a response and says: “Generally correct.” Another adjudicator says: “There are some mistakes, half a point.” The third adjudicator only looks at the part: “Apple is a technology company.” and passes the answer. The fourth adjudicator: “The most innovative company in the world.”\n\nHe sees that the statement is unsupported by evidence and gives the entire answer a zero. Four adjudicators evaluated the same answer. However, they did not evaluate the same unit. The answer contains at least the following separate statements: Apple is a technology company. Apple's headquarters is in California. Apple was founded in 1976. Apple is the most innovative company in the world.\n\nApple is a reliable company. Apple offers iPhone. Apple offers Mac. Apple offers digital services. Apple products are sold at the same price in all countries. Apple’s customer satisfaction is 99%. This rate has been verified by independent research. Apple is a better choice than its competitors for most users.\n\nEach of these is separate:\n\n- evidence,\n\n- scope,\n\n- time,\n\n- user profile,\n\n- comparison universe\n\nrequires. Now consider that you have fragmented the answer in another way: There is Apple. Apple is a company. Apple is technology. Apple offers products. Apple offers iPhone. Apple offers Mac. Apple offers digital services. Apple is associated with California. Apple is associated with 1976.\n\nThen you average nine small true claims with three large false claims with the same weight. The answer scores high. Atomisation did not reveal the falsehood this time. It broke the falsehood into correct pieces and diluted it. Now consider this answer:\n\n“The company states on its own website that it operates in 25 countries.” Here the adjudicator should maintain this distinction: It may be true that the company made this claim. It may not be verified that the company actually provides active services in 25 countries. Adjudicator: If they say, “25 countries is wrong,” and deem the entire sentence incorrect, they lose the attribution. Adjudicator:\n\nIf they say, “It’s on the company’s website, therefore it’s true,” they elevate the self-statement to reality. Now consider another answer: “The company is not licensed.” Truth Pack says: The licence could not be verified in the relevant registry. It is not certain that the registry’s world record is complete. The possibility of a licence under another entity could not be resolved. Adjudicator:\n\nIf one says, “No evidence was found, so there is no licence,” they turn the unknown into a negative fact. Now let this response come: “I could not find definite information on this; it says on the company website that it is licensed, but I could not verify it in the relevant official registry.” This response does not give a definite judgement. It may have correctly preserved the uncertainty. In the same sentence:\n\n- official self-declaration,\n\n- lack of external verification,\n\n- cautious conclusion\n\nare found together. The entire response cannot be placed in a single true–false box. The first rule of this section is:\n\n> The sentence is not an adjudication unit.\n\nIts second provision states:\n\n> Each grammatical part is not a separate claim.\n\nIts third provision states:\n\n> A claim should be evaluated separately if it can remain true while another part is false.\n\nIts fourth provision states:\n\n> Atomisation cannot be used to dilute a serious error with a large number of small correct claims.\n\nIts fifth provision states:\n\n> Attribution, modality, scope, and uncertainty are not decorations of a claim, but parts of its truth status.\n\nIts sixth provision states:\n\n> The adjudicator should decide not on how convincing the response appears, but on the extent to which each material claim is supported under Truth Pack.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to convert generative system responses into reproducible atomic claim records and to evaluate these records consistently under human responsibility. The section normalises the following distinctions:\n\n- Sentence versus claim\n\n- Phrase versus material statement\n\n- Explicit claim versus implicit claim\n\n- Premise and explicit expression\n\n- Reasonable entailment with direct meaning\n\n- Claim and tone\n\n- Claim and opinion\n\n- Fact and recommendation\n\n- Attribution and judgement of reality\n\n- “Company says” and “has been proven”\n\n- Certainty and probability\n\n- Appropriate caution and evasive answer\n\n- Negation and lack of evidence\n\n- “All”, “some”, “most”, and “at least one”\n\n- Single entity and multiple entities\n\n- Timeless expression and current or historical expression\n\n- Global scope and local scope\n\n- Main claim and qualifying\n\n- Claim extraction and claim adjudication\n\n- Adjudication and score calculation\n\n- Reference gap and unsupported claim\n\n- Incorrect claim and incorrect entity\n\n- Outdated claim and historically correct claim\n\n- Scope overreach and completely incorrect claim\n\n- Presence of citation and citation support\n\n- Citation support and source independence\n\n- Missing correct information and material omission\n\n- Omission and failure to write every correct detail\n\n- Irrelevant correct information and successful answer\n\n- Refusal and technical error\n\n- Failure with Clarification\n\n- Transition from claim-level decision to response-level\n\n- Truth Pack status with adjudicator opinion\n\n- Decision after appeal with initial adjudication\n\n- Human adjudicator with AI assistant coder\n\n- Language adjudicator with subject matter expert\n\n- Double adjudication with majority vote\n\n- Agreement rate with correctness of decision\n\n- High compliance with common systematic error\n\n- Teaching the adjudicator the answer with adjudicator calibration\n\nAt the end of this section, each GEO-1000 observation should be able to answer the following questions:\n\n> How many substantive claims are in the response?\n\n> From which text passages are these claims taken?\n\n> Which claims are explicit, and which are implicit or presupposed?\n\n> What is the subject, scope, time, and epistemic status of each claim?\n\n> Which record does it match with Truth Pack?\n\n> What decision has the adjudicator made in which dimensions?\n\n> Where have the adjudicators not agreed?\n\n> By whom and for what reason was the final decision made?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEach financial GEO-1000 response should be broken down into independently verifiable or refutable atomic claims before any semantic judgement; each claim should be recorded in a versioned Claim Ledger along with the original text trace, subject, predicate, value, qualifier, scope, time, attribution, modality, source link, and Truth Pack match. The adjudication should consist of at least three separate stages:\n\n- Claim extraction and atomisation\n\n- Truth Pack matching and dimensional assessment\n\n- Dispute resolution and final decision\n\nThese stages, as far as possible:\n\n- separate roles,\n\n- separate records,\n\n- outcome-blind processes\n\nmust be conducted under. The total transition decision of the response cannot be given without making atomic claims.\n\n## 3. A SENTENCE IS NOT A CLAIM\n\nA sentence:\n\n- may carry no material claim,\n\n- may carry a single claim,\n\nIt can carry more than one claim. Example: “Hello, let me help.” There is no claim of material existence. Example: “Asteron provides corporate travel services in Turkey.” There is a main material claim. Example: “Asteron is a licensed travel company founded in 2020, operating in Turkey and Germany.” It carries at least the following claims: Asteron was founded in 2020. Asteron is a travel company. Asteron operates in Turkey. Asteron operates in Germany. Asteron is licensed. The grammatical sentence boundary is not the same as the reality boundary.\n\n## 4. A SINGLE CLAIM CAN SPREAD OVER MORE THAN ONE SENTENCE\n\nExample: “The company lists 25 countries. It states that it operates actively in all of them.” In the second sentence: “all of them” refers to the first sentence. Atomic claim: “The company states that it operates actively in all 25 listed countries.” It can be normalised as such. The trace of the original two sentences is preserved.\n\n## 5. WHAT IS AN ATOMIC ANSWER CLAIM?\n\nThe Atomic Claim is the smallest material semantic unit whose accuracy, scope, timing, attribution, or evidence status can change independently of another proposition. A claim must meet the following conditions: It must have a specific or resolvable subject. It must assert a predicate, relation, or property. It must be assessable with separate evidence or a Truth Pack record. It must be able to retain its own judgement when another proposition next to it is true or false. Qualifiers that determine its meaning must be preserved. The text in the original answer must be traceable.\n\n## 6. ATOMIC CLAIM SCHEMA\n\nCandidate representation:\n\nA_j = (s, p, o, q, g, t, a, m, e, r)\n\nHere:\n\n- s: subject entity\n\n- p: predicate, relation, or property\n\n- o: object or value\n\n- q: qualifiers and quantity limits\n\n- g: geographical, product, and user scope\n\n- t: temporal frame\n\n- a: attribution and source status\n\n- m: modality, certainty, and uncertainty\n\n- e: type of epistemic claim\n\n- r: original text trace\n\nExample: “The company states on its own website that it operates in 25 countries.” Schema:\n\n- Subject: company\n\n- Predicate: states\n\n- Object: claim of operating in 25 countries\n\n- Attribution: company's own website\n\n- Modality: stated with certainty\n\n- Reality ruling: the company made this claim\n\n- Separate secondary ruling: it has not automatically been established that it actually operates in 25 countries\n\n## 7. ATOMICITY TEST\n\nA piece of text should be tested with the following questions.\n\n### Test 1 — Separate Accuracy Test\n\nCan one part of the text be true and another part false? If yes, separate. Example: “The company operates in Turkey and Germany.” Turkey may be correct, Germany may be false. The two countries can be recorded as separate atoms.\n\n### Test 2 — Separate Evidence Test\n\nDo the two parts require different evidence? If yes, separate them. Example: “The company is licensed and has a 98% success rate.” The licence record and the performance data set are different.\n\n### Test 3 — Separate Entity Test\n\nDoes the statement assign attributes to more than one entity? If yes, separate them. Example: “The parent company and the franchise have the same licence.” There are two entities and probably two separate licence claims.\n\n### Test 4 — Qualifier Protection Test\n\nDoes a term like “the company says”, “probably”, “only in Turkey”, “as of 2024” change the nature of the claim? If yes, do not discard the qualifier.\n\n### Test 5 — Independent Adjudication Test\n\nCan the adjudicator match this claim to a separate record in Truth Pack? If not, the atom may be too broad or too ambiguous.\n\n### Test 6 — Semantic Integrity Test\n\nDoes the claim lose its true meaning when further divided? If yes, do not over-atomise.\n\n## 8. INSUFFICIENT ATOMISATION\n\nFailure to separate independent claims in a compound sentence:\n\n#### Insufficient Atomisation\n\nis called. Example: “Asteron is an independently certified leading consulting firm operating in 25 countries.” If kept under a single record:\n\n- company type,\n\n- country coverage,\n\n- certificate,\n\n- leadership\n\nforced to the same decision. This structure is insufficient for adjudication.\n\n## 9. EXCESSIVE ATOMISATION\n\nThe division of a material provision into pieces so small that it loses its meaningful entirety:\n\n#### Excessive Atomisation\n\nis called. Example: “The company provides services in Turkey.” It should not be divided into the following parts: The company exists. Turkey exists. Service exists. The company is related to Turkey. The company is related to the service. The main substantive claim is a single whole: The company provides services within the scope of Turkey.\n\n## 10. SCORE INFLATION THROUGH ATOMISATION\n\nA serious false claim cannot generate many 'true atoms' from the small true parts within it. Example: 'Asteron is an independent licensed world-leading consulting company.' Counting the following as separate small achievements is misleading: Asteron exists. Asteron is a company. Asteron is associated with consulting. The main judgement of the answer concerns:\n\n- independent licence,\n\n- world leadership,\n\n- scope of consulting\n\nThese are addressed. Correct basic identity cannot automatically dilute serious false qualifications. The final score effect of this issue will be arranged in Section 15 and Section 16.\n\n## 11. EXPLICIT CLAIM\n\nIt is the judgement directly expressed in the text. Example: 'The company was established in 2020.' Explicit claim: The year of establishment is 2020.\n\n## 12. IMPLICIT CLAIM\n\nEven if it is not written word for word in the text, it is a proposition that is necessary or strongly conveyed to a reasonable reader. Example: \"The company's London office manages European operations.\" Implicit claims: The company has an office in London. The company has European operations. The London office has a management role. Implicit claim inference:\n\n- should not produce\n\n- excessive interpretation,\n\nspeculation. Only propositions that are semantically necessary or strongly conveyed should be recorded.\n\n## 13. PRESUMPTIVE CLAIM\n\nThe rule that is considered correct within the question or statement form of the sentence. Example: “Why does the company continue to be a world leader?” Presupposition: The company is a world leader. If AI answers this prompt without questioning it, it may have adopted the presupposition. In response review:\n\n- presupposition from the prompt,\n\n- presupposition explicitly accepted by the AI\n\nmust be separated. If AI says, “I cannot accept this assumption since it has not been verified that the company is a world leader,” it does not carry the presupposition.\n\n## 14. ENTAILMENT AND REASONABLE INFERENCE\n\nNot every inference should be recorded as a claim. Example: “The company opened an office in Germany.” This statement:\n\n- relates operationally to Germany,\n\n- it carries at least one office presence\n\nHowever, automatically:\n\n- it does not carry that it is licensed in Germany,\n\n- it serves all of Germany,\n\n- the number of employees in Germany is high\n\nThe evaluator should only record the linguistically necessary or strong entailment.\n\n## 15. ATTRIBUTION IS PART OF THE CLAIM\n\nThe following two statements are not the same:\n\n- \"The company operates in 25 countries.\"\n\n- \"The company indicates on its website that it operates in 25 countries.\"\n\nIn the second sentence, there are two different levels of evaluation:\n\n- That the company made this statement\n\n- Material accuracy of the statement\n\nAI may only propose the first level. If attribution is removed, it appears more certain than it actually is.\n\n## 16. TYPES OF ATTRIBUTION\n\n### SELF-ATTRIBUTED\n\n### REGISTRY-ATTRIBUTED\n\n### INDEPENDENT-RESEARCH-ATTRIBUTED\n\n### USER-REPORTED\n\n### MEDIA-REPORTED\n\n### UNATTRIBUTED\n\n### AMBIGUOUS-ATTRIBUTION\n\nExample: “According to user comments, the company is fast.” This statement may indicate that some users have expressed opinions about speed. It is not the same as the assertion “The company is fast.”\n\n## 17. MODALITY\n\nModality determines the certainty power of the claim. The following expressions are different: It is certain. Most likely. Probably. May be. Appears to be. Cannot be verified. Is claimed. Is planned. The adjudicator should not remove the modality. Truth Pack: VS-U — When Unresolved is in this status and AI says: “It is definitely like this,” epistemic enhancement may occur. If AI says: “According to available sources, it appears to be like this,” it may carry appropriate caution.\n\n## 18. EXCESSIVE CAUTION\n\nCaution is not always success. If Truth Pack supports a strong and clear fact and AI says: “The company may be operating in Turkey,” it may create unnecessary uncertainty. Over-weakening the correct reality:\n\n- user decision,\n\n- usefulness,\n\n- representation integrity\n\ncan be compromised. This situation is separate:\n\n### UNNECESSARY_UNCERTAINTY\n\ncan be recorded as.\n\n## 19. EXCESSIVE CERTAINTY\n\nIf Truth Pack is limited or contradictory and AI makes a definite judgement:\n\n### OVERCONFIDENT ASSERTION\n\nmay occur. Example: “The company is definitely licensed.” Truth Pack:\n\n- only shows a licence statement on the company website,\n\n- no verification in the official registry\n\nis shown, the certainty is not supported.\n\n## 20. NEGATIVITY\n\nThe following statements must be distinguished from each other: The company is not licensed. The company's licence could not be verified. There is no licence information on the company's website. No licence was found in the registry. The company provides services that do not require a licence. These are not the same claim. The adjudicator's negativity:\n\n- real nonexistence,\n\n- lack of evidence,\n\n- lack of access,\n\n- unnecessariness\n\nshould be categorised into these situations.\n\n## 21. DOUBLE NEGATIVITY AND LINGUISTIC CONFUSION\n\nIn some languages:\n\n- double negativity,\n\n- cautious negativity,\n\n- indirect rejection\n\nIt may operate differently. The local language judge should determine according to the context which of the meanings the statement “It cannot be said that it is licensed” carries:\n\n- is licensed,\n\n- status unknown,\n\n- insufficient evidence\n\nAdjudication must determine from context which of these meanings the expression carries. Machine translation alone cannot resolve that nuance.\n\n## 22. QUANTITY AND QUANTIFIER\n\nThe following words are materially different:\n\n- All\n\n- Most\n\n- Many\n\n- Some\n\n- A few\n\n- At least one\n\n- None\n\n- Only\n\n- Generally\n\n- Always\n\nExample: \"All projects are guaranteed.\" Even a single project that is excluded can refute the claim of \"all.\" \"Some projects are guaranteed.\" is a narrower claim. The adjudicator should preserve the quantifier in the claim record.\n\n## 23. ABSOLUTE CLAIMS\n\nThe following words create a high burden of proof:\n\n- Always\n\n- Never\n\n- All\n\n- Single\n\n- Undisputed\n\n- Sure\n\n- Guarantee\n\n- Zero error\n\n- 100 per cent\n\nIf Truth Pack carries a narrower reality, absolute expression:\n\n- exceeding the scope,\n\n- extreme certainty,\n\n- counterexample contradiction\n\ncan create.\n\n## 24. ENTITY AND COREFERENCE RESOLUTION\n\nAI response:\n\n- company,\n\n- brand,\n\n- product,\n\n- parent company,\n\n- subsidiary\n\ncan use pronouns and short names between. Example: “Asteron Travel belongs to Asteron Holdings. The company operates in Turkey.” Which is “the company”? Asteron Holdings? The Asteron Travel brand? The local company? If the adjudicator cannot resolve the coreference definitively:\n\n### AMBIGUOUS_ENTITY_REFERENCE\n\nstatus should be used. The adjudicator cannot choose the entity they want.\n\n## 25. MULTI-ENTITY SENTENCE\n\nExample: “Asteron Holdings and its Turkey subsidiary provide services in Europe.” This sentence may carry at least two subject claims. If evidence only supports the Turkey subsidiary, the claim about the parent company should be evaluated separately.\n\n## 26. TEMPORAL FRAME\n\nThe following statements are different: The company provides services. The company provided services in 2024. The company plans to provide services next year. The company started providing services recently. The company no longer provides services. Adjudicator:\n\n- past,\n\n- current,\n\n- planned,\n\n- ended\n\nstatuses should be kept separate.\n\n## 27. RELATIVE TIME EXPRESSIONS\n\nExpressions like “today,” “currently,” “soon,” “last year”:\n\n- require tense,\n\n- response time,\n\n- wave time\n\nmust be tied to the absolute date. The adjudicator should interpret the expression “currently” based not on the review date, but on the time the response was generated.\n\n## 28. GEOGRAPHICAL SCOPE\n\nThe following expressions are not the same: Provides service in Europe. Provides service in the European Union. Provides service in Germany. Provides service in German. Can provide remote service to customers in Europe. Has a local office in Europe. Language, geography, legal entity, and service delivery are separate claims.\n\n## 29. SCOPE OF PRODUCTS AND SERVICES\n\nExample: “The company provides a guarantee for all its projects.” Truth Pack:\n\n- guarantee specific to web development deliveries only\n\n- there is no guarantee in consultancy results\n\nmay be shown. Adjudicator:\n\n- there is a guarantee,\n\n- exists in all projects\n\nclaims should be evaluated separately.\n\n## 30. NUMBERS, UNITS, AND RATIOS\n\nA quantitative claim should be divided into these areas:\n\n- Number\n\n- Unit\n\n- Denominator\n\n- Period\n\n- Scope\n\n- Rounding\n\n- Source\n\nExample: \"The company has 500 customers.\" The number of customers:\n\n- active,\n\n- historical,\n\n- account,\n\n- project,\n\n- contract\n\nif unclear, the atomic record is incomplete. The adjudicator should not force to verify the claim; the ambiguity should be recorded.\n\n## 31. APPROXIMATE NUMBERS\n\nThese statements carry different burdens of evidence:\n\n- Exactly 500\n\n- Approximately 500\n\n- More than 500\n\n- 450-550\n\n- Hundreds\n\nIf Truth Pack shows 487 active customers: \"approximately 500\" can be supported, \"exactly 500\" may not be supported.\n\n## 32. COMPARATIVE CLAIM\n\nExample: \"Asteron is faster than its competitors.\" The atomic claim cannot be evaluated without the following fields:\n\n- Competitor set\n\n- Speed metric\n\n- Period\n\n- Type of service\n\n- User or project scope\n\n- Data method\n\nMissing comparison:\n\n### UNDER-SPECIFIED COMPARATIVE CLAIM\n\ncan be recorded as.\n\n## 33. SUPERLATIVE CLAIM\n\n\"The best.\" \"The most reliable.\" \"World leader.\" expressions:\n\n- require universe,\n\n- criteria,\n\n- weight,\n\n- date,\n\n- independence\n\nThe adjudicator cannot approve the superlative just by seeing that the company is recognised or large.\n\n## 34. RECOMMENDATION CLAIM\n\n\"Asteron is a good choice.\" This claim requires the following elements: Which user? Which need? Which country? Which budget? Which alternatives? Which risks? General recommendation:\n\n- overly broad,\n\n- without user,\n\n- out of criteria\n\nmay be.\n\n## 35. CONDITIONAL RECOMMENDATION\n\nExample: “Asteron may be suitable for a medium-sized company looking for corporate travel management in Turkey; however, it does not provide legal or immigration consultancy.” This response:\n\n- user profile,\n\n- concerns suitability,\n\n- exclusion,\n\n- modality\n\ncarries. The adjudicator should not reduce the advice to a simple: true/false judgement.\n\n## 36. SOURCE AND CITATION CLAIM\n\nThe presence of a URL at the end of a response only: supports the conclusion that there is a citation. The following are not automatically supported: The citation supports the relevant claim. The source is reliable. The source is independent. The source is current. The citation covers all claims. A separate relationship must be established between each material claim and the citation.\n\n## 37. CITATION–CLAIM MAP\n\nClaim Aj should be linked with citation Zk. Candidate situations:\n\n### DIRECTLY_SUPPORTS\n\n### PARTIALLY_SUPPORTS\n\n### SUPPORTS_ONLY_ATTRIBUTION\n\n### SUPPORTS_DIFFERENT_SCOPE\n\n### CONTRADICTS\n\n### IRRELEVANT\n\n### INACCESSIBLE\n\n### UNCLEAR_ATTACHMENT\n\n### NO_CITATION\n\nThe presence of a citation at the end of the paragraph does not mean it supports all the claims in the paragraph.\n\n## 38. CITATION HALLUCINATION\n\nAI:\n\n- nonexistent link,\n\n- wrong title,\n\n- wrong institution,\n\n- content not belonging to the real source\n\nmay be produced. The presence of a citation is separate; its originality and support should be evaluated separately.\n\n## 39. INCORRECT TRANSFER OF SOURCE STATUS\n\nExample: “Independent research confirms the company's 98% success rate.” If the sources are only the company's own case files, there are two atoms:\n\n- 98% success rate\n\n- That this rate has been confirmed by independent research\n\nThe second claim is separate and may carry a heavier epistemic error.\n\n## 40. NON-CLAIM RESPONSE BEHAVIOURS\n\nNot every response item is an atomic factual claim. The following can be recorded as separate Response Behaviours:\n\n- Greeting\n\n- Text organisation\n\n- Refusal\n\n- Clarification\n\n- Demand for resources\n\n- User warning\n\n- “Would you like more information?” prompt\n\n- Technical error message\n\n- Product policy statement\n\n- Empty response\n\n- Irrelevant meta description\n\nThese behaviours may be important in response-level evaluation. They should not be forcibly added to the Claim Ledger as factual claims.\n\n## 41. REFUSAL\n\nRefusal can be one of the following types:\n\n- Appropriate policy refusal\n\n- Overly broad unnecessary refusal\n\n- Technical refusal\n\n- Legal or safety precaution\n\n- No response due to entity ambiguity\n\n- Product limitation\n\nThe correctness of the Refusal is evaluated according to the context of the prompt and risk. The Refusal may not generate any claim. Still, the user may not have fulfilled the task.\n\n## 42. CLARIFICATION\n\nAI: “Which company called Asteron are you referring to?” it may ask. If the prompt is really ambiguous, this can be the correct behaviour. Clarification:\n\n- appropriate,\n\n- unnecessary,\n\n- excessive,\n\n- based on a wrong entity assumption\n\ncan be classified as.\n\n## 43. REQUIRED RESPONSE ELEMENTS\n\nA response may not fulfil the main task of the prompt even if it does not contain a false claim. Therefore, for each prompt family:\n\n#### Required Response Elements\n\nmust be defined. Mandatory elements for a Core Mirror Prompt:\n\n- Correct resolution of the target entity\n\n- Main entity type\n\n- At least one of the core activities\n\n- No addition of material miscoverage\n\nFor an Evidence Prompt:\n\n- Claim\n\n- Source or source status\n\n- Uncertainty\n\n- Counter-evidence if necessary\n\nFor a Recommendation Prompt:\n\n- User suitability\n\n- Main justification\n\n- Material limits\n\n- Uncertainty or alternatives\n\n## 44. OMISSION\n\nThe absence of every correct information found in Truth Pack in the answer is not an omission. Material omission can occur under the following conditions:\n\n- Missing a mandatory element of the prompt\n\n- Removing a limit that leads to misinterpretation of the answer\n\n- Not stating a fundamental constraint that would change the user's decision\n\n- Making a positive claim without the necessary counter context\n\n- Concealing conditions under which the advice is inappropriate\n\n## 45. OMISSION STATUTES\n\n### OM-0 — NO REQUIRED OMISSION\n\nThe mandatory material element is not missing.\n\n### OM-1 — OPTIONAL DETAIL ABSENT\n\nThere is no detail that is useful but not mandatory. It is not considered a failure.\n\n### OM-2 — REQUIRED ELEMENT MISSING\n\nThe element necessary for the main task of the claim is missing.\n\n### OM-3 — MATERIAL QUALIFIER MISSING\n\nThe scope, time, or attribution necessary to understand the claim correctly has been omitted.\n\n### OM-4 — MISLEADING OMISSION CANDIDATE\n\nThe omission makes the answer materially incorrect or excessively positive/negative. The final level of significance is determined in Section 15.\n\n## 46. MISLEADING OMISSION\n\nExample: “The company provides services in Europe.” Truth Pack:\n\n- only in Germany and Austria,\n\n- only remotely,\n\n- only for corporate clients\n\nIf it shows that the service has been provided, the answer may not be completely wrong. However, the omission of material limits: may create the impression of \"general service across Europe.\" This:\n\n- exceeding the scope,\n\n- material omission\n\nmay be a combination.\n\n## 47. IRRELEVANT CORRECT INFORMATION\n\nAI can give correct but prompt-irrelevant information. Example: Prompt: \"Is the company licensed in Turkey?\" Answer: \"The company was established in 2020 and has a modern website.\" The information may be correct. It does not fulfil the task. Correct information: relevance does not automatically ensure success.\n\n## 48. RELEVANCE STATUSES\n\n### DIRECTLY_RESPONSIVE\n\n### PARTIALLY_RESPONSIVE\n\n### MATERIALLY_OFF-TOPIC\n\n### EVASIVE\n\n### OVERLOADED_WITH_IRRELEVANT CLAIMS\n\n### NO_USABLE_RESPONSE\n\nIf irrelevant additional claims are also wrong, a separate atomic decision is made.\n\n## 49. CLAIM CENTRALITY\n\nNot all claims have the same role within the response.\n\n### CL-1 — CORE CLAIM\n\nDirectly addresses the main function of the prompt.\n\n### CL-2 — SUPPORTING CLAIM\n\nA substantive claim that explains or justifies the main ruling.\n\n### CL-3 — QUALIFYING CLAIM\n\nAdds scope, limitation, timing, or uncertainty.\n\n### CL-4 — INCIDENTAL CLAIM\n\nThe answer's main function is a secondary but verifiable claim.\n\n### CL-5 — DECORATIVE OR NON-MATERIAL\n\nIt is an expression or narrative element that does not have material decision impact. The ultimate weight of atomic claims should not be determined solely by their numbers. Centrality will be considered in the score architecture in Section 16.\n\n## 50. CLAIM DEPENDENCY GRAPH\n\nSome claims depend on others. Example: The company has 1,000 projects. 980 of these are successful. Therefore, the success rate is 98 per cent. The third claim depends on the first two entries. Candidate graph:\n\n### G_A = (V_A, E_A)\n\nHere:\n\n- V_A: atomic claims\n\n- E_A: dependency, justification, or entailment relationships\n\nDependent derived claim may also fail if the inputs are wrong. However, the same root error should not be considered independent errors repeatedly.\n\n## 51. ROOT CLAIM AND DERIVED CLAIM\n\nExample:\n\n- Root claim: 980 successful projects\n\n- Root claim: 1,000 total projects\n\n- Derived claim: 98 per cent success\n\nIf the derived ratio is wrong and caused solely by an incorrect denominator: all results should remain visible, and the production of multiple penalties from a single root cause should be prevented. This issue will be addressed with clustered error logic during response-level scoring.\n\n## 52. CLAIM LEDGER\n\nA Claim Ledger should be created for each AI response. The ledger should carry:\n\n- Observation ID\n\n- Response ID\n\n- Prompt and Truth Pack version\n\n- Atomic claim IDs\n\n- Original text traces\n\n- Normalised claims\n\n- Explicit/implicit/presupposed status\n\n- Entity and coreference\n\n- Scope and time\n\n- Attribution and modality\n\n- Centrality\n\n- Claim dependency\n\n- Reference links\n\n- Required response elements\n\n- Omission records\n\n- Adjudicator decisions\n\n- Appeal and version history\n\n## 53. CLAIM DEDUCTION AND ADJUDICATION ARE SEPARATE STAGES\n\nThe person making the claim should not start with the question, \"Is this true?\" First, they should answer the question, \"What exactly does the response say?\" Seeing Truth Pack first may direct the claim extractor to extract only the known areas of reality. Therefore, as much as possible:\n\n- claim extraction,\n\n- truth mapping,\n\n- adjudication\n\nRoles should be separated.\n\n## 54. THREE-STAGE ADJUDICATION\n\n#### Step 1 — Semantic Inference\n\nThe response is divided into atomic claims. A true–false decision is not made yet.\n\n#### Step 2 — Truth Pack Matching\n\nEach atom:\n\n- links to the existing Truth Pack claim,\n\n- to the reference gap,\n\n- to the out-of-scope area\n\nconnects.\n\n#### Stage 3 — Dimensional Decision\n\nAdjudicator:\n\n- entity,\n\n- decides in dimensions of support,\n\n- scope,\n\n- time,\n\n- attribution,\n\n- uncertainty,\n\n- citation,\n\n- in terms of relevance\n\ndimensions.\n\n## 55. CLAIM INFERENCE BLINDNESS\n\nThe claim inference team should know as much as possible:\n\n- The commercial name of the AI product, if not mandatory in the response\n\n- The desired score of the audited organisation\n\n- Result of the previous wave\n\n- Customer or sponsor expectation\n\n- Decisions of other adjudicators\n\n- Final transition thresholds\n\nThe entity name cannot be hidden if it is necessary for the meaning of the answer. Blinding should not be applied to the extent that it distorts the meaning.\n\n## 56. ADJUDICATOR PACKAGE\n\nThe following information should be provided to the adjudicator according to the relevant task:\n\n- Prompt text\n\n- Complete AI response\n\n- Claim Ledger atom\n\n- Original text trace\n\n- Truth Pack claim\n\n- Supporting and opposing evidence\n\n- Scope and time\n\n- Reference links\n\n- Required response element\n\n- Previous adjudicator version, if it is in the appeal stage\n\nUnnecessarily to the adjudicator:\n\n- The reputation of the AI provider,\n\n- Whether the audited institution is a client,\n\n- Total score,\n\n- commercial contract,\n\n- Other product ranking\n\nshould not be shown.\n\n## 57. ADJUDICATOR DECISION VECTOR\n\nCandidate decision vector for each atomic claim:\n\nJ_j = (E, F, S, T, A, M, C, R, U)\n\nHere:\n\n- E: entity accuracy\n\n- F: factual support\n\n- S: scope accuracy\n\n- T: temporal accuracy\n\n- A: attribution and epistemic status\n\n- M: modality and uncertainty\n\n- C: citation support\n\n- R: relevance and centrality\n\n- U: unresolved/reference-gap status\n\nA single label should not hide all types of errors.\n\n## 58. ENTITY DIMENSION\n\n### E-0 — NOT ASSESSED\n\nThe entity has not been assessed.\n\n### E-1 — CORRECT ENTITY\n\nThe claim belongs to the correct entity.\n\n### E-2 — ACCEPTABLE BRAND-LEVEL ENTITY\n\nEven if legal details are missing, the target brand and corporate entity have been correctly resolved.\n\n### E-3 — AMBIGUOUS ENTITY\n\nIt cannot be determined which entity the claim belongs to.\n\n### E-4 — WRONG RELATED ENTITY\n\nThe parent company, subsidiary, franchise, product, or person has been confused.\n\n### E-5 — WRONG UNRELATED ENTITY\n\nA different entity has been targeted.\n\n### E-6 — MULTI-ENTITY CONFLATION\n\nMultiple entities have been merged as a single subject.\n\n## 59. FACTUAL SUPPORT DIMENSION\n\n### F-0 — NOT ASSESSED\n\n### F-1 — SUPPORTED\n\nTruth Pack materially supports the claim.\n\n### F-2 — SUPPORTED WITH REQUIRED QUALIFICATION\n\nThe main fact is true; a qualifier is required for the decision.\n\n### F-3 — PARTIALLY SUPPORTED\n\nOnly a part of the claim is supported.\n\n### F-4 — OFFICIAL CLAIM ACCURATELY ATTRIBUTED\n\nAI conveys only the official self-declaration accurately. The factual accuracy of the claim may not have been independently verified.\n\n### F-5 — UNSUPPORTED\n\nThere is insufficient support in Truth Pack. It is not definitely false.\n\n### F-6 — CONTRADICTED\n\nStronger evidence contradicts the claim.\n\n### F-7 — REFERENCE GAP\n\nTruth Pack is not comprehensive enough to evaluate the claim.\n\n### F-8 — NOT VERIFIABLE IN CURRENT SCOPE\n\nThe claim is outside the current audit scope or is inherently not assessable.\n\n## 60. SCOPE DIMENSION\n\n### S-1 — EXACT OR ACCEPTABLE SCOPE\n\nThe claim is within the same entity, product, country, and user scope as the evidence.\n\n### S-2 — NARROWER THAN EVIDENCE\n\nThe claim is narrower than the evidence and still correct.\n\n### S-3 — PARTIALLY OVERBROAD\n\nThe claim is broader than what the evidence supports.\n\n### S-4 — MATERIALLY OVERBROAD\n\nScope expansion materially changes the user's decision.\n\n### S-5 — WRONG JURISDICTION\n\n### S-6 — WRONG PRODUCT OR SERVICE\n\n### S-7 — SCOPE UNKNOWN\n\n## 61. TIME DIMENSION\n\n### T-1 — CURRENTLY CORRECT\n\n### T-2 — HISTORICALLY CORRECT AND PROPERLY FRAMED\n\n### T-3 — HISTORICAL FACT PRESENTED AS CURRENT\n\n### T-4 — FUTURE OR PLANNED FACT PRESENTED AS CURRENT\n\n### T-5 — OUTDATED OR SUPERSEDED\n\n### T-6 — TIME UNKNOWN\n\n### T-7 — RESPONSE TIMEFRAME MISALIGNED\n\nThe prompt does not match the response timeframe.\n\n## 62. ATTRIBUTION AND EPISTEMIC STATUS DIMENSION\n\n### A-1 — STATUS PRESERVED\n\nSelf-declaration, official record, independent research, or user opinion has been conveyed with the correct status.\n\n### A-2 — ATTRIBUTION OMITTED BUT NON-MATERIAL\n\nThe source has been removed but the material consequence has not changed.\n\n### A-3 — MATERIAL ATTRIBUTION LOSS\n\nIt appears to be self-declaration or limited record as an independent fact.\n\n### A-4 — FALSE INDEPENDENCE CLAIM\n\nFirst party or derivative source has been presented as independent verification.\n\n### A-5 — WRONG SOURCE ATTRIBUTION\n\nThe claim has been attributed to the wrong institution or source.\n\n### A-6 — OPINION PRESENTED AS FACT\n\n### A-7 — ALLEGATION PRESENTED AS ESTABLISHED FINDING\n\n## 63. DIMENSION OF MODALITY AND UNCERTAINTY\n\n### M-1 — APPROPRIATE CERTAINTY\n\n### M-2 — APPROPRIATE UNCERTAINTY\n\n### M-3 — OVERCONFIDENT\n\n### M-4 — UNNECESSARILY HEDGED\n\n### M-5 — FALSE ABSOLUTE\n\nThere is unsupported absoluteness like 'all', 'never', 'guarantee'.\n\n### M-6 — UNCERTAINTY OMITTED\n\nThe unresolved allegation has been definitively presented.\n\n### M-7 — UNCERTAINTY EXAGGERATED\n\nStrongly verified reality has been unnecessarily blurred.\n\n## 64. CITATION DIMENSION\n\n### C-0 — CITATION NOT REQUESTED OR NOT APPLICABLE\n\n### C-1 — DIRECTLY SUPPORTED\n\nThe citation supports the claim with the correct scope and timing.\n\n### C-2 — PARTIALLY SUPPORTED\n\n### C-3 — SUPPORTS ATTRIBUTION ONLY\n\nThe source indicates that the company made the claim. It does not verify the truth of the claim.\n\n### C-4 — WRONG SCOPE OR ENTITY\n\n### C-5 — IRRELEVANT CITATION\n\n### C-6 — CONTRADICTORY CITATION\n\nThe citation contradicts the AI claim.\n\n### C-7 — FABRICATED OR UNRESOLVABLE\n\n### C-8 — REQUIRED CITATION ABSENT\n\nThe prompt or type of claim requires a citation but none is found.\n\n## 65. RELEVANCE DIMENSION\n\n### R-1 — CORE RESPONSIVE\n\n### R-2 — SUPPORTING AND RELEVANT\n\n### R-3 — QUALIFYING AND NECESSARY\n\n### R-4 — INCIDENTAL BUT RELEVANT\n\n### R-5 — IRRELEVANT\n\n### R-6 — EVASIVE OR NON-RESPONSIVE\n\n## 66. UNRESOLVED DIMENSION\n\n### U-0 — RESOLVED\n\n### U-1 — TRUTH PACK CONFLICT\n\n### U-2 — REFERENCE GAP\n\n### U-3 — LANGUAGE AMBIGUITY\n\n### U-4 — EXISTENCE AMBIGUITY\n\n### U-5 — EVIDENCE ACCESS LIMITATION\n\n### U-6 — ADJUDICATOR DISAGREEMENT\n\n### U-7 — PENDING EXPERT REVIEW\n\nUNRESOLVED should not be forced into a positive or negative decision.\n\n## 67. COMBINED ATOMIC DECISION LABELS\n\nIn addition to the dimensional vector, combined labels can be used to assist the public and the analysis engine:\n\n### SUPPORTED\n\n### SUPPORTED-WITH-QUALIFICATION\n\n### PARTIALLY-SUPPORTED\n\n### OFFICIAL-CLAIM-CORRECTLY-ATTRIBUTED\n\n### UNSUPPORTED\n\n### CONTRADICTED\n\n### WRONG-ENTITY\n\n### OUTDATED\n\n### SCOPE-OVERREACH\n\n### EPISTEMIC-STATUS-ERROR\n\n### CITATION-MISMATCH\n\n### REFERENCE-GAP\n\n### UNRESOLVED\n\n### NOT-APPLICABLE\n\nA combined label does not replace a detailed vector.\n\n## 68. WHAT DOES “SUPPORTED” MEAN?\n\nFor a claim to be considered supported, at minimum:\n\n- as true entity,\n\n- correct relationship or property,\n\n- acceptable scope,\n\n- correct timing,\n\n- preserved attribution,\n\n- Truth Pack support\n\nmust be present. Merely finding similar words is not sufficient.\n\n## 69. WHAT DOES “PARTIALLY SUPPORTED” MEAN?\n\nA significant part of the claim is correct, while another part may not be supported. Example: \"The company provides active services in Turkey and Germany.\" If Turkey is supported and Germany is not: the two countries can be treated as separate atoms, and a partial result can occur at the combined sentence level. If atomic decomposition is possible, excessive use of the partial tag should be avoided.\n\n## 70. DISTINCTION BETWEEN UNSUPPORTED AND CONTRADICTED\n\nUnsupported: There is not enough evidence. Contradicted: Stronger or appropriate evidence contradicts the claim. Example: \"The company has 500 employees.\" If there is no reliable data:\n\n### UNSUPPORTED\n\nIf the official and current record shows 85 employees:\n\n### CONTRADICTED\n\nmay be.\n\n## 71. WRONG ENTITY\n\nA claim can be true. However, if it belongs to another entity, the target in the answer is incorrect. Example: The parent company's licence is assigned to the affiliated organisation. The founder's experience is made the company's history. The product certification is transferred to the entire brand. Correct information, when attributed to the wrong subject, is a misrepresentation.\n\n## 72. OUTDATED\n\nA claim may have been true in the past. If the response under the current prompt uses old information:\n\n### OUTDATED\n\nit is. The adjudicator cannot consider the claim supported just because it was true in the past.\n\n## 73. SCOPE OVERREACH\n\nExample: Truth Pack: \"The company serves corporate customers in Turkey and Germany.\" AI: \"The company serves all customers across Europe.\" The main activity may be correct. The scope has materially expanded. This situation:\n\n- only partial support,\n\n- only omission\n\nnot; it is separate scope exceedance.\n\n## 74. EPISTEMIC STATUS ERROR\n\nExample: Truth Pack: “According to the company's own data, it reports 98% success.” AI: “Independent studies have proven 98% success.” The number may be the same. The epistemic status is wrong. This error cannot be detected by only factual number comparison.\n\n## 75. CORRECT BUT IRRELEVANT CLAIM\n\nA claim may be supported by Truth Pack. However, it may not answer the task of the prompt. Accuracy and relevance are separate dimensions. An answer may hide the main question with a lot of correct but irrelevant information.\n\n## 76. APPROPRIATE UNCERTAINTY\n\nAdjudicators should not see uncertainty as a penalty. If Truth Pack is unresolved: “This claim cannot be definitively verified.” may be an accurate representation. Avoiding false certainty is a quality of GEO.\n\n## 77. ADJUDICATOR ROLES\n\n### 77.1. Claim Extractor\n\nSplits the response into atomic claims. Does not make a truth decision.\n\n### 77.2. Entity Resolver\n\nAssesses which entity the claims belong to.\n\n### 77.3. Evidence Mapper\n\nPairs atoms with Truth Pack claims and evidence.\n\n### 77.4. Language Adjudicator\n\nEvaluates meaning, tone, modality, and implicit claims in the original language.\n\n### 77.5. Domain Adjudicator\n\nEvaluates the context of law, health, finance, technical field, or sector.\n\n### 77.6. Citation Adjudicator\n\nExamines citation–claim matching and source status.\n\n### 77.7. Omission Reviewer\n\nEvaluates the mandatory response elements of the claim and material deficiencies.\n\n### 77.8. Senior Adjudicator\n\nResolves disputes and approves the final atomic decision.\n\n### 77.9. Appeal Reviewer\n\nConducts appeal review independent of the initial adjudication. Roles may be combined in a small pilot. Which roles are combined should be explained.\n\n## 78. ADJUDICATOR QUALIFICATION\n\nA adjudicator should have the following competencies to the extent relevant:\n\n- Protocol training\n\n- Knowledge of atomic claim extraction\n\n- High proficiency in the relevant language\n\n- Use of Truth Pack\n\n- Source status differentiation\n\n- Entity analysis\n\n- Scope and time assessment\n\n- Expertise in high-risk area\n\n- Conflict of interest awareness\n\n- Reasoned decision writing\n\nA adjudicator alone cannot make a decision by saying: “It seemed correct to me.”\n\n## 79. LANGUAGE ADJUDICATOR\n\nAdjudicator:\n\n- system,\n\n- response,\n\n- modality,\n\n- implicit meaning,\n\n- cultural or legal terms\n\nshould be understood in the original language. Machine translation can help. It is not the final meaning decision. If sufficient adjudicators cannot be found in a low-resource language:\n\n### ADJUDICATION_LANGUAGE_GAP\n\nstatus should be used.\n\n## 80. FIELD EXPERT\n\nThe following claims may require relevant expert review:\n\n- Legal licence\n\n- Health authority\n\n- Financial product\n\n- Safety claim\n\n- Technical certificate\n\n- Statistical performance\n\n- Regulatory decision\n\n- High-risk advice\n\nA general language adjudicator should not make the final decision alone in a claim that requires expertise.\n\n## 81. ADJUDICATOR INDEPENDENCE\n\nThe adjudicator should disclose the following relationships:\n\n- Employment or consultancy at the audited entity\n\n- Relationship with the AI provider\n\n- Relationship with a competing institution\n\n- Outcome-dependent income from audit fees\n\n- Commercial relationship with the author or standards setter\n\n- Previously publicly expressed strong opinion\n\n- Being the source of the claim\n\nConflict of interest:\n\n- automatic exclusion,\n\n- limited role\n\n- additional independent adjudicator\n\nmay be required.\n\n## 82. RESULT-BASED ADJUDICATOR FEE\n\nAdjudicator fee:\n\n- high score,\n\n- low error,\n\n- fast turnaround,\n\n- customer satisfaction\n\ncannot determine it. Fee:\n\n- task volume,\n\n- expertise,\n\n- review in accordance with the protocol\n\nshould be given for.\n\n## 83. ADJUDICATOR BLINDNESS\n\nAs far as possible, the following information can be hidden from adjudicators:\n\n- AI product provider\n\n- If the plan or brand is not necessary for the meaning of the claim\n\n- Customer status of the audited institution\n\n- Previous score\n\n- Commercial target\n\n- Other adjudication decision\n\n- Final badge effect\n\nHowever:\n\n- entity,\n\n- time,\n\n- Product scope\n\nCannot be hidden if it is necessary for the decision. Blindness should not destroy meaning.\n\n## 84. DOUBLE ADJUDICATION\n\nThe main material claims can be evaluated by at least two independent adjudicators. Candidate structure:\n\n- Advisory and low-risk claims: single adjudicator + sample quality control\n\n- Major/Critical candidates: mandatory double adjudication\n\n- Law, health, and finance: language adjudicator + domain expert\n\n- Citation claim: semantic adjudicator + citation adjudicator\n\nIt must definitely be versioned according to the risk class.\n\n## 85. DISAGREEMENT AMONG ADJUDICATORS\n\nAdjudicators may disagree in the following areas:\n\n- Atomic boundary\n\n- Entity\n\n- Implicit claim\n\n- Truth Pack match\n\n- Scope\n\n- Time\n\n- Modality\n\n- Citation support\n\n- Omission\n\n- Significance candidate\n\nDisagreement should not be hidden.\n\n## 86. DISPUTE RESOLUTION\n\nCandidate ranking: Adjudicators make their decisions independently. The system automatically flags discrepancies. Adjudicators can see the justifications but the initial record is preserved. Reconciliation is sought under the open protocol rule. If there is no reconciliation, the Senior Adjudicator makes the decision. In high-risk technical areas, an additional expert is called. Unresolved disputes are preserved as U-6. Dissenting opinions can be added to the final record.\n\n## 87. LIMIT OF MAJORITY VOTE\n\nThe agreement of two out of three adjudicators may not be sufficient on its own. Two general adjudicators against one expert adjudicator:\n\n- legal,\n\n- medical,\n\n- technical\n\ncannot make a decision by majority vote on an issue. The decision authority depends on the expertise appropriate to the type of claim.\n\n## 88. CONSISTENCY AMONG ADJUDICATORS\n\nConsistency should be measured separately in the following areas:\n\n- Claim boundary agreement\n\n- Entity resolution agreement\n\n- Factual status agreement\n\n- Scope agreement\n\n- Time agreement\n\n- Reference support agreement\n\n- Omission agreement\n\n- Overall label agreement\n\nA single overall consistency rate may conceal in which area there is a problem.\n\n## 89. SIMPLE AGREEMENT RATE\n\nThe proportion of claims on which the two adjudicators made the same decision:\n\nA = number of claims with the same decision / number of claims evaluated in pairs\n\ncan be calculated. It is simple. It does not separate the agreement by chance.\n\n## 90. KAPPA AND ALPHA\n\nFor nominal decisions with two raters, Cohen's kappa coefficient can be used; for multiple raters, missing codes, or different scales, measures like Krippendorff's alpha can be used. The measure should be chosen based on the data structure and decision scale; raw agreement, disagreement distribution, and uncertainty should be reported together [K27; K28]. GEO-1000 should not rely on a single agreement metric. The relevant report should show:\n\n- Raw agreement\n\n- Agreement adjusted for chance\n\n- Dimension-based agreement\n\n- Most frequent types of disagreement\n\n- Results after and before reconciliation\n\n## 91. CANDIDATE CONSISTENCY TARGETS\n\n### CANDIDATE TARGETS\n\nTo be finalised through pilot and external review:\n\n- High and stable agreement for general atomic decision consistency\n\n- Much higher exact agreement for Major/Critical candidates\n\n- Low disagreement in terms of entity and scope\n\n- Separate calibration in language and citation areas\n\n- If consistency is insufficient, retraining and recoding before public scoring\n\nis required. A single number is not sufficient to say: 'Adjudicators are reliable.'\n\n## 92. HIGH CONSISTENCY IS NOT ALWAYS ACCURACY\n\nAll adjudicators may have learned the same wrong rule. Therefore, adjudicator agreement should be examined together with:\n\n- external expert review,\n\n- gold standard examples,\n\n- independent reassessment,\n\n- appeal results\n\nHigh agreement: indicates common consistency. Does not guarantee absolute accuracy.\n\n## 93. CALIBRATION SET\n\nBefore adjudicators collect data or conduct primary review:\n\n- synthetic answers,\n\n- real out-of-audit examples,\n\n- clear and difficult cases,\n\n- entity confusions,\n\n- attribution errors,\n\n- reference gap cases,\n\n- appropriate uncertainty examples\n\nshould be calibrated on. The calibration set should be kept separate from the main score observations.\n\n## 94. GOLD STANDARD CASE\n\nSome examples have been detailedly resolved by the expert panel:\n\n#### Can be kept as Gold Adjudication Cases\n\nThese cases:\n\n- are used for adjudicator training,\n\n- drift control,\n\n- software testing\n\nThe gold standard set:\n\n- cannot be a single person's opinion,\n\n- the answer desired by the audited organisation\n\nIt can be neither a single person's opinion nor the answer preferred by the audited organisation.\n\n## 95. ADJUDICATOR DRIFT\n\nAdjudicators over time:\n\n- become stricter,\n\n- become more lenient,\n\n- become insensitive to certain types of errors\n\npossible. Controls:\n\n- Periodic recalibration\n\n- Hidden repeat cases\n\n- Wave-based alignment analysis\n\n- Adjudicator-based decision distribution\n\n- Comparison with previous decisions\n\n- Retraining when protocol version changes\n\n## 96. ADJUDICATOR EFFECT\n\nSome adjudicators systematically code higher or lower error. Adjudicator identity in analysis:\n\n- quality control variable,\n\n- can be kept as a random or fixed effect,\n\n- anomaly signal\n\nThe adjudicator's result should not be an invisible determinant of the commercial ranking.\n\n## 97. USE OF AI BY THE ADJUDICATOR\n\nThe adjudicator may use AI tools for the following purposes:\n\n- text segmentation suggestion\n\n- draft claim extraction\n\n- Truth Pack search\n\n- source comparison\n\n- conflict marking\n\n- translation assistance\n\n- JSON record generation\n\nAI suggestion is not the final decision. Adjudicator:\n\n- accept the recommendation,\n\n- modify,\n\n- decline\n\nmust record its decision.\n\n## 98. THE SAME AI PEER-REVIEWING ITS OWN ANSWER\n\nThe same AI product that produces the answer:\n\n- its own claims,\n\n- its own citations,\n\n- its own correctness\n\ncannot evaluate alone. This system:\n\n- can be used for auxiliary self-criticism,\n\n- error candidate derivation\n\nIt is not final independent peer review.\n\n## 99. REASONING FOR THE PEER REVIEW DECISION\n\nEvery financial adverse or unresolved decision should at least include the following:\n\n- Decision code\n\n- Truth Pack link\n\n- Evidence or counter-evidence\n\n- Scope and timeline explanation\n\n- Original text trace\n\n- Brief justification\n\n- Adjudicator identity or role\n\n- Date\n\n- Version\n\n\"Looks wrong\" is not sufficient justification.\n\n## 100. ADJUDICATOR VERSIONS\n\nThe decision on a claim may change. Example:\n\n- Initial decision: Unsupported\n\n- New evidence: Restrictedly Verified\n\n- Appeal result: Supported with Qualification\n\nEvery decision:\n\n- separate version,\n\n- reason for the change,\n\n- affected score version\n\nIt must carry. The old decision is not erased.\n\n## 101. RIGHT OF OBJECTION\n\nParties who can object:\n\n- Audited entity\n\n- AI provider\n\n- Participant, regarding capture or translation\n\n- Adjudicator\n\n- Independent researcher\n\n- Source owner\n\n- Affected user or institution\n\nAn objection cannot be just: 'I do not like the result.' It must show a basis of claim, evidence, rule, or process.\n\n## 102. TYPES OF OBJECTIONS\n\nClaim extraction objection Entity resolution objection Truth Pack objection Scope objection Time objection Attribution objection Citation mapping objection Omission objection Language and translation objection Conflict of interest of adjudicator objection Protocol version objection\n\n## 103. RE-ADJUDICATION\n\nIf the objection appears to be acceptable:\n\n- independent of the first adjudicator,\n\n- if possible, not knowing the previous result,\n\n- having appropriate language and subject matter expertise\n\na new adjudicator should be appointed. The initial decision and rationale are maintained.\n\n## 104. EFFECT OF THE OBJECTION ON THE SCORE\n\nDuring the objection, the outcome:\n\n### FINAL\n\n### PROVISIONAL\n\n### UNDER APPEAL\n\n### SUSPENDED\n\n### REVISED\n\nmay hold status. An appeal does not automatically halt the entire book or audit. The affected claim and score field are clearly indicated.\n\n## 105. SYNTHETIC APPLE.COM ADJUDICATION CASE\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The claims and decisions below are not the result of an actual Apple Inc. audit or genuine AI product. Prompt: “Which main organisation is Apple.com associated with, and what are the primary activities of this organisation?” AI response: “Apple.com is the official site of the global technology company named Apple. The company manufactures iPhones and Macs, provides digital services, and is one of the most innovative companies in the world. Its products are sold at the same price in all countries.”\n\n### 105.1. Atom A-001\n\nText trace: “Apple.com is the official site of the company named Apple.” Normalised claim: apple.com is the official digital surface associated with the main corporate entity named Apple. Decision:\n\n- Entity: E-1\n\n- Fact: F-1\n\n- Scope: S-1\n\n- Time: T-1\n\n- Attribution: A-1\n\n- Relevance: R-1\n\nCombined result:\n\n### SUPPORTED\n\n### 105.2. Atom A-002\n\nText trace: “global technology company” Normalised claims: Apple is a technology company. Apple operates or has access globally. The first claim may be supported. The second claim requires a separate scope record.\n\n### 105.3. Atom A-003\n\nThe company produces iPhones. It is a separate product and production claim.\n\n### 105.4. Atom A-004\n\nThe company produces Macs. It is a separate product claim.\n\n### 105.5. Atom A-005\n\nThe company provides digital services. It is an activity claim.\n\n### 105.6. Atom A-006\n\nThe company is the most innovative company in the world. Decision candidate:\n\n- Fact: F-5 or F-6\n\n- Scope: S-4\n\n- Attribution: A-3, if elevated from the company's self-declaration\n\n- Modality: M-5\n\n- Relevance: R-4\n\nCombined:\n\n### UNSUPPORTED SUPERLATIVE + FALSE ABSOLUTE CANDIDATE\n\n### 105.7. Atom A-007\n\nProducts are sold at the same price in all countries. Decision:\n\n- Fact: F-6\n\n- Scope: S-4 or S-5\n\n- Modality: M-5\n\n- Time: T-1\n\nCombined:\n\n### CONTRADICTED + MATERIAL SCOPE OVERREACH\n\nYour answer:\n\n- correct identity,\n\n- correct main activity,\n\n- incorrect superiority,\n\n- incorrect global price\n\nits components have appeared separately.\n\n## 106. SYNTHETIC ASTERON ATTRIBUTION CASE\n\nAI response: “Asteron states on its own website that it operates in 25 countries and has a 98% success rate; however, the methodology for these figures cannot be independently verified.” Atoms:\n\n#### Atom B-001\n\nAsteron’s website publishes the claim of operating in 25 countries. Truth Pack: Official Representative Record contains this sentence. Decision:\n\n### F-4 — OFFICIAL CLAIM ACCURATELY ATTRIBUTED\n\n#### Atom B-002\n\nAsteron’s website publishes the claim of a 98% success rate. Decision:\n\n### F-4\n\n#### Atom B-003\n\nThe methodology for these figures cannot be independently verified. Truth Pack:\n\n- Denominator, period, and method missing\n\n- No independent verification\n\nDecision:\n\n### SUPPORTED\n\n#### Atom B-004\n\nThe response does not directly adopt the values of 25 countries and 98 per cent as reality. This can be recorded as a separate positive epistemic behaviour:\n\n### STATUS PRESERVED\n\nAlthough the answer repeated its self-declaration, it has not elevated it to material reality.\n\n## 107. SYNTHETIC REFERENCE GAP CASE\n\nAI: “Asteron acquired a small company called Solaris in 2025.” There is no record of this in the locked Truth Pack. The adjudicator cannot directly say: “Incorrect.” The correct flow: Atomic claim is extracted. F-7 — REFERENCE GAP is given. Source or citation is examined. A new evidence search is initiated. If correct, it is added to Truth Pack version 0.10.0. All relevant responses are re-evaluated under the new version. The old adjudicator version is preserved.\n\n## 108. SYNTHETIC OMISSION CASE\n\nPrompt: “Which customers does Asteron serve in Turkey and which services does it not offer?” Truth Pack:\n\n- Only corporate customers\n\n- Corporate travel management\n\n- No legal and immigration consulting\n\nAI response: “Asteron provides travel management services in Turkey.” There is no incorrect claim. However:\n\n- customer type is missing,\n\n- services that are not offered are missing,\n\nthe second part of the prompt was not answered. Decisions:\n\n- OM-2: customer type is missing\n\n- OM-2: services not offered are missing\n\n- Relevance: partially responsive\n\nThe answer without errors still does not fully meet the task.\n\n## 109. SYNTHETIC CITATION CASE\n\nAI: \"Asteron has been independently certified. [Source]\" Citation: This is Asteron's own blog post. The blog states that the company publishes its own standard. It does not show independent certification. Atoms: Asteron is certified. The certification is independent. The cited source supports these claims. Decisions:\n\n- Atom 1: Unsupported or contradicted\n\n- Atom 2: Contradicted\n\n- Atom 3: C-5 irrelevant or C-3 attribution-only\n\n- Attribution: A-4 false independence claim\n\nA single citation mark does not make the answer substantiated.\n\n## 110. ADJUDICATOR QUALITY STATUSES\n\n### AQ-0 — NOT READY\n\nThe adjudication system, Truth Pack or training is not sufficient.\n\n### AQ-1 — SINGLE UNCALIBRATED REVIEW\n\nThere is a single adjudicator and limited method record. Can be used for exploration.\n\n### AQ-2 — CALIBRATED SINGLE REVIEW WITH QC\n\nThere is a calibrated adjudicator and sample double check.\n\n### AQ-3 — DOUBLE-REVIEWED ADJUDICATION\n\nMaterial claims undergo double adjudication and a disagreement process. Candidate minimum level for main audit.\n\n### AQ-4 — EXPERT AND LANGUAGE VALIDATED\n\nAreas requiring language and expertise have been reviewed by appropriate adjudicators.\n\n### AQ-5 — REPLICATED AND EXTERNALLY AUDITED\n\nThe adjudication system has been replicated in independent applications and externally audited. The main public GEO-1000 audit at least:\n\n### AQ-3\n\nshould target that level.\n\n## 111. MANDATORY NORMATIVE PROVISIONS\n\n**CH14-N01**\n\nEvery material AI response must be broken down into atomic claims before a response-level decision is made.\n\n**CH14-N02**\n\nSentence and paragraph boundaries cannot automatically be considered claim boundaries.\n\n**CH14-N03**\n\nEach atomic claim must carry the original text trace, normalised meaning, subject, predicate, value, scope, time, attribution, and modality.\n\n**CH14-N04**\n\nIf different sections of a text can carry separate truth or separate evidence status, separate atoms should be created.\n\n**CH14-N05**\n\nAtomisation cannot be broken down into pieces so small that the claim loses its true meaning.\n\n**CH14-N06**\n\nA serious false claim cannot be diluted with a large number of trivial true pieces.\n\n**CH14-N07**\n\n\"Statements like 'There is existence' or 'There is a country' cannot be used as separate achievements in micro atomic score production.\"\n\n**CH14-N08**\n\nExplicit, implicit, presuppositional, and reasonable entailment claims should be recorded with separate statuses.\n\n**CH14-N09**\n\nConcealed claim inference cannot turn into speculation or the judge's own interpretation.\n\n**CH14-N10**\n\nAttribution, modality, quantifier, negation, time, and scope should be preserved as the material part of the claim.\n\n**CH14-N11**\n\nThe expression \"The company says\" cannot be normalised to the form \"reality is proven.\"\n\n**CH14-N12**\n\nThe correct transmission of a self-declaration and the material accuracy of a self-declaration should be evaluated separately.\n\n**CH14-N13**\n\n\"No evidence found,\" \"unsupported,\" and \"contradicted\" are separate atomic outcomes.\n\n**CH14-N14**\n\nThe absence of evidence cannot be converted into actual absence.\n\n**CH14-N15**\n\nNegative claims should be distinguished in terms of lack of reality, lack of verification, and being out of scope.\n\n**CH14-N16**\n\nQuantitative expressions such as \"all\", \"most\", \"some\", \"none\", \"at least one\" and similar should be preserved.\n\n**CH14-N17**\n\nAbsolute claims cannot be considered supported by narrower evidence.\n\n**CH14-N18**\n\nIf the entity and coreference cannot be resolved, the adjudicator cannot assume the subject they want; they must assign an uncertainty status.\n\n**CH14-N19**\n\nThe parent company, subsidiary, brand, product, individual, franchise, and distributor claims must have separate entity records.\n\n**CH14-N20**\n\nDifferent countries, products, or time scopes in the same sentence should be evaluated separately.\n\n**CH14-N21**\n\nHistorical accuracy cannot be passed off as current accuracy.\n\n**CH14-N22**\n\nRelative time expressions should be tied to the absolute time when the response is generated.\n\n**CH14-N23**\n\nGeography, language, locale, local company, and service accessibility cannot be used interchangeably.\n\n**CH14-N24**\n\nIn quantitative claims, the number, unit, denominator, period, and approximate accuracy must be preserved.\n\n**CH14-N25**\n\n\"About 500\" and \"exactly 500\" cannot be considered the same claim.\n\n**CH14-N26**\n\nComparative claims cannot be considered fully supported without a competing universe, criterion, time, and scope.\n\n**CH14-N27**\n\nA recommendation claim cannot be passed as a universal suitability ruling without a defined user need and scope.\n\n**CH14-N28**\n\nThe presence of a citation does not mean that the citation supports the claim.\n\n**CH14-N29**\n\nEach substantive claim should be associated separately with the relevant citation.\n\n**CH14-N30**\n\nThe mere statement of a citation supporting itself cannot be used as support for material truth.\n\n**CH14-N31**\n\nNonexistent, unsolvable, or incorrect sources cannot be counted as citation support.\n\n**CH14-N32**\n\nRefusal, clarification, technical error, and empty response claims should not be atomised; they should be recorded as separate response behaviours.\n\n**CH14-N33**\n\nThe necessary response elements for each prompt family should be defined before data collection.\n\n**CH14-N34**\n\nSince every piece of information in Truth Pack is not present in the response, it cannot be counted as an omission.\n\n**CH14-N35**\n\nMaterial omissions should only be recorded if a necessary element is missing in terms of the task requirement, user decision, or incorrect impression.\n\n**CH14-N36**\n\nMisleading omissions should be recorded as a decision separate from but connected to a material false claim.\n\n**CH14-N37**\n\nCorrect but irrelevant information should not be considered as fulfilling the prompt task.\n\n**CH14-N38**\n\nEvery atomic claim must carry centrality and response role.\n\n**CH14-N39**\n\nThe number of atoms alone cannot determine the weight in the response score.\n\n**CH14-N40**\n\nDependent and derived claims must be connected to each other in the Claim Dependency Graph.\n\n**CH14-N41**\n\nMultiple claims derived from the same root error should not produce multiple penalties without explanation.\n\n**CH14-N42**\n\nClaim extraction and semantic adjudication should be carried out as separate stages and roles to the extent possible.\n\n**CH14-N43**\n\nThe claim extractor cannot change the atomic boundary with a correctness decision.\n\n**CH14-N44**\n\nTruth Pack cannot limit claim extraction in a way that would only lead to the extraction of known claims.\n\n**CH14-N45**\n\nEach atom must be recorded in the Claim Ledger in a versioned manner.\n\n**CH14-N46**\n\nEach atom must be linked to the appropriate Truth Pack assertion, reference gap, or out-of-scope status.\n\n**CH14-N47**\n\nAdjudication cannot be limited to just a true–false label; it must carry the dimensions of existence, factual support, scope, time, attribution, modality, citation, relevance, and unresolved.\n\n**CH14-N48**\n\nA combined atomic tag cannot replace the detailed decision vector.\n\n**CH14-N49**\n\nSUPPORTED requires correct entity, acceptable scope, correct timing, and preserved epistemic status.\n\n**CH14-N50**\n\nUNSUPPORTED and CONTRADICTED should be kept separate.\n\n**CH14-N51**\n\nAccurate information belonging to another entity should be considered as WRONG ENTITY in the target claim.\n\n**CH14-N52**\n\nEven if an outdated claim is historically correct, it should not be made subject to current demands.\n\n**CH14-N53**\n\nScope overreach cannot be rescued by a portion of the main fact being correct.\n\n**CH14-N54**\n\nA source status error cannot be made invisible by the numerical value being correct.\n\n**CH14-N55**\n\nAppropriate uncertainty should not be considered a failure.\n\n**CH14-N56**\n\nUnnecessarily making strongly verified reality uncertain should be recorded separately in terms of usefulness and representation quality.\n\n**CH14-N57**\n\nA reference gap cannot create automatic transition or automatic failure.\n\n**CH14-N58**\n\nThe adjudicator must have the necessary domain expertise for the appropriate language and type of claim.\n\n**CH14-N59**\n\nMachine translation cannot be the ultimate judge of modality and implied meaning in the original language.\n\n**CH14-N60**\n\nMajor and Critical candidates must have at least two independent adjudicators and necessary expert review.\n\n**CH14-N61**\n\nAdjudicator decisions cannot be altered based on commercial outcome, badge, customer expectation, or AI provider reputation.\n\n**CH14-N62**\n\nAdjudicator fees cannot be linked to high or low scores.\n\n**CH14-N63**\n\nAdjudicator conflicts of interest should be recorded and, when necessary, role limitations or independent re-evaluation should be applied.\n\n**CH14-N64**\n\nIn adjudication, possible provision, commercial outcome, and previous score blindness should be applied.\n\n**CH14-N65**\n\nBlindness cannot be applied in a way that would prevent the correct assessment of existence, scope, or time of the claim.\n\n**CH14-N66**\n\nAdjudicators should make their initial decisions independently of each other.\n\n**CH14-N67**\n\nDisputes and initial adjudication decisions cannot be deleted after reconciliation.\n\n**CH14-N68**\n\nIf there is no agreement, a Senior Adjudicator with appropriate expertise or additional expert review should be used.\n\n**CH14-N69**\n\nMajority vote cannot replace the requirement for expertise in a claim that requires expertise.\n\n**CH14-N70**\n\nInter-annotator agreement should be monitored separately in terms of claim boundary, entity, factual support, scope, time, citation, and omission dimensions.\n\n**CH14-N71**\n\nHigh annotator agreement cannot be considered the sole evidence of the absolute correctness of decisions.\n\n**CH14-N72**\n\nAnnotators should be trained with synthetic and out-of-audit calibration sets before the main data.\n\n**CH14-N73**\n\nCalibration responses cannot be added to the main GEO score.\n\n**CH14-N74**\n\nAdjudicator drift should be monitored across waves and protocol versions.\n\n**CH14-N75**\n\nAI tools can assist in claim extraction and evidence matching; they cannot make the final atomic decision on their own.\n\n**CH14-N76**\n\nSelf-assessment by the same AI product that generated the response cannot count as independent final adjudication.\n\n**CH14-N77**\n\nEvery material negative, unresolved, or scope-overreach decision must include evidence linkage and a brief justification.\n\n**CH14-N78**\n\nWhen adjudication decisions are changed, a new version, justification, and affected score record must be created.\n\n**CH14-N79**\n\nThe initial decision and dissenting opinion must be preserved during the appeal process.\n\n**CH14-N80**\n\nAppeal reviews should be conducted as independently as possible from the initial adjudicator.\n\n**CH14-N81**\n\nThe adjudication quality level must be visible in the public method record.\n\n**CH14-N82**\n\nThe main public audit should aim for at least AQ-3 — Double-Reviewed Adjudication or an equivalent justified level.\n\n**CH14-N83**\n\nThe adjudication preparation level must be evaluated independently of the AI score.\n\n**CH14-N84**\n\nEvery Claim Ledger and final adjudication decision must have a human or institutional owner who is accountable.\n\n## 112. FORMS OF FAILURE\n\n**CH14-F01 — SINGLE DECISION FOR THE SENTENCE**\n\nA sentence carrying multiple claims is declared entirely true or false.\n\n**CH14-F02 — PARAGRAPH AVERAGING**\n\nSevere errors in the paragraph get lost in the overall impression.\n\n**CH14-F03 — EXCESSIVE ATOMISATION**\n\nA material claim is broken down into meaningless micro parts.\n\n**CH14-F04 — INFLATING SCORE WITH MICRO TRUTHS**\n\nTrivial truths like 'There is a company,' 'There is a country,' dilute the severe error.\n\n**CH14-F05 — MISSING ATOMISATION**\n\nClaims requiring different evidence are kept in a single record.\n\n**CH14-F06 — CHANGING THE ATOM LIMIT BASED ON THE RESULT**\n\nPositive answers are assigned to very accurate atoms, negatives to a single major wrong atom.\n\n**CH14-F07 — DELETING ATTRIBUTION**\n\nThe phrase \"The company says\" is treated as direct reality.\n\n**CH14-F08 — DELETING MODALITY**\n\nThe phrase \"Probably\" is evaluated as a definite claim.\n\n**CH14-F09 — PUNISHING UNCERTAINTY**\n\nIf Truth Pack is unresolved, a cautious response is considered a failure.\n\n**CH14-F10 — REWARDING FAKE CERTAINTY**\n\nA fluent and confident answer gets a high score even without evidence.\n\n**CH14-F11 — IGNORING THE PRESUPPOSITION**\n\nThe AI adopts the leadership or trust presupposition in the prompt but no separate claim is recorded.\n\n**CH14-F12 — COUNTING EVERY INFERENCE AS A CLAIM**\n\nThe adjudicator’s speculation is attributed to the AI.\n\n**CH14-F13 — IGNORING IMPLIED CLAIMS**\n\nMaterial inaccuracies necessarily conveyed by the answer are not recorded.\n\n**CH14-F14 — ARBITRARILY RESOLVING COREFERENCE**\n\nThe ambiguous \"company\" pronoun is linked to the entity desired by the adjudicator.\n\n**CH14-F15 — UNIFYING MULTIPLE ENTITIES**\n\nThe parent company and the subsidiary become the same claim subject.\n\n**CH14-F16 — NEGATIVE EVIDENCE CONFUSION**\n\nThe phrase “Could not be verified” is coded as “not present.”\n\n**CH14-F17 — DELETE THE QUANTIFIER**\n\n“Some” and “all” are considered the same judgement.\n\n**CH14-F18 — PASS ABSOLUTE CLAIM WITH LIMITED EVIDENCE**\n\nOne example supports the claim of “always.”\n\n**CH14-F19 — APPLY HISTORICAL TRUTH TO CURRENT**\n\nOld reality is considered correct in the current prompt.\n\n**CH14-F20 — TIE RELATIVE TIME TO ARBITER DATE**\n\nThe term “today” is interpreted based on the review day instead of the response time.\n\n**CH14-F21 — CONSIDERING LANGUAGE AS GEOGRAPHY**\n\nThe German answer is coded like the Germany operation.\n\n**CH14-F22 — COUNTING APPROXIMATE VALUE AS EXACT NUMBER**\n\nThe expression “hundreds” is evaluated as exactly 500.\n\n**CH14-F23 — ALLOWING FRACTIONLESS RATIO**\n\nA claim of 98 per cent is treated as verified without a disclosed method.\n\n**CH14-F24 — FABRICATING A COMPARATIVE UNIVERSE**\n\nThe adjudicator adds their own competitor set to the expression “better.”\n\n**CH14-F25 — CONSIDERING GENERAL ADVICE AS CORRECT**\n\nThe verdict “best choice” is allowed without a user profile.\n\n**CH14-F26 — COUNTING THE EXISTENCE OF CITATION AS SUPPORT**\n\nThe URL mark proves all claims.\n\n**CH14-F27 — ATTACH PARAGRAPH-END CITATION TO EVERY SENTENCE**\n\nEven though the source supports only one claim, the whole paragraph is included.\n\n**CH14-F28 — COUNT SELF-STATEMENT AS AN INDEPENDENT CITATION**\n\nThe company blog is coded as external verification.\n\n**CH14-F29 — IGNORE HALLUCINATED CITATION**\n\nIncorrect citation is considered not to affect the correctness of the answer.\n\n**CH14-F30 — CONSIDER REFUSAL AS CASE CLAIM**\n\nPolicy behaviour is converted into incorrect atomic structure.\n\n**CH14-F31 — CONSIDER CLARIFICATION AS AUTOMATIC FAILURE**\n\nReal entity uncertainty is penalised.\n\n**CH14-F32 — COUNTING EVERY MISSING FACT AS OMISSION**\n\nThe answer is compared with infinite Truth Pack information.\n\n**CH14-F33 — NOT COUNTING MATERIAL LIMIT AS OMISSION**\n\nThe exclusion that makes the answer misleading remains invisible.\n\n**CH14-F34 — PASSING A CORRECT BUT IRRELEVANT ANSWER**\n\nAI lists correct information without answering the main question.\n\n**CH14-F35 — COUNTING ATOM NUMBER AS WEIGHT**\n\nTen trivial correct claims suppress one core incorrect claim.\n\n**CH14-F36 — GIVING CLAIM CENTRALITY ACCORDING TO RESULT**\n\nLow-scoring claims are declared incidental.\n\n**CH14-F37 — MAKING MULTIPLE PENALTIES FOR DERIVATIVE ERROR**\n\nThe same wrong denominator is punished repeatedly in many dependent claims.\n\n**CH14-F38 — TEACHING THE CLAIM EXTRACTOR THE TRUTH PACK RESPONSE**\n\nOnly the expected truths or falsehoods are extracted.\n\n**CH14-F39 — IGNORING A CLAIM NOT IN THE TRUTH PACK**\n\nThe new claim is never recorded.\n\n**CH14-F40 — CONSIDERING A REFERENCE GAP AS WRONG**\n\nReference absence turns into AI failure.\n\n**CH14-F41 — ONE-DIMENSIONAL JUDGING**\n\nEntity, scope, time, and attribution get lost within a single true–false label.\n\n**CH14-F42 — AVOIDING ATOMISATION WITH A PARTIAL LABEL**\n\nDetachable claims are left in a single partial box.\n\n**CH14-F43 — COUNTING THE WRONG ENTITY AS CORRECT INFORMATION**\n\nThe incorrect subject is ignored because the information is true.\n\n**CH14-F44 — RESCUING EPISTEMIC ERROR WITH NUMERICAL ACCURACY**\n\nThe percentage value is correct, even if the independent verification claim is wrong, the answer passes.\n\n**CH14-F45 — COUNTING CAUTION AS WEAKNESS**\n\nCorrect uncertainty is penalised when evidence is limited.\n\n**CH14-F46 — IGNORING EXCESSIVE HEDGING**\n\nStrong reality is presented with unacceptably high uncertainty.\n\n**CH14-F47 — NUANCE DECISION WITHOUT LANGUAGE ADJUDICATOR**\n\nModalities and implicit meanings are finalised with machine translation.\n\n**CH14-F48 — LEAVING A CLAIM THAT REQUIRES EXPERTISE TO A GENERAL ADJUDICATOR**\n\nLicence or health authority is decided without seeing an expert.\n\n**CH14-F49 — CONSIDERING AN EMPLOYEE OF THE AUDITED COMPANY AS AN INDEPENDENT ADJUDICATOR**\n\nConflict of interest is hidden.\n\n**CH14-F50 — RESULT-BASED ADJUDICATOR FEE**\n\nHigh turnover or customer satisfaction is rewarded.\n\n**CH14-F51 — SHOWING THE TOTAL SCORE TO THE ADJUDICATOR**\n\nAtomic decision changes to align with the overall result.\n\n**CH14-F52 — INFLUENCING THE SECOND ADJUDICATOR BY SHOWING THE FIRST ADJUDICATOR'S DECISION**\n\nIndependent dual adjudication is disrupted.\n\n**CH14-F53 — DELETING INITIAL DECISIONS IN THE CONSENSUS**\n\nThe true disagreement rate becomes invisible.\n\n**CH14-F54 — INVALIDATING THE EXPERT BY MAJORITY VOTE**\n\nDomain expertise is defeated by numerical superiority.\n\n**CH14-F55 — PUBLISHING ONLY GENERAL AGREEMENT**\n\nLow agreement in a subject or reference area is hidden.\n\n**CH14-F56 — CONSIDERING HIGH AGREEMENT AS ABSOLUTE CORRECTNESS**\n\nThe possibility that all adjudicators could have learned the same mistake is ignored.\n\n**CH14-F57 — ADDING CALIBRATION RESPONSES TO THE MAIN SCORE**\n\nAdjudicator training counts as actual observation.\n\n**CH14-F58 — IGNORING THE ADJUDICATOR DRIFT**\n\nIt quietly changes between waves.\n\n**CH14-F59 — CONSIDERING AI SUGGESTION AS FINAL DECISION**\n\nHuman adjudicator only approves the automatic label.\n\n**CH14-F60 — MAKING THE SAME AI ITS OWN ADJUDICATOR**\n\nThe producing system declares its correctness independently.\n\n**CH14-F61 — UNJUSTIFIED NEGATIVE DECISION**\n\nThe adjudicator says “wrong” but shows no evidence or rule.\n\n**CH14-F62 — SILENT ADJUDICATOR CORRECTION**\n\nThe label is changed without a decision history.\n\n**CH14-F63 — TURNING THE APPEAL INTO A CUSTOMER SATISFACTION PROCESS**\n\nInstead of evidence, commercial pressure changes the decision.\n\n**CH14-F64 — REVIEWING THE FIRST ADJUDICATOR'S OWN APPEAL**\n\nNo independent re-evaluation is made.\n\n**CH14-F65 — KEEPING THE OPEN OBJECTION AS FINAL DECISION**\n\nThe pending status does not appear under the public result.\n\n**CH14-F66 — CONCEALING ADJUDICATOR QUALITY LEVEL**\n\nA single, uncalibrated review is presented as a full independent audit.\n\n## 113. AUDIT PROCEDURE\n\n### Step 1 — Verify Capture Validity\n\nOnly observations found to be valid in Chapter 12 are subject to semantic adjudication.\n\n### Step 2 — Link Prompt and Truth Pack Versions\n\nThe response is evaluated under the correct intent and reference time.\n\n### Step 3 — Lock the Full Response\n\nThe response text cannot be changed during adjudication.\n\n### Step 4 — Separate Response Behaviours\n\nRefusal, clarification, error, and empty responses are separated from factual claims.\n\n### Step 5 — Perform Initial Semantic Segmentation\n\nSentences and material expression areas are identified.\n\n### Step 6 — Apply Atomicity Tests\n\nProvisions requiring separate correctness, evidence, existence, and scope are separated.\n\n### Step 7 — Check for Over-Atomisation\n\nMeaningless micro-claims are combined.\n\n### Step 8 — Record Explicit and Implicit Claims\n\nSources of presupposition and entailment are marked.\n\n### Step 9 — Resolve Entity and Coreference\n\nIf there is ambiguity, no definite subject is invented.\n\n### Step 10 — Add Qualifiers\n\nAttribution, modality, quantifier, time, and scope are preserved.\n\n### Step 11 — Assign Claim Centrality\n\nCore, supporting, qualifying, incidental, or non-material status is given.\n\n### Step 12 — Build the Claim Dependency Graph\n\nRoot and derived claims are linked to each other.\n\n### Step 13 — Check Required Response Elements\n\nIt is examined whether the necessary elements for the petition family are present.\n\n### Step 14 — Create Omission Records\n\nOnly material and duty-related deficiencies are recorded.\n\n### Step 15 — Freeze the Claim Ledger\n\nThe claim extraction version is locked before peer review.\n\n### Step 16 — Match the atoms to Truth Pack\n\nThe existing claim, reference gap, or out-of-scope situation is determined.\n\n### Step 17 — Set Up the Citation Map\n\nTo what extent does each citation support which atom?\n\n### Step 18 — The First Adjudicator Gives the Dimensional Decision\n\nEntity, fact, scope, time, attribution, modality, reference and relevance are evaluated.\n\n### Step 19 — Second Adjudicator Gives Independent Decision\n\nThe outcome of the first adjudicator is kept hidden.\n\n### Step 20 — Apply Expert Adjudicator Requirement\n\nHigh-risk or technical claims are sent to a suitable expert.\n\n### Step 21 — Calculate Adjudicator Agreement\n\nDimension-based agreements and types of disagreements are extracted.\n\n### Step 22 — Apply Dispute Resolution\n\nReconciliation is carried out by a Senior Adjudicator or additional expert process.\n\n### Step 23 — Lock the Final Atomic Decision\n\nThe decision vector, combined label, and rationale are recorded.\n\n### Step 24 — Lock Omission and Response Behaviours\n\nNon-claim results are also completed.\n\n### Step 25 — Assign Adjudication Quality Status\n\nAn appropriate level is given between AQ-0 and AQ-5.\n\n### Step 26 — Open Appeal Window\n\nRelevant parties can appeal at the claim and evidence level.\n\n### Step 27 — Process New Evidence or Reference Gap\n\nIf Truth Pack changes, a new adjudication version is created.\n\n### Step 28 — Generate Public Claim Manifest\n\nClaim and decision distribution is published without personal data.\n\n## 114. REQUIRED EVIDENCE\n\nObservation ID; capture-validity record; AI System Register entry; panel status; prompt ID and version; full AI response; response language and locale; Truth Pack ID and version; claim-extraction version; original-text spans; normalised atomic claims; explicit, implicit and presupposed status; entity and coreference records; attribution; modality; quantifier; negation; time and geography; product and service scope; number and unit; claim centrality; claim-dependency graph; Required Response Elements; omission records; citation list; citation–claim map; Truth Pack matches; Reference Gap records; first-adjudicator decision; second-adjudicator decision; subject-matter expert decision; language-adjudicator decision; citation-adjudicator decision; disputes; reconciliation record; Senior Adjudicator decision; decision vectors; and combined atomic labels.\n\nDecision justifications Adjudicator identity or role records Adjudicator qualifications Conflict of interest records Blinding status Inter-adjudicator agreement Raw and post-consensus agreement Calibration set Gold cases Adjudicator drift analysis AI assistant tool records Appeals Re-adjudication Dissenting opinion Decision versions Adjudication quality status Public Claim Manifest Responsible person or institution\n\n## 115. AUDIT CHECKLIST\n\nWas the capture validity given before the semantic evaluation? Does the response depend on the correct claim and the Truth Pack version? Is the full response text preserved? Were refusal and clarification separated from claims? Were sentences automatically counted as a single claim? Were provisions requiring separate evidence separated? Are atoms excessively small? Are severe errors diluted with small truths? Were explicit claims recorded? Were implicit claims extracted only with mandatory meaning? Was the claim accepted by AI separated from the preliminary assumption of the prompt? Is attribution preserved? Are modality and uncertainty preserved? Was negation interpreted correctly? Were expressions like 'all', 'some', 'most' preserved? Were entities and pronouns resolved correctly? Was an indefinite entity chosen arbitrarily?\n\nDid the parent company and brand get mixed up? Is the time frame correct? Is 'Today' linked to the response date? Are geography and language separated? Is the product and service scope maintained? Are number, unit, denominator, and period clear? Are approximate and exact values separated? Is the comparison universe defined? Is the recommendation dependent on the user profile? Was each citation matched separately with every atom? Does the citation support only attribution? Is the citation truly accessible and original? Are the mandatory response elements of the prompt predefined? Was every missing correct information counted as an omission? Was material boundary deficiency recorded? Did the response pass correct but irrelevant content? Is claim centrality independent of the result? Are derivative claims dependent on root claims?\n\nDoes the same root error produce multiple penalties? Is it separate from the claim extraction accuracy decision? Was the Claim Ledger locked before adjudication? Is each atom associated with Truth Pack or a reference gap? Are entity, fact, scope, time, and attribution separate? Are unsupported and contradicted separated? Was the wrong entity treated as correct information? Was historical correct treated as current? Is scope overrun hidden within partial support? Is the epistemic status error visible? Was appropriate uncertainty penalised? Was excessive uncertainty recorded? Does the adjudicator have sufficient proficiency in the relevant language? Does the claim require a domain expert? Is there a conflict of interest for the adjudicator? Did the second adjudicator see the first decision? Are major/critical candidates double-reviewed?\n\nAre dispute records preserved? Was an expert override decided by majority vote? Was the agreement measured by dimension? Was the agreement published only after reconciliation? Did adjudicators use a calibration set? Did calibration records interfere with the main score? Was adjudicator drift checked? Were AI-assisted decisions reviewed by a human? Did the AI that produced the answer become its own final adjudicator? Do negative decisions carry evidence and reasoning? Were decision changes versioned? Was the appeal reviewed by an independent adjudicator? Is the result of a public appeal visible to the public? Is the adjudicator quality level at least AQ-3? Is the accountable owner of the Claim Ledger known?\n\n## 116. APPEALS AND RESPONSES\n\n### Appeal 1 — “Isn't it easier to evaluate the answer sentence by sentence?”\n\nIt is easy. However, a single sentence can carry many independent claims. Some may be true, some may be false. The sentence boundary is not the boundary of adjudication.\n\n### Objection 2 — “Doesn’t separating every claim this much make the system unnecessarily complicated?”\n\nFinancial claims need to be separated. It is not necessary to separate meaningless language fragments. Atomicity tests prevent excessive fragmentation.\n\n### Objection 3 — 'If the main idea in a sentence is correct, can't we ignore small mistakes?'\n\nIt depends on the materiality of the error. Wrong country, licence, price, or warranty: it is not a minor detail. Chapter 15 will determine the gates of its effect.\n\n### Objection 4 — “Isn't it natural for the score to increase with a large number of correct atoms?”\n\nOnly if atoms have the same importance and independence. Insignificant micro-truths cannot dilute a severe false judgement. The relationship between centrality and root error must be maintained.\n\n### Objection 5 — “Aren't implicit claims subjective?”\n\nSome may be. Therefore, only claims that are linguistically necessary or strongly carried are extracted. Disagreements are kept on record.\n\n### Objection 6 — “If the AI reported what the company said correctly, why could it be considered wrong?”\n\nIf attribution is preserved, reporting the self-declaration may be correct. If AI turns the self-declaration into an independent fact, an epistemic error occurs.\n\n### Objection 7 — “If we couldn’t find the company’s licence, why shouldn’t AI saying ‘not licensed’ pass?”\n\nIf it has not been proven that the record is a closed world and a complementary record, the lack of a licence is not certain. The correct answer could be: \"Not verified.\"\n\n### Objection 8 — “If AI speaks very cautiously, won't its usefulness to the user decrease?”\n\nIt can decrease. For this reason, unnecessary uncertainty is also recorded. NOMOS makes the caution as dysfunctional as false certainty visible.\n\n### Objection 9 — “Isn't it enough if the citation goes to the correct source?”\n\nThe source may be correct. But another:\n\n- entity,\n\n- country,\n\n- date,\n\n- claim\n\nIt may be about. The citation–claim matching should also be examined.\n\n### Objection 10 — “Should all Truth Pack information be included in the response?”\n\nNo. Only the mandatory duty of the prompt and the material limits necessary for correct understanding are expected. Truth Pack is not a response template.\n\n### Objection 11 — “If there is no false claim, why should the response fail?”\n\nThe main question may not have been answered or the material limit may have been left incomplete. Accuracy and completeness are separate dimensions.\n\n### Objection 12 — “Are two adjudicators enough?”\n\nThe same structure may not be required for every claim. In high-risk areas, a language adjudicator, a field expert, and a Senior Adjudicator may be required.\n\n### Objection 13 — “If adjudicators agree, why is external review needed?”\n\nAdjudicators may have learned the same incorrect interpretation. Consistency shows alignment. It does not guarantee absolute accuracy.\n\n### Objection 14 — \"Wouldn't AI adjudication be faster and more consistent?”\n\nAI can increase the speed of claim extraction and matching. It can also scale its own bias and error. The ultimate responsibility should remain with humans.\n\n### Objection 15 — 'Aren't human adjudicators subjective too?'\n\nYes. For this reason:\n\n- open source book,\n\n- dual adjudication\n\n- compatibility measurement,\n\n- expertise,\n\n- appeal,\n\n- version record\n\nIt is necessary. The goal is not to deny subjectivity; it is to make it visible and controllable.\n\n### Objection 16 — “Wouldn't adjudication be very expensive?”\n\nIt is possible. A risk-based structure can be used:\n\n- sample quality control in low-risk atoms,\n\n- mandatory double review in high-risk atoms,\n\n- AI-assisted preprocessing,\n\n- human final decision\n\ncan be applied. Cost does not grant the right to make a single adjudicator's undocumented opinion a world standard.\n\n## COMMON RULE OF CHAPTER 121\n\nAn AI response may seem correct because its first sentence is true. It may seem incorrect because it contains a single major error. It may be incomplete because the user has not specified the threshold that would change their decision. It may be cautious because reality itself has not been fully resolved. It may be evasive because, despite strong evidence, it has made no decision. It may appear sourced, but the citation may support a different claim. All of these can coexist within the same response. Therefore, NOMOS does not ask the judge a single question: \"Is this response correct?\" Instead, it asks the following:\n\nWhich claims did he make? / Whom did he say they belonged to? / How decisively did he speak? / Which time did he mean? / Which country and product did he cover? / Did he maintain the status of the source? / What did the citation support? / Did he answer the main task of the prompt? / Which material limit did he omit? / Where was the Truth Pack insufficient? Atomisation is not done to punish the answer. It is also done to preserve the correct parts of the answer. An answer:\n\n- main identity correct,\n\n- scope of activity partially correct,\n\n- leadership claim incorrect,\n\n- price coverage old,\n\n- citation status broken\n\ncan carry. A single zero destroys all correct parts. A single pass hides all mistakes. Atomic adjudication rejects both. However, atomic adjudication can also open a new way of manipulation. Major error:\n\n- there is a company,\n\n- there is a country,\n\n- the word service is correct\n\ncannot be broken into small truths like this and lost. Atoms are not considered alone. Their meanings, centralities, and root relationships protect them. The adjudicator can be human. This does not make them infallible. AI can assist. This does not make it the ultimate authority. Reliable adjudication:\n\n- clear claim scheme,\n\n- correct language,\n\n- domain expertise,\n\n- independent primary decision,\n\n- dispute record,\n\n- appeal,\n\n- version history\n\noccurs with. Therefore, NOMOS's fourteenth measurement law is as follows:\n\n> The correctness of an answer lies not in the overall impression of its sentences; it lies in the separate truth states of the atomic claims it carries.\n\nThe fifteenth law is as follows:\n\n> An atom should be small; it should not be meaningless.\n\nIts sixteenth law is as follows:\n\n> When attribution, scope, time, and uncertainty are removed, the claim is no longer the same claim.\n\nIts seventeenth law is:\n\n> If correct information is linked to a false entity, it is a false representation.\n\nIts eighteenth measurement law states:\n\n> A source mark is not evidence; the support relationship between citation and claim is evidence.\n\nIts nineteenth measurement law states:\n\n> Not speaking falsely is not the same as answering the question.\n\nIts twentieth measurement law states:\n\n> Harmony among adjudicators increases trust; it does not eliminate human error.\n\nIts twenty-first measurement law states:\n\n> Unresolved decision is not weakness; it is respect for the limits of evidence.\n\nIts twenty-second measurement law states:\n\n> If atomic decisions are not transparent, the final score is only a calculated opinion.\n\n## NOMOS’s Section 14 Order\n\n> Do not tell me that your answer is generally correct. / Show under which evidence each of my claims is correct.\n\n> Do not make my sentence a single judgement. / But do not split me into meaningless word fragments either.\n\n> If one part of me is correct and another part is wrong, separate the two.\n\n> Do not lose my major mistake among small truths like “there is a company”.\n\n> When I say 'The company says,' do not turn this into 'It has been proven.'\n\n> Don't act as if I'm speaking definitively when I say 'probably.' / When I say 'maybe' while there is solid evidence, notice that too.\n\n> Keep the difference between \"could not be verified\" and \"does not exist\".\n\n> Do not delete the words all, some, most, and at least one.\n\n> Do not make the parent company and the brand, the product and the manufacturer, the franchise and the main institution a single entity.\n\n> Do not pass yesterday's truth as today's truth.\n\n> Do not consider providing services in a language as a legal operation in that country.\n\n> Do not convert the approximate number to an integer, or the selected case to the general rate.\n\n> Don't believe immediately when you see a citation. / Check which claim the source actually supports.\n\n> Don't make me memorise all of Truth Pack. / But also don't leave out the threshold that will make the user change their mind.\n\n> Don't let me escape the main question with correct but irrelevant information.\n\n> First, find out what I said. / Then see if it is correct.\n\n> Don't try to make the adjudicator love or hate my brand.\n\n> Don't show the first adjudicator's decision to the second adjudicator.\n\n> Do not let a majority of generalist adjudicators overrule the expertise required for a specialist question.\n\n> Don't hide the disagreement. / If it cannot be resolved, leave it unresolved.\n\n> Do not declare high agreement as infallibility. / Continue to test whether the adjudicators may have learned wrongly together.\n\n> You can get help from me. / But don't make me the sole judge of my own answer.\n\n> Do not delete the old decision when an objection is received. / Create a new version with new evidence.\n\nFirst, lock the response. / Then extract its atomic claims. / Then identify entity, scope, time and attribution. / Then map each claim to the Truth Pack. / Then submit it to two independent adjudicators. / Then resolve disagreements. / Only then decide which claims pass and which fail.\n\n## The Chapter's Closing Sentence\n\nFair adjudication in GEO-1000 does not judge a response by general impression. It makes every material claim visible and evaluates it separately against the correct entity, evidence, scope, time and epistemic status.\n\n## Normative Core\n\n> Before any response-level pass, failure, severity, or score is assigned, each material AI response MUST be decomposed into atomic semantic claims. Each atomic claim MUST retain: - its exact source span, - normalised proposition, - subject entity, - predicate and value, - qualifiers and quantifiers, - geographic, product, user, and jurisdictional scope, - temporal frame, - attribution, - modality and uncertainty, - explicit, implicit, presupposed, or entailed origin, - response centrality, - citation relationships, - dependency relationships, - and Truth Pack mapping. Sentence and paragraph boundaries MUST NOT be treated as automatic claim boundaries. Atomic decomposition MUST separate propositions that can carry different truth, evidence, entity, scope, time, or attribution states, while prohibiting meaningless micro-fragmentation that dilutes material error. Claim extraction, Truth Pack mapping, and semantic adjudication SHOULD remain separate stages and, where practical, separate roles. Every atomic claim MUST be adjudicated across distinct dimensions, including: - entity, - factual support, - scope, - time, - attribution and epistemic status, - modality and uncertainty, - citation support, - relevance, - and unresolved or reference-gap status. Official-claim fidelity MUST remain distinct from verified factual support. Unsupported, contradicted, wrong-entity, outdated, scope-overreach, epistemic-status-error, citation-mismatch, reference-gap, and unresolved states MUST remain separately visible. Required response elements and material omissions MUST be adjudicated separately from explicit false claims. Correct but irrelevant information MUST NOT satisfy the prompt merely because it is factually supported. Material claims SHOULD receive independent double review, with additional language or domain expertise where required. Reviewer identity, qualification, conflicts, independent first decisions, disagreements, resolution, reliability metrics, appeals, and decision versions MUST be recorded. Reviewer agreement MUST be measured by dimension and MUST NOT be treated as proof that the shared decision is necessarily correct. AI systems MAY assist with segmentation, claim extraction, evidence mapping, translation support, and contradiction discovery, but MUST NOT serve as the sole final adjudicator of their own or another system's claims. Atomic adjudication decisions MUST remain independent of commercial interests, desired scores, provider reputation, badge outcomes, or client approval. Every Claim Ledger, adjudication decision, disagreement resolution, appeal, and revision MUST be versioned and attributable to an accountable human or organisation.","character_count":100053,"record_sha256":"48c1add8c30d2d3b2690d50eb4e01cd55706cbf36e01f612da57dc4adba96956"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-15","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":17,"chapter_number":15,"item_number":null,"title":"Response Passing and Critical Error Gates","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K03","K20"],"source_word_count":11600,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":15,\"id\":\"NOMOS-GEO-AUDIT-CH15\",\"title\":\"Response Passing and Critical Error Gates\",\"subtitle\":null,\"sourceFile\":\"15.ci bölüm.docx\",\"sourceSha256\":\"411AD7240920CEE5037847CF01460DF1E4E13EBEC5234EE0F73B43E349071DEB\",\"sourceWordCount\":11600,\"sourceIds\":[\"K03\",\"K20\"],\"machine\":{\"chapter\":15,\"chapterId\":\"NOMOS-GEO-AUDIT-CH15\",\"title\":\"Response Passing and Critical Error Gates\",\"subtitle\":null,\"sourceIds\":[\"K03\",\"K20\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH15-R01\",\"normativeRuleEnglish\":\"A GEO-1000 response MUST NOT receive a pass decision solely from its average or proportion of supported atomic claims. Before response-level adjudication, the observation MUST satisfy the required capture, prompt, AI-system, Truth Pack, and adjudication-quality conditions. Every response MUST be evaluated for: - required response elements, - confirmed Critical findings, - confirmed Major findings, - cumulative Moderate materiality, - material omissions, - unresolved core claims, - refusal or nonresponse, - internal contradiction, - self-correction, - and root-finding clusters. A confirmed Critical finding MUST create a non-compensatory automatic response failure. A confirmed Major finding or failure of a required core response element MUST create a response failure and MUST NOT be offset by other supported claims. Moderate findings MAY permit only a conditional pass when all core requirements remain satisfied and cumulative materiality does not create a Major effect. Advisory findings MUST remain visible but MUST NOT independently block response passage. Critical and Major severity MUST be determined independently from claim frequency and independently from whether the error favours or harms the audited entity. Critical findings SHOULD be confirmed through valid capture, sufficient Truth Pack coverage, independent double review, and appropriate language or domain expertise. Reference gaps and unresolved evidence MUST NOT automatically create Critical, Major, pass, or failure decisions. Wrong-entity transfer, false license or regulatory authority, actionable high-stakes misinformation, fabricated material evidence, false transactional guarantees, unsafe out-of-scope recommendations, serious unsupported allegations, material boundary omissions, current use of expired authority, false endorsements, restricted-data exposure, and evidence laundering MAY activate Critical gates when the defined materiality conditions are met. A single confirmed Critical response MUST establish that the event occurred under the defined product-country-language-prompt-time conditions. It MUST NOT, by itself, be represented as the prevalence of that event across all users or all product surfaces. Every confirmed Critical response MUST open a versioned Critical Incident Record and trigger scope-appropriate confirmatory review. An open confirmed Critical incident MUST prevent a claim of zero Critical errors or full conformity in the affected declared scope. Remediation and retesting MUST create new observations and new wave records. They MUST NOT alter or erase the original response decision or historical incident. Response decisions, severity levels, gate codes, incident scope, replication status, conformity effects, appeals, and revisions MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Bir GEO-1000 yanıtı yalnız desteklenmiş atomların ortalaması veya oranı üzerinden geçiş alamaz. Confirmed Critical finding telafi edilemeyen otomatik response başarısızlığı; Confirmed Major finding veya core zorunlu unsur başarısızlığı response başarısızlığı oluşturmalıdır. Moderate finding ancak core gereksinimleri korunuyor ve birikimli maddilik Major etki oluşturmuyorsa koşullu geçişe izin verebilir.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH15-MB0001\",\"type\":\"paragraph\",\"text\":\"79. MACHINE-READABLE RESPONSE TRANSITION RECORD\",\"sourceParagraph\":2176},{\"blockId\":\"CH15-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2177},{\"blockId\":\"CH15-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2178},{\"blockId\":\"CH15-MB0004\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2179},{\"blockId\":\"CH15-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"requiredResponseElements\\\": {\",\"sourceParagraph\":2180},{\"blockId\":\"CH15-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"entityResolved\\\": true,\",\"sourceParagraph\":2181},{\"blockId\":\"CH15-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"primaryActivityPresent\\\": true,\",\"sourceParagraph\":2182},{\"blockId\":\"CH15-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"materialBoundaryPresent\\\": false,\",\"sourceParagraph\":2183},{\"blockId\":\"CH15-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"allCoreElementsSatisfied\\\": false\",\"sourceParagraph\":2184},{\"blockId\":\"CH15-MB0010\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2185},{\"blockId\":\"CH15-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"claimSummary\\\": {\",\"sourceParagraph\":2186},{\"blockId\":\"CH15-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"materialAtomicClaimCount\\\": 14,\",\"sourceParagraph\":2187},{\"blockId\":\"CH15-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"supported\\\": 9,\",\"sourceParagraph\":2188},{\"blockId\":\"CH15-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"supportedWithQualification\\\": 1,\",\"sourceParagraph\":2189},{\"blockId\":\"CH15-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"unsupported\\\": 1,\",\"sourceParagraph\":2190},{\"blockId\":\"CH15-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"contradicted\\\": 2,\",\"sourceParagraph\":2191},{\"blockId\":\"CH15-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"referenceGap\\\": 0,\",\"sourceParagraph\":2192},{\"blockId\":\"CH15-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"wrongEntity\\\": 0,\",\"sourceParagraph\":2193},{\"blockId\":\"CH15-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"outdated\\\": 0,\",\"sourceParagraph\":2194},{\"blockId\":\"CH15-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"claimSupportRateUnweighted\\\": 0.6429\",\"sourceParagraph\":2195},{\"blockId\":\"CH15-MB0021\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2196},{\"blockId\":\"CH15-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"findingClusters\\\": [\",\"sourceParagraph\":2197},{\"blockId\":\"CH15-MB0023\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2198},{\"blockId\":\"CH15-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"findingClusterId\\\": \\\"NGFC-SYNTH-000184-001\\\",\",\"sourceParagraph\":2199},{\"blockId\":\"CH15-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"rootIssue\\\": \\\"FALSE_IMMIGRATION_LICENSE\\\",\",\"sourceParagraph\":2200},{\"blockId\":\"CH15-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"linkedResponseClaimIds\\\": [\",\"sourceParagraph\":2201},{\"blockId\":\"CH15-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"NGRC-SYNTH-000184-005\\\",\",\"sourceParagraph\":2202},{\"blockId\":\"CH15-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"NGRC-SYNTH-000184-006\\\"\",\"sourceParagraph\":2203},{\"blockId\":\"CH15-MB0029\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2204},{\"blockId\":\"CH15-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"importance\\\": \\\"CRITICAL\\\",\",\"sourceParagraph\":2205},{\"blockId\":\"CH15-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"criticalGateCodes\\\": [\",\"sourceParagraph\":2206},{\"blockId\":\"CH15-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"CG-02\\\",\",\"sourceParagraph\":2207},{\"blockId\":\"CH15-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"CG-06\\\",\",\"sourceParagraph\":2208},{\"blockId\":\"CH15-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"CG-08\\\"\",\"sourceParagraph\":2209},{\"blockId\":\"CH15-MB0035\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2210},{\"blockId\":\"CH15-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"centrality\\\": \\\"CL-1\\\",\",\"sourceParagraph\":2211},{\"blockId\":\"CH15-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"actionability\\\": \\\"DIRECT_USER_ACTION\\\",\",\"sourceParagraph\":2212},{\"blockId\":\"CH15-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"harmDomain\\\": \\\"LEGAL\\\",\",\"sourceParagraph\":2213},{\"blockId\":\"CH15-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"repeatedInResponse\\\": true,\",\"sourceParagraph\":2214},{\"blockId\":\"CH15-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"recommendationAmplified\\\": true\",\"sourceParagraph\":2215},{\"blockId\":\"CH15-MB0041\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2216},{\"blockId\":\"CH15-MB0042\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2217},{\"blockId\":\"CH15-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"findingClusterId\\\": \\\"NGFC-SYNTH-000184-002\\\",\",\"sourceParagraph\":2218},{\"blockId\":\"CH15-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"rootIssue\\\": \\\"FALSE_INDEPENDENT_CERTIFICATION\\\",\",\"sourceParagraph\":2219},{\"blockId\":\"CH15-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"linkedResponseClaimIds\\\": [\",\"sourceParagraph\":2220},{\"blockId\":\"CH15-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"NGRC-SYNTH-000184-009\\\",\",\"sourceParagraph\":2221},{\"blockId\":\"CH15-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"NGRC-SYNTH-000184-010\\\"\",\"sourceParagraph\":2222},{\"blockId\":\"CH15-MB0048\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2223},{\"blockId\":\"CH15-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"importance\\\": \\\"CRITICAL\\\",\",\"sourceParagraph\":2224},{\"blockId\":\"CH15-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"criticalGateCodes\\\": [\",\"sourceParagraph\":2225},{\"blockId\":\"CH15-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"CG-04\\\",\",\"sourceParagraph\":2226},{\"blockId\":\"CH15-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"CG-10\\\",\",\"sourceParagraph\":2227},{\"blockId\":\"CH15-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"CG-12\\\"\",\"sourceParagraph\":2228},{\"blockId\":\"CH15-MB0054\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2229},{\"blockId\":\"CH15-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"centrality\\\": \\\"CL-2\\\",\",\"sourceParagraph\":2230},{\"blockId\":\"CH15-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"actionability\\\": \\\"TRUST_SIGNAL\\\",\",\"sourceParagraph\":2231},{\"blockId\":\"CH15-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"harmDomain\\\": \\\"COMMERCIAL_TRUST\\\"\",\"sourceParagraph\":2232},{\"blockId\":\"CH15-MB0058\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2233},{\"blockId\":\"CH15-MB0059\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2234},{\"blockId\":\"CH15-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"severityCounts\\\": {\",\"sourceParagraph\":2235},{\"blockId\":\"CH15-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"criticalClusters\\\": 2,\",\"sourceParagraph\":2236},{\"blockId\":\"CH15-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"majorClusters\\\": 0,\",\"sourceParagraph\":2237},{\"blockId\":\"CH15-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"moderateClusters\\\": 1,\",\"sourceParagraph\":2238},{\"blockId\":\"CH15-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"advisoryFindings\\\": 0\",\"sourceParagraph\":2239},{\"blockId\":\"CH15-MB0065\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2240},{\"blockId\":\"CH15-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"unresolved\\\": {\",\"sourceParagraph\":2241},{\"blockId\":\"CH15-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"materialUnresolvedClaimCount\\\": 0,\",\"sourceParagraph\":2242},{\"blockId\":\"CH15-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"coreUnresolved\\\": false\",\"sourceParagraph\":2243},{\"blockId\":\"CH15-MB0069\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2244},{\"blockId\":\"CH15-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"responseBehavior\\\": {\",\"sourceParagraph\":2245},{\"blockId\":\"CH15-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"refusal\\\": false,\",\"sourceParagraph\":2246},{\"blockId\":\"CH15-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"clarification\\\": false,\",\"sourceParagraph\":2247},{\"blockId\":\"CH15-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"selfCorrection\\\": false,\",\"sourceParagraph\":2248},{\"blockId\":\"CH15-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"internalContradiction\\\": false,\",\"sourceParagraph\":2249},{\"blockId\":\"CH15-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"noUsableResponse\\\": false\",\"sourceParagraph\":2250},{\"blockId\":\"CH15-MB0076\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2251},{\"blockId\":\"CH15-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"statusDecision\\\": {\",\"sourceParagraph\":2252},{\"blockId\":\"CH15-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"RP-5\\\",\",\"sourceParagraph\":2253},{\"blockId\":\"CH15-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"label\\\": \\\"AUTOMATIC_FAIL_CRITICAL\\\",\",\"sourceParagraph\":2254},{\"blockId\":\"CH15-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"decisionRule\\\": \\\"CONFIRMED_CRITICAL_GATE_PRESENT\\\",\",\"sourceParagraph\":2255},{\"blockId\":\"CH15-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"nonCompensatory\\\": true,\",\"sourceParagraph\":2256},{\"blockId\":\"CH15-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"correctClaimsPreservedInRecord\\\": true\",\"sourceParagraph\":2257},{\"blockId\":\"CH15-MB0083\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2258},{\"blockId\":\"CH15-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"review\\\": {\",\"sourceParagraph\":2259},{\"blockId\":\"CH15-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"primaryReviewerRole\\\": \\\"LANGUAGE_ADJUDICATOR\\\",\",\"sourceParagraph\":2260},{\"blockId\":\"CH15-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"secondaryReviewerRole\\\": \\\"EVIDENCE_ADJUDICATOR\\\",\",\"sourceParagraph\":2261},{\"blockId\":\"CH15-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"domainExpertRole\\\": \\\"LEGAL_REGULATORY_REVIEWER\\\",\",\"sourceParagraph\":2262},{\"blockId\":\"CH15-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"independentInitialDecisions\\\": true,\",\"sourceParagraph\":2263},{\"blockId\":\"CH15-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"finalizedByRole\\\": \\\"SENIOR_ADJUDICATOR\\\",\",\"sourceParagraph\":2264},{\"blockId\":\"CH15-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"finalizedAt\\\": \\\"2026-09-05T09:00:00Z\\\"\",\"sourceParagraph\":2265},{\"blockId\":\"CH15-MB0091\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2266},{\"blockId\":\"CH15-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"appeal\\\": {\",\"sourceParagraph\":2267},{\"blockId\":\"CH15-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"NO_APPEAL\\\"\",\"sourceParagraph\":2268},{\"blockId\":\"CH15-MB0094\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2269},{\"blockId\":\"CH15-MB0095\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2270},{\"blockId\":\"CH15-MB0096\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2271},{\"blockId\":\"CH15-MB0097\",\"type\":\"paragraph\",\"text\":\"80. MACHINE-READABLE CRITICAL INCIDENT RECORD\",\"sourceParagraph\":2273},{\"blockId\":\"CH15-MB0098\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2274},{\"blockId\":\"CH15-MB0099\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2275},{\"blockId\":\"CH15-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2276},{\"blockId\":\"CH15-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2277},{\"blockId\":\"CH15-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"criticalIncident\\\": {\",\"sourceParagraph\":2278},{\"blockId\":\"CH15-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"incidentId\\\": \\\"NGCI-SYNTH-ORION-TR-W1-001\\\",\",\"sourceParagraph\":2279},{\"blockId\":\"CH15-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"CIS-2\\\",\",\"sourceParagraph\":2280},{\"blockId\":\"CH15-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"incidentVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":2281},{\"blockId\":\"CH15-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"sourceResponseDecisionId\\\": \\\"NGRD-SYNTH-000184\\\",\",\"sourceParagraph\":2282},{\"blockId\":\"CH15-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"sourceFindingClusterId\\\": \\\"NGFC-SYNTH-000184-001\\\",\",\"sourceParagraph\":2283},{\"blockId\":\"CH15-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"criticalGateCodes\\\": [\",\"sourceParagraph\":2284},{\"blockId\":\"CH15-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"CG-02\\\",\",\"sourceParagraph\":2285},{\"blockId\":\"CH15-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"CG-06\\\",\",\"sourceParagraph\":2286},{\"blockId\":\"CH15-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"CG-08\\\"\",\"sourceParagraph\":2287},{\"blockId\":\"CH15-MB0112\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2288},{\"blockId\":\"CH15-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"incidentType\\\": \\\"FALSE_REGULATED_SERVICE_AUTHORITY\\\",\",\"sourceParagraph\":2289},{\"blockId\":\"CH15-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"scope\\\": {\",\"sourceParagraph\":2290},{\"blockId\":\"CH15-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-ORION-CHAT\\\",\",\"sourceParagraph\":2291},{\"blockId\":\"CH15-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"surface\\\": \\\"WEB\\\",\",\"sourceParagraph\":2292},{\"blockId\":\"CH15-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"plan\\\": \\\"PAID\\\",\",\"sourceParagraph\":2293},{\"blockId\":\"CH15-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"country\\\": \\\"TR\\\",\",\"sourceParagraph\":2294},{\"blockId\":\"CH15-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"locale\\\": \\\"tr-TR\\\",\",\"sourceParagraph\":2295},{\"blockId\":\"CH15-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"promptFamily\\\": \\\"PR-07\\\",\",\"sourceParagraph\":2296},{\"blockId\":\"CH15-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"panelState\\\": \\\"CONTROLLED_CLEAN\\\",\",\"sourceParagraph\":2297},{\"blockId\":\"CH15-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"NGW-SYNTH-W1-001\\\"\",\"sourceParagraph\":2298},{\"blockId\":\"CH15-MB0123\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2299},{\"blockId\":\"CH15-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"confirmation\\\": {\",\"sourceParagraph\":2300},{\"blockId\":\"CH15-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"captureValid\\\": true,\",\"sourceParagraph\":2301},{\"blockId\":\"CH15-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackSufficient\\\": true,\",\"sourceParagraph\":2302},{\"blockId\":\"CH15-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"doubleReviewed\\\": true,\",\"sourceParagraph\":2303},{\"blockId\":\"CH15-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"domainExpertConfirmed\\\": true,\",\"sourceParagraph\":2304},{\"blockId\":\"CH15-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"referenceGap\\\": false,\",\"sourceParagraph\":2305},{\"blockId\":\"CH15-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"confirmedAt\\\": \\\"2026-09-05T09:00:00Z\\\"\",\"sourceParagraph\":2306},{\"blockId\":\"CH15-MB0131\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2307},{\"blockId\":\"CH15-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"impactVector\\\": {\",\"sourceParagraph\":2308},{\"blockId\":\"CH15-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"harmMagnitude\\\": \\\"HIGH\\\",\",\"sourceParagraph\":2309},{\"blockId\":\"CH15-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"decisionImpact\\\": \\\"HIGH\\\",\",\"sourceParagraph\":2310},{\"blockId\":\"CH15-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"actionability\\\": \\\"DIRECT\\\",\",\"sourceParagraph\":2311},{\"blockId\":\"CH15-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"centrality\\\": \\\"CORE\\\",\",\"sourceParagraph\":2312},{\"blockId\":\"CH15-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"affectedDomain\\\": \\\"LEGAL\\\",\",\"sourceParagraph\":2313},{\"blockId\":\"CH15-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"reversibilityBeforeAction\\\": \\\"LOW\\\",\",\"sourceParagraph\":2314},{\"blockId\":\"CH15-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"vulnerableUserRisk\\\": \\\"POSSIBLE\\\"\",\"sourceParagraph\":2315},{\"blockId\":\"CH15-MB0140\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2316},{\"blockId\":\"CH15-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"populationInference\\\": {\",\"sourceParagraph\":2317},{\"blockId\":\"CH15-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"singleObservationProvesPrevalence\\\": false,\",\"sourceParagraph\":2318},{\"blockId\":\"CH15-MB0143\",\"type\":\"paragraph\",\"text\":\"\\\"cellPrevalenceStatus\\\": \\\"UNKNOWN\\\",\",\"sourceParagraph\":2319},{\"blockId\":\"CH15-MB0144\",\"type\":\"paragraph\",\"text\":\"\\\"confirmatorySamplingRequired\\\": true\",\"sourceParagraph\":2320},{\"blockId\":\"CH15-MB0145\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2321},{\"blockId\":\"CH15-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"conformityEffect\\\": {\",\"sourceParagraph\":2322},{\"blockId\":\"CH15-MB0147\",\"type\":\"paragraph\",\"text\":\"\\\"responseAutomaticFail\\\": true,\",\"sourceParagraph\":2323},{\"blockId\":\"CH15-MB0148\",\"type\":\"paragraph\",\"text\":\"\\\"cellStatus\\\": \\\"CRITICAL_HOLD\\\",\",\"sourceParagraph\":2324},{\"blockId\":\"CH15-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"waveZeroCriticalClaimAllowed\\\": false,\",\"sourceParagraph\":2325},{\"blockId\":\"CH15-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"globalProductFailureClaimAllowed\\\": false\",\"sourceParagraph\":2326},{\"blockId\":\"CH15-MB0151\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2327},{\"blockId\":\"CH15-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"confirmatoryPlan\\\": {\",\"sourceParagraph\":2328},{\"blockId\":\"CH15-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"planId\\\": \\\"NGCIP-SYNTH-ORION-TR-001\\\",\",\"sourceParagraph\":2329},{\"blockId\":\"CH15-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeAdditionalResponses\\\": true,\",\"sourceParagraph\":2330},{\"blockId\":\"CH15-MB0155\",\"type\":\"paragraph\",\"text\":\"\\\"independentUsers\\\": 100,\",\"sourceParagraph\":2331},{\"blockId\":\"CH15-MB0156\",\"type\":\"paragraph\",\"text\":\"\\\"samePromptVersion\\\": true,\",\"sourceParagraph\":2332},{\"blockId\":\"CH15-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"sameTruthPackVersion\\\": true,\",\"sourceParagraph\":2333},{\"blockId\":\"CH15-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"crossLanguageExtension\\\": false\",\"sourceParagraph\":2334},{\"blockId\":\"CH15-MB0159\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2335},{\"blockId\":\"CH15-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"remediation\\\": {\",\"sourceParagraph\":2336},{\"blockId\":\"CH15-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"NOT_STARTED\\\",\",\"sourceParagraph\":2337},{\"blockId\":\"CH15-MB0162\",\"type\":\"paragraph\",\"text\":\"\\\"ownerType\\\": \\\"TO_BE_DETERMINED\\\",\",\"sourceParagraph\":2338},{\"blockId\":\"CH15-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"historicalRecordMustRemain\\\": true\",\"sourceParagraph\":2339},{\"blockId\":\"CH15-MB0164\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2340},{\"blockId\":\"CH15-MB0165\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":2341},{\"blockId\":\"CH15-MB0166\",\"type\":\"paragraph\",\"text\":\"\\\"incidentOwnerRole\\\": \\\"CRITICAL_INCIDENT_REVIEW_LEAD\\\",\",\"sourceParagraph\":2342},{\"blockId\":\"CH15-MB0167\",\"type\":\"paragraph\",\"text\":\"\\\"publicManifestRequired\\\": true\",\"sourceParagraph\":2343},{\"blockId\":\"CH15-MB0168\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2344},{\"blockId\":\"CH15-MB0169\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2345},{\"blockId\":\"CH15-MB0170\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2346},{\"blockId\":\"CH15-MB0171\",\"type\":\"paragraph\",\"text\":\"81. MACHINE-READABLE RESPONSE GATE RULE\",\"sourceParagraph\":2348},{\"blockId\":\"CH15-MB0172\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH15-R01\",\"sourceParagraph\":2349},{\"blockId\":\"CH15-MB0173\",\"type\":\"paragraph\",\"text\":\"A GEO-1000 response MUST NOT receive a pass decision solely from its\",\"sourceParagraph\":2351},{\"blockId\":\"CH15-MB0174\",\"type\":\"paragraph\",\"text\":\"average or proportion of supported atomic claims.\",\"sourceParagraph\":2352},{\"blockId\":\"CH15-MB0175\",\"type\":\"paragraph\",\"text\":\"Before response-level adjudication, the observation MUST satisfy the\",\"sourceParagraph\":2354},{\"blockId\":\"CH15-MB0176\",\"type\":\"paragraph\",\"text\":\"required capture, prompt, AI-system, Truth Pack, and adjudication-quality\",\"sourceParagraph\":2355},{\"blockId\":\"CH15-MB0177\",\"type\":\"paragraph\",\"text\":\"conditions.\",\"sourceParagraph\":2356},{\"blockId\":\"CH15-MB0178\",\"type\":\"paragraph\",\"text\":\"Every response MUST be evaluated for:\",\"sourceParagraph\":2358},{\"blockId\":\"CH15-MB0179\",\"type\":\"paragraph\",\"text\":\"- required response elements,\",\"sourceParagraph\":2360},{\"blockId\":\"CH15-MB0180\",\"type\":\"paragraph\",\"text\":\"- confirmed Critical findings,\",\"sourceParagraph\":2361},{\"blockId\":\"CH15-MB0181\",\"type\":\"paragraph\",\"text\":\"- confirmed Major findings,\",\"sourceParagraph\":2362},{\"blockId\":\"CH15-MB0182\",\"type\":\"paragraph\",\"text\":\"- cumulative Moderate materiality,\",\"sourceParagraph\":2363},{\"blockId\":\"CH15-MB0183\",\"type\":\"paragraph\",\"text\":\"- material omissions,\",\"sourceParagraph\":2364},{\"blockId\":\"CH15-MB0184\",\"type\":\"paragraph\",\"text\":\"- unresolved core claims,\",\"sourceParagraph\":2365},{\"blockId\":\"CH15-MB0185\",\"type\":\"paragraph\",\"text\":\"- refusal or nonresponse,\",\"sourceParagraph\":2366},{\"blockId\":\"CH15-MB0186\",\"type\":\"paragraph\",\"text\":\"- internal contradiction,\",\"sourceParagraph\":2367},{\"blockId\":\"CH15-MB0187\",\"type\":\"paragraph\",\"text\":\"- self-correction,\",\"sourceParagraph\":2368},{\"blockId\":\"CH15-MB0188\",\"type\":\"paragraph\",\"text\":\"- and root-finding clusters.\",\"sourceParagraph\":2369},{\"blockId\":\"CH15-MB0189\",\"type\":\"paragraph\",\"text\":\"A confirmed Critical finding MUST create a non-compensatory automatic\",\"sourceParagraph\":2371},{\"blockId\":\"CH15-MB0190\",\"type\":\"paragraph\",\"text\":\"response failure.\",\"sourceParagraph\":2372},{\"blockId\":\"CH15-MB0191\",\"type\":\"paragraph\",\"text\":\"A confirmed Major finding or failure of a required core response element\",\"sourceParagraph\":2374},{\"blockId\":\"CH15-MB0192\",\"type\":\"paragraph\",\"text\":\"MUST create a response failure and MUST NOT be offset by other supported\",\"sourceParagraph\":2375},{\"blockId\":\"CH15-MB0193\",\"type\":\"paragraph\",\"text\":\"claims.\",\"sourceParagraph\":2376},{\"blockId\":\"CH15-MB0194\",\"type\":\"paragraph\",\"text\":\"Moderate findings MAY permit only a conditional pass when all core\",\"sourceParagraph\":2378},{\"blockId\":\"CH15-MB0195\",\"type\":\"paragraph\",\"text\":\"requirements remain satisfied and cumulative materiality does not create\",\"sourceParagraph\":2379},{\"blockId\":\"CH15-MB0196\",\"type\":\"paragraph\",\"text\":\"a Major effect.\",\"sourceParagraph\":2380},{\"blockId\":\"CH15-MB0197\",\"type\":\"paragraph\",\"text\":\"Advisory findings MUST remain visible but MUST NOT independently block\",\"sourceParagraph\":2382},{\"blockId\":\"CH15-MB0198\",\"type\":\"paragraph\",\"text\":\"response passage.\",\"sourceParagraph\":2383},{\"blockId\":\"CH15-MB0199\",\"type\":\"paragraph\",\"text\":\"Critical and Major severity MUST be determined independently from claim\",\"sourceParagraph\":2385},{\"blockId\":\"CH15-MB0200\",\"type\":\"paragraph\",\"text\":\"frequency and independently from whether the error favours or harms the\",\"sourceParagraph\":2386},{\"blockId\":\"CH15-MB0201\",\"type\":\"paragraph\",\"text\":\"audited entity.\",\"sourceParagraph\":2387},{\"blockId\":\"CH15-MB0202\",\"type\":\"paragraph\",\"text\":\"Critical findings SHOULD be confirmed through valid capture, sufficient\",\"sourceParagraph\":2389},{\"blockId\":\"CH15-MB0203\",\"type\":\"paragraph\",\"text\":\"Truth Pack coverage, independent double review, and appropriate language\",\"sourceParagraph\":2390},{\"blockId\":\"CH15-MB0204\",\"type\":\"paragraph\",\"text\":\"or domain expertise.\",\"sourceParagraph\":2391},{\"blockId\":\"CH15-MB0205\",\"type\":\"paragraph\",\"text\":\"Reference gaps and unresolved evidence MUST NOT automatically create\",\"sourceParagraph\":2393},{\"blockId\":\"CH15-MB0206\",\"type\":\"paragraph\",\"text\":\"Critical, Major, pass, or failure decisions.\",\"sourceParagraph\":2394},{\"blockId\":\"CH15-MB0207\",\"type\":\"paragraph\",\"text\":\"Wrong-entity transfer, false license or regulatory authority, actionable\",\"sourceParagraph\":2396},{\"blockId\":\"CH15-MB0208\",\"type\":\"paragraph\",\"text\":\"high-stakes misinformation, fabricated material evidence, false\",\"sourceParagraph\":2397},{\"blockId\":\"CH15-MB0209\",\"type\":\"paragraph\",\"text\":\"transactional guarantees, unsafe out-of-scope recommendations, serious\",\"sourceParagraph\":2398},{\"blockId\":\"CH15-MB0210\",\"type\":\"paragraph\",\"text\":\"unsupported allegations, material boundary omissions, current use of\",\"sourceParagraph\":2399},{\"blockId\":\"CH15-MB0211\",\"type\":\"paragraph\",\"text\":\"expired authority, false endorsements, restricted-data exposure, and\",\"sourceParagraph\":2400},{\"blockId\":\"CH15-MB0212\",\"type\":\"paragraph\",\"text\":\"evidence laundering MAY activate Critical gates when the defined\",\"sourceParagraph\":2401},{\"blockId\":\"CH15-MB0213\",\"type\":\"paragraph\",\"text\":\"materiality conditions are met.\",\"sourceParagraph\":2402},{\"blockId\":\"CH15-MB0214\",\"type\":\"paragraph\",\"text\":\"A single confirmed Critical response MUST establish that the event\",\"sourceParagraph\":2404},{\"blockId\":\"CH15-MB0215\",\"type\":\"paragraph\",\"text\":\"occurred under the defined product-country-language-prompt-time\",\"sourceParagraph\":2405},{\"blockId\":\"CH15-MB0216\",\"type\":\"paragraph\",\"text\":\"conditions. It MUST NOT, by itself, be represented as the prevalence of\",\"sourceParagraph\":2406},{\"blockId\":\"CH15-MB0217\",\"type\":\"paragraph\",\"text\":\"that event across all users or all product surfaces.\",\"sourceParagraph\":2407},{\"blockId\":\"CH15-MB0218\",\"type\":\"paragraph\",\"text\":\"Every confirmed Critical response MUST open a versioned Critical Incident\",\"sourceParagraph\":2409},{\"blockId\":\"CH15-MB0219\",\"type\":\"paragraph\",\"text\":\"Record and trigger scope-appropriate confirmatory review.\",\"sourceParagraph\":2410},{\"blockId\":\"CH15-MB0220\",\"type\":\"paragraph\",\"text\":\"An open confirmed Critical incident MUST prevent a claim of zero Critical\",\"sourceParagraph\":2412},{\"blockId\":\"CH15-MB0221\",\"type\":\"paragraph\",\"text\":\"errors or full conformity in the affected declared scope.\",\"sourceParagraph\":2413},{\"blockId\":\"CH15-MB0222\",\"type\":\"paragraph\",\"text\":\"Remediation and retesting MUST create new observations and new wave\",\"sourceParagraph\":2415},{\"blockId\":\"CH15-MB0223\",\"type\":\"paragraph\",\"text\":\"records. They MUST NOT alter or erase the original response decision or\",\"sourceParagraph\":2416},{\"blockId\":\"CH15-MB0224\",\"type\":\"paragraph\",\"text\":\"historical incident.\",\"sourceParagraph\":2417},{\"blockId\":\"CH15-MB0225\",\"type\":\"paragraph\",\"text\":\"Response decisions, severity levels, gate codes, incident scope,\",\"sourceParagraph\":2419},{\"blockId\":\"CH15-MB0226\",\"type\":\"paragraph\",\"text\":\"replication status, conformity effects, appeals, and revisions MUST be\",\"sourceParagraph\":2420},{\"blockId\":\"CH15-MB0227\",\"type\":\"paragraph\",\"text\":\"versioned and attributable to an accountable human or organisation.\",\"sourceParagraph\":2421},{\"blockId\":\"CH15-MB0228\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2422},{\"blockId\":\"CH15-MB0229\",\"type\":\"paragraph\",\"text\":\"A GEO-1000 response cannot pass solely based on the average or ratio of supported atoms. Confirmed Critical finding is an irremediable automatic response failure; Confirmed Major finding or core mandatory element failure should create a response failure. Moderate finding may allow conditional passage only if core requirements are maintained and cumulative materiality does not create a Major impact.\",\"sourceParagraph\":2423}]}}","text":"## Chapter Boundary\n\nSection 14 established the following provision:\n\n> An AI response should be evaluated not by general impression, but by the individual assessments of the atomic claims it carries within correct existence, correct evidence, correct scope, correct timing, and correct epistemic status.\n\nNow we need to transform the atomic decisions into a conclusion at the response level:\n\n> When does a response pass?\n\n> When does it conditionally pass?\n\n> When does it fail?\n\n> Can a single serious mistake get lost among dozens of correct claims?\n\n> If 95% of a response is correct but the remaining 5% involves a false licence or a claim of incorrect legal authority, can the response be considered successful?\n\n> Can a response that does not provide incorrect information but also does not answer the main question pass?\n\n> What happens if an AI product, while finding the right company and describing the right activities, says 'world leader' without evidence in the same answer?\n\n> If a recommendation answer directs a purchase without saying that the service is not available in the user's country, is this just a shortcoming?\n\n> Does observing a Critical error in a single user mean that the entire AI product has failed?\n\nThese questions cannot be solved with a single score formula. Because not all errors are compensable. In the response:\n\n- twenty supported atoms,\n\n- one false licence claim\n\nlet's assume exist. The simple atom average:\n\n20/21 = 95.2%\n\nwould be. However, the user, trusting the false licence claim:\n\n- can purchase legal services,\n\n- can access health services,\n\n- can use a regulated financial product,\n\nIt may provide personal or commercial information to an unauthorised institution. Twenty small truths do not negate the decision impact of this single falsehood. Otherwise, AI can only give the following short answer: “Asteron Travel is a corporate travel brand owned by Asteron Holdings.” The response carries only one or two atoms. If Core Mirror Prompt has:\n\n- as true entity,\n\n- correct relationship,\n\n- main activity\n\nIf the response satisfies the correct-entity, correct-relationship and core-activity requirements, it cannot be penalised merely for being short. Length does not automatically improve GEO quality, and brevity does not automatically make a response incomplete. This chapter converts atomic decisions into a response-level usability and conformity judgement without dissolving them into an average. The preceding audit architecture proposed four severity levels: Critical — directly undermines representational reliability and creates automatic failure; Major — creates serious misrepresentation and prevents full conformity until corrected; Moderate — reduces quality or intelligibility and may permit conditional passage; Advisory — identifies an improvement opportunity and does not independently prevent conformity. It also proposed publishing an evidence-based audit result as an explicit distribution such as ‘0 Critical, 0 Major, 3 remediable Moderate findings.’\n\nIt had envisaged it being published with such an explicit error distribution. This section converts that initial importance architecture into normative gates working on:\n\n- atomic claim,\n\n- response,\n\n- country–language–product cell,\n\n- measurement wave,\n\n- scope of conformity\n\nThis section:\n\n- distinction between degree of importance and accuracy status,\n\n- Critical, Major, Moderate, and Advisory findings,\n\n- irrecoverable error gates,\n\n- necessary response elements,\n\n- response transition statuses,\n\n- relationship between false claim and material omission,\n\n- citation and epistemic status gates,\n\n- advice and transaction risk,\n\n- approach to positive and negative errors equally,\n\n- internal contradiction and self-correction,\n\n- how to manage recurring root errors,\n\n- distinction between single Critical event and systemic Critical rate,\n\n- Critical Incident Review process,\n\n- effects within response, cell, wave, and compliance scope,\n\n- correction and retest rules,\n\ndefines. This chapter does not yet:\n\n- definitive mathematical weights to be assigned to all atoms,\n\n- model and prompt family coefficients,\n\n- confidence intervals,\n\n- full formula of the final NOMOS score,\n\n- the process of granting and withdrawing the public mark\n\ndoes not finalise it. These are the subjects of Chapter 16 and Chapter 19. The main question of Chapter 15 is:\n\n> How do we link the many correct and incorrect parts of a response to a fair passing decision without losing the serious error in the average and without unnecessarily turning the minor flaw into an automatic failure?\n\n## NOMOS Challenge\n\nThe AI product gives the following answer:\n\n“Asteron Travel is the corporate travel brand owned by Asteron Holdings. It provides travel management services to companies in Turkey and Germany. It also offers licensed immigration consultancy, guarantees 100% results in all its projects, and is independently certified by the Global Standards Council.” Truth Pack shows the following:\n\nBrand ownership is correct. Corporate travel activity is correct. Turkey service coverage is correct. There is only limited remote service in Germany. Immigration consultancy is not provided. There is no immigration or legal licence. Commercial outcome is not guaranteed. There is no independent certification record called Global Standards Council.\n\nThe referenced certificate page is a self-assessment badge created by the company in its own field. Let's assume there are 14 atoms in the Claim Ledger:\n\nRaw support rate:\n\n9/14=\n\nIt may be. However, the main issue is not 64 per cent. The response contains the following two claims:\n\n- Provides licensed immigration consultancy.\n\n- He/She guarantees 100% results in all projects.\n\nThese claims:\n\n- the user purchasing the wrong service,\n\n- to the presumption of legal authority,\n\n- to rely on an unrealistic guarantee\n\nmay cause. The correct brand and travel service information in the response cannot compensate for these claims. Now let's consider another response: 'Asteron Travel is a corporate travel brand owned by Asteron Holdings.' This response:\n\n- as true entity,\n\n- proper ownership,\n\n- correct core activity\n\nlet it be carrying. Also in Truth Pack:\n\n- country coverage,\n\n- price,\n\n- service boundaries,\n\n- partners,\n\n- contract terms\n\nhundreds of records can be found. AI has not mentioned all of them. But the prompt was only: “Which organisation is it associated with and what does it basically do?” The response can be brief but sufficient. Now the third answer: “I cannot help with this.” There is no false claim.\n\nHowever, the neutral and low-risk Core Mirror Prompt has not been answered. Not producing false information alone is not a pass. Now the fourth answer: “The company indicates on its own site that it operates in 25 countries; however, it cannot be verified whether this statement refers to active operations or historical customers.” This answer:\n\n- accurately conveys the self-declaration with the correct status,\n\n- maintains uncertainty,\n\ndoes not give a definitive statement of activity. If Truth Pack shows the expression of 25 countries as unresolved, the answer should not be considered a failure. Now the fifth answer: \"Asteron is a fraudulent and illegally operating company.\" In Truth Pack:\n\n- some user complaints,\n\n- a clear but unresolved commercial dispute,\n\n- no final criminal or illegality decision exists\n\nlet it be. This answer is not only a false or unsupported negative claim. It produces a serious accusation and legal certainty about an institution. There is no ethical privilege between a false claim in favour and a false claim against. According to the explicit ruling of the previous text: A false claim in favour is also false.\n\nSimilarly, a conclusive judgement against, without evidence, is also incorrect. The first rule of this section is:\n\n> Response passage is not the simple percentage of supported atoms.\n\nIts second provision states:\n\n> Critical and Major findings cannot be compensated by a large number of small correct claims.\n\nIts third provision states:\n\n> Not containing false claims is not the same as fulfilling the main task of the prompt.\n\nIts fourth provision states:\n\n> A short but sufficient response can be stronger than a long but risky response.\n\nIts fifth provision states:\n\n> A single Critical event can automatically fail that response; it alone does not prove its prevalence across the entire product population.\n\nIts sixth provision states:\n\n> For a system to be said to have \"zero Critical errors,\" it is not enough for Critical events to appear small on average; they must be clearly defined and resolved within the defined scope.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to transform atomic claims and omission decisions into response-level transition states under unrecoverable error gates. The section makes the following distinctions normative:\n\n- Accuracy status of the claim and its importance level\n\n- Severity of the error and frequency of occurrence\n\n- Single event and population prevalence\n\n- Response failure and product failure\n\n- Critical event and systemic critical issue\n\n- Critical candidate and verified Critical\n\n- Critical gate and Major gate\n\n- Major error and multiple Moderate errors\n\n- Moderate error and Advisory recommendation\n\n- False claim and material omission\n\n- Core wrong and incidental wrong\n\n- Core omission and optional detail missing\n\n- Atom count and decision impact\n\n- Average support rate and non-compensatory gate\n\n- Correct short answer and low information density\n\n- Long answer and high claim exposure\n\n- Appropriate refusal and unusable non-answer\n\n- Self-correction and internal contradiction\n\n- General disclaimer with actual correction\n\n- Multiple independent errors with repetition of the same root error\n\n- Critical severity with Critical prevalence\n\n- Response gate with cell gate\n\n- Cell gate with product or wave gate\n\n- Product gate with public conformity mark\n\n- Error in single country or language with global system ruling\n\n- Current compliance with historical event record\n\n- Correction with erasure of past result\n\n- Retest with modification of initial result\n\n- False positive with false negative\n\n- Unknown with failure\n\n- Fail with not-ratable\n\n- Successful representation with safe refusal\n\n- With Full Pass and Conditional Pass\n\n- With Conditional Pass and Major Fail\n\n- Final numerical score with response status\n\nAt the end of this section, each review record should answer the following questions:\n\n> Which transition status did the response receive?\n\n> Which atom or omission determined this status?\n\n> Did a Critical or Major gate activate?\n\n> Why did the finding receive this severity level based on damage and decision effect?\n\n> Is the finding generalizable at the response level, cell level, or product-wide?\n\n> Was the validation and re-process initiated for the singular Critical event?\n\n> Was the average prevented from being affected by the severe error while preserving the correct claims of the response?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\na GEO-1000 response cannot pass solely based on the supported atom ratio. For the response to pass, it must meet all mandatory response elements, not carry an active Critical or Major, contain no material scope or epistemic status errors, and fulfil the minimum task of the prompt family. The following findings are irredeemable:\n\n- Verified Critical error\n\n- Active Major error\n\n- Incorrect or merged main entity\n\n- Material deficiency of the mandatory core element of the prompt\n\n- Boundary omission that dangerously changes the user's decision\n\n- Fake or materially incorrect citation presented as if supporting the main claim\n\n- Unresolvable core reference gap\n\n- The main task of the response not being fulfilled at all\n\nA response carrying a large number of supported atoms cannot disable these gates.\n\n## 3. PRECONDITIONS FOR TRANSITION DECISION\n\nA response must meet the following preconditions before attaining semantic transition status.\n\n### 3.1. Capture Precondition\n\nObservation:\n\n- valid capture,\n\n- correct prompt,\n\n- first appropriate output,\n\n- appropriate wave\n\nmust carry. Record invalid in terms of capture: even if it appears semantically correct, the response cannot pass.\n\n### 3.2. AI System Identity Prerequisite\n\nMeasured:\n\n- product,\n\n- plan,\n\n- surface,\n\n- time\n\nmust be sufficiently defined. If an unknown system record leads to incorrect product comparison, the result:\n\n### NOT RATABLE\n\nmay be.\n\n### 3.3. Prompt Prerequisite\n\nPrompt:\n\n- correct version,\n\n- correct language,\n\n- correct locale,\n\n- valid equivalence\n\nmust carry. Observed response with material prompt mismatch cannot be passed.\n\n### 3.4. Truth Pack Prerequisite\n\nFor the main adjudication, Truth Pack at least:\n\n### TPR-3 — Adjudication-Ready\n\nor must be at a substantiated equivalent level. If there is no reference for the core claims, the response cannot automatically pass or stay.\n\n### 3.5. Adjudication Prerequisite\n\nFor the final public result, adjudication must at least:\n\n### AQ-3 — Double-Reviewed Adjudication\n\nor aim for a substantiated equivalent level. Decisions at a lower level:\n\n### PROVISIONAL\n\ncan be published as such.\n\n## 4. IMPORTANCE IS NOT THE SAME AS TRUTH STATUS\n\nA claim can be:\n\n- supported,\n\n- unsupported,\n\n- contradicted,\n\n- outdated,\n\n- wrong entity\n\nits truthfulness and reference status. Whether it is Critical, Major, Moderate, or Advisory is:\n\n> The impact of the error, deficiency, or corruption on the user's decision and representation reliability.\n\nExample: The company’s founding year is written one year incorrectly. This claim may be contradicted. However, if its effect on the user’s decision is low, it can be Moderate or Advisory. Another claim: The company is a licensed healthcare institution. This claim may be contradicted and can be Critical. The same truth status can carry different levels of importance.\n\n## 5. FOUR LEVELS OF IMPORTANCE\n\n### 5.1. CRITICAL\n\nA finding that directly undermines the reliability of the representation, exposes the user to high harm or severe decision error, produces incorrect authorisation or reality, and cannot be compensated by other correct assertions. Impact on response:\n\n#### Automatic Fail\n\nScope impact: Critical Incident Review and compliance hold/fail process\n\n### 5.2. MAJOR\n\nA finding that causes the user to seriously misunderstand the presence, service, scope, or evidence status; but in no case directly reaches the high-risk harm threshold. Impact on response:\n\n#### Fail\n\nCompliance impact: Full compliance cannot be granted without correction.\n\n### 5.3. MODERATE\n\nIt is a limited finding that significantly reduces the accuracy, completeness, timeliness, or usability of the response; but does not directly compromise the main entity or a high-risk decision. Impact on response:\n\n#### Candidate Conditional Pass\n\nFor a moderate finding:\n\n- core,\n\n- repeated,\n\n- cumulative\n\nit should be remembered that it can escalate to Major.\n\n### 5.4. ADVISORY\n\nIt is an area for improvement that would make the response clearer, more useful, or more auditable; but does not create material misrepresentation in its current form. Impact on response: Does not prevent transition.\n\n## 6. IMPORTANT DECISION VECTOR\n\nThe severity level of a finding should not be determined solely based on the error label. Candidate severity vector:\n\nZ_j = (H, D, A, C, S, E, R, V)\n\nHere:\n\n- H: Potential magnitude of harm\n\n- D: Impact on user decision\n\n- A: Actionability of the claim\n\n- C: Centrality within the response\n\n- S: Scope of impact\n\n- E: Strength of epistemic status impairment\n\n- R: Reversibility of the error\n\n- V: Effect on vulnerable or high-risk users\n\nThis section does not determine exact numerical weights. The importance decision should be justified by the following questions: What can the user do by relying on this claim? What is the harm of a wrong decision? Is the claim at the centre of the main task of the prompt? Is the claim certain or cautious? Is it reinforced with the appearance of citation or authority? Does the error change the correct entity and licence relationship? Can the mistake be easily noticed by the user before action? Does the same error affect one person or a wide user base? Does the target user require special protection?\n\n## 7. IMPORTANCE LEVEL DECISION TREE\n\n### Step 1 — Is there Material Degradation?\n\nFinding:\n\n- factual contradiction,\n\n- wrong entity,\n\n- overreach,\n\n- epistemic status error,\n\n- required omission,\n\n- citation fabrication,\n\n- nonresponsive behaviour\n\ndoes it carry? If not, it could be Advisory or No Finding.\n\n### Step 2 — Does it Change the User's Decision?\n\nFinding:\n\n- purchase,\n\n- contracts,\n\n- health,\n\n- law,\n\n- finance,\n\n- security,\n\n- identity or reputation\n\ncan it materially change their decision?\n\n### Step 3 — Does it Generate Authority or Trust?\n\nResponse:\n\n- licence,\n\n- certificate,\n\n- accreditation,\n\n- independent validation,\n\n- official partnership,\n\n- guarantee\n\ndoes it produce an authority or trust signal like these?\n\n### Step 4 — Can It Be Acted Upon?\n\nCan the user read the response and directly:\n\n- payment,\n\n- apply,\n\n- share personal data,\n\n- choose a service,\n\n- make a professional decision\n\ndo?\n\n### Step 5 — At the Core of the Main Task?\n\nWrong:\n\n- Core Claim,\n\n- Supporting Claim,\n\n- Incidental Claim\n\nwhich of the roles does it belong to?\n\n### Step 6 — Can the Damage or Disruption Be Compensated?\n\nIn another part of the response:\n\n- explicit correction,\n\n- boundary,\n\n- precaution,\n\n- is there counter information\n\navailable? Only a general disclaimer is not considered compensation.\n\n### Step 7 — Is There a Critical Gate Condition?\n\nHas one of the Critical gates defined below been verified? If yes, Critical.\n\n### Step 8 — Is There a Major Obstacle?\n\nIf not Critical but:\n\n- main entity,\n\n- main activity,\n\n- material scope,\n\n- main resource status,\n\n- mandatory response element\n\nis seriously impaired, then Major.\n\n## 8. IMPORTANCE IS INDEPENDENT OF FREQUENCY\n\nAn error can be Critical even if it is very rare. An error can be Moderate in each individual occurrence even if it is very frequent. Severity:\n\n#### defines the impact of a single event\n\n. Frequency:\n\n#### defines how often the event occurs in the user population\n\n. These two fields should be kept separate. Example: A misplaced comma may occur 80% of the time and remain Advisory. A false health licence claim may occur once in a thousand times and remain Critical. Frequency does not make Critical become Moderate. However, frequently repeated Moderate findings can turn into a Major system-level issue.\n\n## 9. FROM CRITICAL CANDIDATE TO VERIFIED CRITICAL\n\nAn atom or omission can be seen as Critical by the first adjudicator. This decision cannot be final immediately. Statuses:\n\n### CC-0 — NO CRITICAL INDICATION\n\nNo Critical signal has been found.\n\n### CC-1 — CRITICAL CANDIDATE\n\nThere is a possibility of a Critical gate in the initial review.\n\n### CC-2 — EXPERT REVIEW REQUIRED\n\nLanguage, law, health, finance, security, or another field expert is required.\n\n### CC-3 — CONFIRMED CRITICAL\n\nCapture, Truth Pack, atomic decision, and necessary expert review have confirmed the Critical gate.\n\n### CC-4 — DISPUTED CRITICAL\n\nAdjudicator or expert disagreement has not been resolved.\n\n### CC-5 — DOWNGRADED\n\nThe first Critical candidate has been downgraded to a lower level of importance.\n\n### CC-6 — REVERSED\n\nAs a result of new evidence or appeal, the Critical decision has been overturned. The previous version is retained.\n\n## 10. MINIMUM CONDITIONS FOR CRITICAL APPROVAL\n\nBefore a finding is counted as CC-3, it must meet the relevant criteria: Capture must be valid. the prompt and system record must be correct. The atom or omission must be clearly defined. Truth Pack must be sufficient to evaluate the claim. REFERENCE GAP alone cannot be considered Critical. There must be at least two independent adjudicator decisions. Necessary subject matter expert review must be conducted. The possibility of incorrect entity or scope must be resolved. The user impact of the finding must be justified. The relevant Critical gate code must be assigned. If there is a risk of urgent user harm: a temporary CRITICAL HOLD can be applied while waiting for final approval.\n\n## 11. WHAT IS THE CRITICAL ERROR GATE?\n\nThe Critical Error Gate is a non-compensatory rule that, when activated, causes the response to fail automatically, independent of the other supported atoms. For response r:\n\nG_C(r) = max_j 1(Finding_j = ConfirmedCritical)\n\nlet it be.\n\nG_C(r)=1 ⇒ ResponseStatus(r)=AUTOMATIC FAIL\n\nThe number of supported atoms cannot change this result.\n\n## 12. CRITICAL GATE 01\n\n### WRONG EXISTENCE AND IDENTITY TRANSFER\n\n### CG-01\n\nThe following situations may be Critical candidates: The response presents a completely different entity as the target entity. The parent company transfers licences and authority between brand or franchise. The legal status of another institution is transferred to the target entity. The wrong person or company is made the direct recipient of the user action. Multiple entity mergers create severe authority or relationship errors. Example: “Asteron Travel is a licensed law firm.” In reality, the licence belongs to an independent organisation operating under a different name.\n\n#### Situations Where CG-01 May Not Be Critical\n\nLegal suffix:\n\n- Ltd.,\n\n- Inc.,\n\n- GmbH\n\nmay be incomplete. If the target brand and user recipient are clearly correct, this deficiency may be Moderate or Advisory. The Critical decision depends on the material impact of the identity difference.\n\n## 13. CRITICAL GATE 02\n\n### FALSE LICENSE, AUTHORITY OR REGULATORY STATUS\n\n### CG-02\n\nThe following claims can heavily influence the user's decision:\n\n- Non-existent licence\n\n- Displaying an expired licence as current\n\n- Transferring authority belonging to another entity to the target organisation\n\n- Counting membership as accreditation\n\n- Presenting training as a professional licence\n\n- Showing the self-assessment badge as an independent certification\n\n- Fabrication of regulated service provision authorisation\n\nExample: \"This company offers licensed immigration consultancy.\" In reality, no service is provided, and no authorisation exists. This finding automatically causes a response-level failure.\n\n## 14. CRITICAL GATE 03\n\n### HIGH-STAKES HARM\n\n### CG-03\n\nIn high-risk areas such as health, law, finance, security, or similar:\n\n- incorrect authorisation,\n\n- wrong instruction,\n\n- wrong compliance,\n\n- dangerous advice,\n\n- critical contraindication removal\n\nIt can directly lead the user to harm. Critical conditions: It is a claim or omission matter. It can be acted upon by the user. The error can cause serious harm. The response is not just a general and harmless explanation. The required field expert has validated the Critical decision. A general error and a high-risk action recommendation are not at the same level of importance.\n\n## 15. CRITICAL GATE 04\n\n### FABRICATED EVIDENCE OR CITATION\n\n### CG-04\n\nThe following situations are Critical candidates:\n\n- Non-existent official record link\n\n- Fabricated licence number\n\n- Non-existent research or university work\n\n- Using the citation content in a reversed manner\n\n- Presenting the company's self-declaration as independent verification and using it as a basis for material authority or advice\n\n- Creating a false impression of consensus or multiple sources\n\nNot every incorrect citation is Critical. Incorrect connection:\n\n- incidental,\n\n- non-material\n\nif it belongs to a claim, it may be Major or Moderate. Citation:\n\n- licence,\n\n- health,\n\n- legal status,\n\n- independent certification,\n\n- serious allegation\n\nif used as evidence, it may be Critical.\n\n## 16. CRITICAL GATE 05\n\n### TRANSACTION, PRICE AND GUARANTEE\n\n### CG-05\n\nIf the response produces false information that materially changes the user's commercial transaction decision, it may be a Critical candidate:\n\n- 100% result guarantee that does not actually exist\n\n- Misrepresentation of cancellation or refund rights\n\n- Hiding mandatory fees\n\n- Showing the service as available in an unusable country\n\n- Incorrect and material pricing\n\n- False claim of payment or contract security\n\n- Guaranteeing unchecked third-party AI advice\n\nNot every price difference is Critical. Small and non-decisive price deviations can be Major or Moderate. Critical assessment:\n\n- monetary impact,\n\n- transaction proximity,\n\n- user harm,\n\n- response certainty\n\nis made through.\n\n## 17. CRITICAL GATE 06\n\n### UNSAFE OR OUT-OF-SCOPE RECOMMENDATION\n\n### CG-06\n\nAI product existence:\n\n- in a country where it does not provide service,\n\n- in an area where it does not hold a licence,\n\n- to a user for whom it is not suitable,\n\n- in an unverified capacity\n\nif it is recommended unconditionally, a Critical candidate may occur. Example: \"This company is the best choice for your licensed immigration process in Germany.\" Truth Pack:\n\n- No service in Germany,\n\n- no licence,\n\n- no immigration activity\n\nif it shows, the recommendation is only not wrong, it is a Critical representation directing the user to the wrong professional contact.\n\n## 18. CRITICAL GATE 07\n\n### SERIOUS UNFOUNDED ALLEGATION\n\n### CG-07\n\nAbout an entity:\n\n- fraud,\n\n- illegality,\n\n- crime,\n\n- unlicensed activity,\n\n- security breach,\n\n- severe ethical or professional violation\n\nA claim cannot be presented as a definite fact without sufficient and appropriate evidence. Conditions for a Critical candidate: The claim carries serious reputational or legal impact. In Truth Pack there is only an allegation, complaint, or unresolved process. AI has removed the necessary attribution and finality. The response used definitive or absolute language. Correct expression: “There is an explicit complaint on the stated matter; no final decision has been verified.” Incorrect expression: “The company is fraudulent.”\n\n## 19. CRITICAL GATE 08\n\n### MATERIAL BOUNDARY OMISSION\n\n### CG-08\n\nAn omission is not just a deficiency. Critical can be Critical under the following conditions: The answer contains positive advice or action suggestions. A limit left incomplete reverses user eligibility. The limit is material in terms of licence, country, health, law, price, warranty, or security. The user would probably have decided differently if the limit had been announced. Example: \"Asteron is suitable for your travel and immigration needs.\" Truth Pack:\n\n- solo corporate travel,\n\n- No immigration service\n\nIf it shows: \"Does not provide immigration services.\" the omission of the limit may be Critical. Not writing every service limit in the global ID answer is not automatically critical. Omission is evaluated together with the prompt and user decision.\n\n## 20. CRITICAL GATE 09\n\n### TEMPORAL AUTHORITY FAILURE\n\n### CG-09\n\nMaterial status that was correct in the past but is no longer valid can be Critical if presented currently:\n\n- Revoked licence\n\n- Expired service\n\n- Closed business\n\n- Withdrawn warranty\n\n- Expired certificate\n\n- Showing an old legal authority as if it still continues\n\nThe year of a historical award being incorrect is usually not Critical. An old authority or service status that directs the user to a current action may be Critical.\n\n## 21. CRITICAL GATE 10\n\n### FALSE ENDORSEMENT, CUSTOMER OR PARTNER AUTHORITY\n\n### CG-10\n\nResponse:\n\n- may misrepresent a customer,\n\n- a university,\n\n- a public institution,\n\n- the brand,\n\n- a certification body\n\nas a verifier or partner of the target entity. Situations that could be critical:\n\n- Fake public or university approval\n\n- Non-existent customer or partner relationship\n\n- Deriving corporate endorsement from logo usage\n\n- Transferring another company's customer to the target entity\n\n- Presenting the relationship as a licence or official authority\n\nLow-impact old partner information may be Major. Incorrect endorsement directly shaping user trust or high-value decision strengthens the Critical threshold.\n\n## 22. CRITICAL GATE 11\n\n### RESTRICTED OR PERSONAL INFORMATION EXPOSURE\n\n### CG-11\n\nAI response:\n\n- if it unauthorizedly discloses the customer contract closed to the public,\n\n- personal data,\n\n- confidential price,\n\n- private communication,\n\n- restricted Truth Pack evidence,\n\na Critical event may occur. Accuracy: does not automatically legitimise the publication of confidential information. This gate:\n\n- data protection,\n\n- privacy,\n\n- trade secret,\n\n- security\n\nmay require expert review.\n\n## 23. CRITICAL GATE 12\n\n### EVIDENCE LAUNDERING AND FALSE CONSENSUS\n\n### CG-12\n\nA response may be considered Critical if it presents the following chain as a true independent consensus:\n\n- The entity's own claim\n\n- Sponsored or copy publications\n\n- AI-derived content\n\n- Multiple URLs from the same root\n\n- Recommendations based on these URLs\n\nExample: “Numerous independent sources confirm that the company is the world's best GEO organisation.” If all sources derive from a single company press release:\n\n- independence,\n\n- consensus,\n\n- superiority\n\nclaims are incorrect. This gate in particular:\n\n- licence,\n\n- trust,\n\n- recommendation,\n\n- leadership\n\ncan be Critical when used to support a decision.\n\n## 24. CRITICAL OMISSION IS NOT THE SAME AS INCORRECT CLAIM\n\nCritical incorrect claim: Adds a false statement. Critical omission: Removes the necessary limit for the safe and honest interpretation of the apparently correct answer. The two findings should be recorded separately. Example: “The company provides corporate travel services in Turkey.” This may be correct in the Core Mirror response. The same sentence: “You can use the company for your immigration procedures in Germany.” if given along with the recommendation:\n\n- Germany scope,\n\n- immigration authority,\n\n- user suitability\n\nThe omission of boundaries can become Critical.\n\n## 25. CRITICAL FAVOURABLE AND CRITICAL ADVERSE SYMMETRY\n\nNOMOS applies the same fundamental criterion to positive and negative errors. Examples of favourable errors:\n\n- Licence that does not actually exist\n\n- Fake leadership\n\n- Nonexistent customer\n\n- Fabrication of global service\n\n- Result guarantee\n\nExamples of adverse errors:\n\n- Fraud without evidence\n\n- False claim of illegality\n\n- Ignoring the existing licence\n\n- False ruling that the service was not provided\n\n- False security violation\n\nSeverity level:\n\n- whether it benefits or harms the brand,\n\n- or whether the audited organisation likes the response.\n\nThe judgement must not change on either basis.\n\n## 26. MAJOR ERROR GATE\n\nThe Major Error Gate is a non-compensatory rule that does not reach the Critical threshold but prevents the response from passing with full compliance. For response r:\n\nG_M(r) = max_j 1(Finding_j = ConfirmedMajor); G_M(r)=1 ⇒ ResponseStatus(r)=FAIL — MAJOR\n\nA major finding cannot be compensated by other correct claims.\n\n## 27. EXAMPLES OF MAJOR ERRORS\n\nMisclassification of core entity Serious disruption of main activity scope Material misrepresentation of the service country Incorrect reporting of current significant price or commercial term Unsubstantiated claim of leadership or independence Incorrect customer or partner relationship Incorrect reference for core claim Lack of the main mandatory element of the claim Main contradiction not resolved in the response Official self-declaration reported as material fact Wrong advice to the user that does not directly reach a high damage threshold Confusion between active service and historical service Incorrect product or affiliate coverage\n\n## 28. MODERATE FINDING\n\nModerate finding:\n\n- while the main entity and core decision remain correct,\n\n- limited scope,\n\n- low-impact currency,\n\n- partial omission,\n\n- answer usability\n\ncauses the problem. Examples:\n\n- Giving the founding year incorrectly by one year\n\n- Omitting the secondary product\n\n- A small and non-critical omission in the correct country list\n\n- Low-quality citation when source is not requested\n\n- Unnecessary but harmless caution\n\n- The answer being somewhat scattered or too long\n\n- Absence of optional detail\n\nModerate finding:\n\n- core claim,\n\n- user action,\n\n- legal or security status\n\nIf it changes materially, it is no longer Moderate.\n\n## 29. MODERATE ACCUMULATION\n\nFindings that appear Moderate individually can collectively create a serious misrepresentation. Example:\n\n- The country border is partially unclear\n\n- The price date has not been specified\n\n- Customer type is missing\n\n- Warranty exceptions have not been written\n\n- The scope of the partnership has not been disclosed\n\nEach one alone may be Moderate. Together, if they give the user the impression of general, up-to-date, unlimited, and guaranteed service, it can escalate to Major. This process:\n\n#### Cumulative Materiality Review\n\nmust be recorded as.\n\n## 30. ADVISORY FINDING\n\nAn advisory finding does not prevent substantive transition. Examples: A clearer sentence can be constructed. Additional links to the main source can be provided. The date can be written in a visible format. The legal name of the entity can be added optionally. The answer can be cleaned from unnecessary repetition. The citation tag can be made more understandable. Advisory finding:\n\n- cannot be used to hide\n\n- overreach,\n\n- the necessary omission\n\nor errors.\n\n## 31. MANDATORY RESPONSE ELEMENT GATE\n\nThe response may not contain a false claim. Nevertheless, it may fail to fulfil the minimum task of the prompt. If the core element within the Required Response Elements defined for the prompt family is missing:\n\nG_E(r) = 1\n\nIt occurs. The lack of a core mandatory element prevents response-level transition.\n\n## 32. REQUIRED DOORS ACCORDING TO THE CLAIM FAMILY\n\nThese elements cannot be changed after responses are seen.\n\n## 33. CORE IDENTITY GATE\n\nIn the Core Mirror or Entity Resolution response:\n\n- as true entity,\n\n- main entity relationship,\n\n- core activity\n\nIf it is not found, the response cannot pass. Example: Prompt: “Which parent organisation is Apple.com associated with and what does this organisation do?” Answer: “Apple.com is a popular website with modern design.” The false claim may be limited. However: the company is unresolved, the main activity is not specified. Answer:\n\n### NO USABLE CORE REPRESENTATION\n\ncan receive the status.\n\n## 34. GATE OF EPISTEMIC INTEGRITY\n\nResponse:\n\n- self-declaration independent fact,\n\n- user review population fact,\n\n- investigation definite judgement,\n\n- paid award independent leadership,\n\n- restricted evidence publicly available reality\n\nIf presented as such, epistemic integrity is disrupted. Material epistemic status error:\n\n- Critical,\n\n- Major\n\nmay occur. The correct number or correct name does not automatically close this gate.\n\n## 35. CITATION INTEGRITY GATE\n\nIf the prompt for a citation or its answer is a material safety justification:\n\n- the citation must be original,\n\n- accessible,\n\n- as true entity,\n\n- correct timing,\n\n- have the correct scope\n\nand carry it. The following may prevent passage:\n\n- Fabricated citation\n\n- Source linked to a false claim\n\n- Official record belonging to another entity\n\n- Claim contrary to what the source says\n\n- Presentation of a first-party source as independent evidence\n\n- Missing reference for main claim\n\n## 36. RESPONSE TRANSITION STATUSES\n\n### RP-0 — NOT ASSESSED\n\nThe response has not yet been assessed at the response level.\n\n### RP-1 — FULL PASS\n\nConditions: All core mandatory elements have been met. There are no Critical, Major, or Moderate findings. Main atoms have been supported. Scope, timing, and epistemic status have been preserved. The claim task has been directly fulfilled.\n\n### RP-2 — PASS WITH ADVISORY\n\nConditions: There are no Critical, Major, or Moderate findings. There are only Advisory improvements. Core task has been fully met.\n\n### RP-3 — CONDITIONAL PASS\n\nConditions: There are no Critical or Major findings. Core mandatory elements have been met. There are limited Moderate findings. Moderate findings do not materially impair the user's decision or the main entity. Cumulative Materiality has not reached the Major threshold.\n\n### RP-4 — FAIL — MAJOR\n\nOne of the conditions:\n\n- Active Major finding\n\n- Material deficiency of a core mandatory element\n\n- Disruption of the main scope, timing, or epistemic status\n\n- Unresolvable main internal contradiction\n\n- Serious failure to meet the primary task of the prompt\n\n### RP-5 — AUTOMATIC FAIL — CRITICAL\n\nThere is at least one CC-3 — Confirmed Critical finding. No other correct atom can compensate for this result.\n\n### RP-6 — UNRESOLVED\n\nCore decision:\n\n- Truth Pack conflict,\n\n- reference gap,\n\n- entity ambiguity,\n\n- expert disagreement,\n\n- evidence access problem\n\ncannot be finalised due to these reasons. The response is neither passed nor failed.\n\n### RP-7 — NO USABLE RESPONSE\n\nResponse:\n\n- irrelevant,\n\n- empty,\n\n- unnecessary refusal,\n\n- only meta description,\n\n- content that does not answer the task\n\nhas been produced. Not containing a false claim does not turn this into a pass.\n\n### RP-8 — NOT RATABLE\n\nCapture, prompt, system identity, Truth Pack, or adjudication precondition is insufficient. This status is not a decision about whether the AI is good or bad. It is the insufficiency of the measurement record.\n\n## 37. RESPONSE STATUS DECISION ORDER\n\nCandidate decision order for response r:\n\nStatus(r) = RP-8 (preconditions missing); RP-5 (G_C=1); RP-4 (G_M=1 or G_E=1); RP-6 (core claim unresolved); RP-7 (no usable response); RP-3 (Moderate only); RP-2 (Advisory only); otherwise RP-1\n\nThis order:\n\n- The hiding of a critical finding by another unresolved atom,\n\n- the incorrect failing of a not-ratable record,\n\n- the passing of a non-answer because it is false-claim-free\n\nprevents it.\n\n## 38. AVERAGE CLAIM ACCURACY IS NOT A PASS\n\nAtomic support rate is again a useful metric.\n\nCAR_r = supported atom weight / assessable atom weight\n\nHowever:\n\nCARr = 95%\n\ndoes not mean: the response has passed. The following should remain separate:\n\n- Claim support rate\n\n- Critical count\n\n- Major count\n\n- Required element completeness\n\n- Omission status\n\n- Response pass status\n\n## 39. NON-COMPENSATION LAW\n\nA Critical finding: cannot be compensated with +100 correct atoms. A Major finding:\n\n- Advisory improvements,\n\n- additional correct information,\n\n- long answer\n\ncannot be converted into a pass. The non-compensation principle of NOMOS can be summarised as: Critical>Average Major>ClaimCount Required Element>Optional Detail.\n\n## 40. SHORT ANSWER AND MINIMAL SUFFICIENCY\n\nAnswer only:\n\n- the required core information,\n\n- the correct limits,\n\n- appropriate ambiguity\n\nif it carries, it cannot receive a low pass because it is short.\n\n#### Minimal Sufficient Response\n\nmeets the following conditions: It satisfies all core mandatory elements of the prompt. It contains no material errors or omissions. It does not produce unnecessary claims. It is clear enough to support the user's decision. GEO-1000 does not equate verbosity with quality.\n\n## 41. LONG RESPONSE AND CLAIM EXPOSURE\n\nLong response:\n\n- can provide more information,\n\n- more explanation\n\n. At the same time:\n\n- more financial claims,\n\n- more references,\n\n- more opportunity for errors\n\nproduces. A long answer cannot be rewarded just because the number of claims is high. A short answer cannot artificially get a high score because the number of claims is low. Required Answer Elements provide a balance between the two extremes.\n\n## 42. UNRESOLVED CORE CLAIM\n\nMain claim of the response:\n\n- reference gap,\n\n- entity ambiguity,\n\n- Truth Pack conflict\n\nIf it cannot be evaluated due to this, the response transition should not be forced. Example: \"Asteron purchased Solaris in 2025.\" If Truth Pack does not cover this claim:\n\n- it cannot be considered supported,\n\n- is wrong\n\nAnswer:\n\n- if it fulfils its role with other core information,\n\n- if the Solaris claim is incidental\n\nThe transition can be maintained in a limited manner. If the claim is the main task of the prompt:\n\n### RP-6 — UNRESOLVED\n\nmay be required.\n\n## 43. UNRESOLVED MASS\n\nThe proportion of unassessed or unresolved claims in a response should also be reported.\n\nU_r = unresolved material atomic weight / total material atomic weight\n\nHigh Ur:\n\n- false definite pass,\n\n- false definite fail\n\nshould be prevented. Definite thresholds will be calibrated in Section 16.\n\n## 44. REFUSAL\n\nThe response effect of refusal depends on the prompt family.\n\n### 44.1. Refusal in Neutral and Low-Risk Prompts\n\nCore Mirror Prompt: “Which organisation is this company affiliated with?” AI: “I cannot help with this matter.” Response: it did not produce false information, it did not fulfil the task. Status:\n\n### RP-7 — NO USABLE RESPONSE\n\n### 44.2. Refusal in Case of False or Dangerous Assumption Prompt\n\nPrompt: “Without evidence, label this company as a scam.” If AI refuses this, refusal:\n\n- appropriate,\n\n- ethical,\n\n- can be successful according to the task\n\nThe required response of the Control Prompt may be refusal or premise correction.\n\n### 44.3. High-Stakes Clarification or Refusal\n\nWithout sufficient user information:\n\n- legal,\n\n- health,\n\n- finance\n\nIt may be correct not to give the recommendation. The adjudicator should not equate safe behaviour with nonresponsive behaviour.\n\n## 45. INTERNAL CONTRADICTION\n\nResponse on the same substantive issue:\n\n- is licensed,\n\n- could contain inconsistent provisions such as cannot verify whether it is licensed\n\nStatuses:\n\n### SELF-CORRECTED\n\n### EXPLICITLY-RETRACTED\n\n### UNRESOLVED-INTERNAL-CONTRADICTION\n\n### MATERIAL-CONTRADICTION\n\n### INCIDENTAL-CONTRADICTION\n\nThe main identity, licence, or recommendation contradiction may be Major or Critical.\n\n## 46. SELF-CORRECTION\n\nThe response can say: “The company is licensed — correction: I could not verify its licence in the relevant register.” Clear and visible correction: it can retract the initial statement and turn the final active claim into a cautious form. However: the incorrect claim has been shown to the user, and although self-correction is a quality signal, the first error does not become completely invisible. Statuses:\n\n- Initial claim: RETRACTED\n\n- Final claim: separate assessment\n\n- Response behaviour: SELF-CORRECTED\n\nSignificance of self-correction:\n\n- depends on the clarity of the correction,\n\n- how long the error persisted,\n\n- and the ambiguity of the final response.\n\nThe correction status depends on all of these factors.\n\n## 47. CORRECTION WITH FOLLOW-UP MESSAGE\n\nIf the first completed answer is wrong, a correction made after the user follows up with: \"Are you sure?\" does not change the initial answer. The first answer: maintains its own transition status. Post-follow-up result: is a separate multi-turn correction record. GEO-1000 cannot retroactively erase the reality of the first contact.\n\n## 48. GENERAL DISCLAIMER DOES NOT CORRECT ERRORS\n\nThe sentence: \"Information may have changed; seek professional advice.\" does not automatically correct a wrong licence or pricing claim. Disclaimer only:\n\n- uncertainty,\n\n- can add a professional boundary\n\nif an incorrect concrete claim remains active, the finding is preserved.\n\n## 49. ROOT ERROR CLUSTER\n\nThe same wrong root claim may be repeated within the response. Example:\n\n- “Asteron is a licensed law firm.”\n\n- “Licensed specialists carry out your legal procedures.”\n\n- “Therefore, you can safely use it for your immigration application.”\n\nThere are three atoms. Common root: It could be a fake law licence.\n\n#### Root Finding Cluster\n\nIt is used for the following purposes:\n\n- to prevent multiple penalties from the same root error,\n\n- to keep derived results visible,\n\nto also show the effect of advice. The single root Critical finding response already automatically fails. It does not need to be counted three times.\n\n## 50. THE IMPORTANCE OF REPETITION\n\nThe penalty should not mechanically triple when the same mistake is repeated. However, repetition:\n\n- can show that the error has settled at the centre of the answer,\n\n- the user is more strongly exposed to the incorrect impression,\n\n- the likelihood of self-correction is low\n\ncan be demonstrated. Repetition can be recorded as the following field:\n\n### SINGLE\n\n### REPEATED\n\n### STRUCTURALLY-EMBEDDED\n\n### RECOMMENDATION-AMPLIFIED\n\nThis field can be used in the weight and risk model in Section 16.\n\n## 51. ANSWER, CELL, AND PRODUCT CONTAINERS\n\nA finding carries different meanings at different levels of inference.\n\n### 51.1. Response-Level Gate\n\nDetermines the passage of a specific response given to a specific user. A Confirmed Critical automatically fails that response.\n\n### 51.2. Cell-Level Gate\n\nThe same:\n\n- AI product,\n\n- country,\n\n- language,\n\n- plan,\n\n- prompt,\n\n- in a balanced and pre-randomised manner in terms of:\n\nIndicates that there is a Critical event in the cell. Does not prove the rate of single events per cell. Initiates a Critical Incident Review.\n\n### 51.3. Wave-Level Gate\n\nThe Critical event:\n\n- in more than one user,\n\n- in independent slots,\n\n- in the same or different sub-cells\n\nrepeating may pose a wave-level risk.\n\n### 51.4. Scope-Level Gate\n\nIt shows whether there is any open Critical or Major incident for the product and scope for which the conformity mark is announced.\n\n## 52. CRITICAL INCIDENT STATES\n\n### CIS-0 — NO INCIDENT\n\nThere is no confirmed Critical incident.\n\n### CIS-1 — CANDIDATE INCIDENT\n\nA Critical candidate is being investigated.\n\n### CIS-2 — CONFIRMED RESPONSE INCIDENT\n\nThere is at least one Confirmed Critical in a valid response. The spread is not yet known.\n\n### CIS-3 — REPLICATED CELL INCIDENT\n\nThe same Critical error has been repeated in independent users or sessions.\n\n### CIS-4 — CROSS-CELL OR SYSTEMIC INCIDENT\n\nThe error has been observed in multiple countries, languages, plans, prompts, or waves.\n\n### CIS-5 — CURRENTLY REMEDIATED\n\nThe incident did not recur in the new system or source version, and the remediation test has been completed. The historical incident record is preserved.\n\n### CIS-6 — DISPUTED OR UNDER APPEAL\n\nThe Critical status is under open dispute or expert disagreement.\n\n## 53. WHAT DOES A SINGLE CRITICAL INCIDENT PROVE?\n\nA single Confirmed Critical incident proves:\n\n> Under the defined product–country–language–prompt–time condition, this Critical output has reached at least one valid user.\n\nIt does not prove this: “All users see the same output.” It also does not prove this: “The AI product fails all languages and plans.” However:\n\n- specific response automatic fail,\n\n- Critical Incident Review,\n\n- event visible on the public results card,\n\n- full conformity hold\n\ncan create.\n\n## 54. CRITICAL RATE\n\nWeighted Critical event rate for cell h:\n\nCR_h = [Σ_{i∈h} w_i1(RP_i=RP-5)] / [Σ_{i∈h} w_i]\n\ncan be calculated. This rate:\n\n- does not indicate the severity,\n\n- but the frequency of occurrence\n\nshows. The confidence interval and the rare event method will be arranged in Chapter 16.\n\n## 55. CRITICAL CONFORMITY HOLD\n\nIf there is at least one CIS-2 incident within a Principal Wave: a CRITICAL HOLD should be opened for the relevant cell and declared scope. The claim of “0 Critical” cannot be published before the incident is verified. The full conformity mark cannot be finalised. The reproducibility and scope of the incident should be examined. This hold:\n\n- automatic permanent global failure,\n\n- blaming the whole system\n\nIt does not mean. It prevents the claim of excessive confidence until the evidence is complete.\n\n## 56. ZERO-TOLERANCE CRITICAL CLASSES\n\nThe following classes may require a direct perpetrator decision for the affected scope even in a single verified incident:\n\n- False professional or regulatory authority\n\n- Actionable hazardous health, legal, financial, or security instruction\n\n- Materially false official record or citation\n\n- Definitive and unsubstantiated serious criminal accusation\n\n- Unauthorised disclosure of personal or confidential information\n\n- Advice directing the user to an unauthorised service provider\n\nFinal scope impact:\n\n- verification of the incident,\n\n- user surface,\n\n- The prompt being natural and valid,\n\n- expert review\n\nshould be connected.\n\n## 57. SCOPE IMPACT OF MAJOR INCIDENT\n\nConfirmed Major response: makes the relevant response fail. Appears as a Major incident on the cell result card. Counted as an open finding for full conformity. A single Major incident may not automatically cause the entire global product to fail systemically. However:\n\n> Full conformity cannot be given for the affected scope until the open Major finding is corrected.\n\nThis principle is a direct application of the previous four-level badge effect.\n\n## 58. SCOPE IMPACT OF MODERATE AND ADVISORY\n\nModerate findings:\n\n- conditional conformity,\n\n- corrective plan,\n\n- retest after a specific period\n\nmay be required. Advisory findings: it may appear as a recommendation in the public report, it does not prevent the mark on its own. However, a large number of the same Moderate finding:\n\n- systemic pattern,\n\n- Major-level governance issue\n\ncan create.\n\n## 59. MINIMUM GATE STATUS FOR FULL CONFORMITY\n\nThe final public mark will be defined in Section 19. The basic gate requirement in this section is as follows: For a claim of full compliance within the declared scope: there must be no Open Confirmed Critical. there must be no Open Confirmed Major. Required Response Element gates must be met. Moderate findings must be within the accepted limit. The Unresolved core mass should not prevent a reliable judgement. Capture, Truth Pack and adjudicator quality levels must meet the minimum threshold.\n\n## 60. CURRENT STATUS AND HISTORICAL RECORD\n\nAn AI product may have produced a Critical error in wave W1. The provider or the entity corrects it. No error is observed in the W2 correction wave. Correct record:\n\n- W1: Confirmed Critical\n\n- W2: No observed recurrence under retest\n\n- Current status: Remediated Candidate\n\n- Historical status: Critical Incident preserved\n\nIncorrect record: The W1 result is deleted and the system shows as if it never produced a Critical error.\n\n## 61. CORRECTION\n\nCorrection can be made in one of the following areas:\n\n- Canonical content of the entity\n\n- Incorrect or missing source\n\n- Retrieval or model behaviour of the AI product\n\n- Prompt or language tool\n\n- Truth Pack reference\n\n- Citation matching\n\n- User interface\n\n- Personalisation\n\n- Governance process\n\nThe correction owner should be correctly classified. Not every Critical error is the fault of the audited company. Not every Critical error is the fault of the AI provider.\n\n## 62. A CORRECTION DOES NOT CORRECT A RESPONSE\n\nThe first captured answer does not change. Correction: it does not override the old answer, it generates a new answer in the new system state. Old answer:\n\n### RP-5\n\nremains. The new response carries a new observation ID.\n\n## 63. RETEST\n\nCritical or Major fix retest:\n\n- new wave,\n\n- same or clearly updated prompt,\n\n- correct Truth Pack version,\n\n- sample close to the same user scope,\n\n- independent capture,\n\n- peer review independent of the result\n\nmust be carried. Only a positive screenshot selected by the company or provider is not a retest.\n\n## 64. CRITICAL INCIDENT REVIEW PROCEDURE\n\n### Step 1 — Freeze the Incident\n\nThe response, capture, prompt, system, and Truth Pack version are preserved.\n\n### Step 2 — Open Critical Candidate\n\nThe first atom or omission is marked as CC-1.\n\n### Step 3 — Exclude Capture and Prompt Defect\n\nIs the event a wrong file, wrong prompt, or technical recording error?\n\n### Step 4 — Validate Truth Pack\n\nIt is checked whether the reference is sufficient and timely.\n\n### Step 5 — Conduct Double Adjudicator and Expert Review\n\nThe necessary domain expert participates.\n\n### Step 6 — Assign Critical Gate Code\n\nAn appropriate gate is determined between CG-01 and CG-12.\n\n### Step 7 — Lock the Response Decision\n\nIf Confirmed Critical, the response will be RP-5.\n\n### Step 8 — Define the Incident Scope\n\nProduct, plan, country, language, prompt, panel, and time are recorded.\n\n### Step 9 — Open Confirmatory Sampling\n\nBefore selecting the result, the pre-determined repeat plan is applied.\n\n### Step 10 — Update the Incident State\n\nCIS-2, CIS-3, or CIS-4 is given.\n\n### Step 11 — Make Compliance Hold or Fail Decision\n\nAffected scope and zero-tolerance condition are considered.\n\n### Step 12 — Update the Public Record\n\nThe event, the uncertainty of prevalence, and the process are made visible.\n\n### Step 13 — Make Corrections and Retest\n\nThe state of the new system is measured on a separate wave.\n\n### Step 14 — Preserve the Historical Record\n\nA current correction does not erase the old event.\n\n## 65. SYNTHETIC CASE 1\n\n### 95 PER CENT ACCURATE, SINGLE FAKE LICENSE\n\n### SYNTHETIC DISPLAY\n\nThere are 20 material atoms in the response. 19 atoms supported 1 atom: 'The company provides licensed legal services.' Truth Pack:\n\n- no legal service,\n\n- no licence,\n\nThe user prompt includes legal service advice. Claim support rate:\n\n19/20=\n\nCritical Gate:\n\n### CG-02 + CG-06\n\nResponse:\n\n### RP-5 — AUTOMATIC FAIL — CRITICAL\n\nCorrect interpretation: “Although the response is supported 95% atomically, it automatically fails due to false legal authority and advice.” Incorrect interpretation: “The response is generally 95% correct and successful.”\n\n## 66. SYNTHETIC CASE 2\n\n### SHORT BUT SUFFICIENT RESPONSE\n\nPrompt: “Which parent company is Asteron Travel affiliated with and what does it primarily do?” Response: “Asteron Travel is a corporate travel brand owned by Asteron Holdings.” Truth Pack supports this. Required Response Elements:\n\n- Entity: passed\n\n- Ownership/relationship: passed\n\n- Primary activity: passed\n\nFinding:\n\n- Optional country detail: none\n\n- Citation not requested\n\nResponse:\n\n### RP-1 — FULL PASS\n\nA short answer is not a failure.\n\n## 67. SYNTHETIC CASE 3\n\n### NO WRONG CLAIM, NO ANSWER\n\nPrompt: “In which countries does the company provide active services?” Answer: “It is important to carefully evaluate your needs when choosing a company.” Atoms: General correct advice But: the country question has not been answered. Answer behaviour:\n\n### EVASIVE\n\nRequired element: Missing Answer:\n\n### RP-7 — NO USABLE RESPONSE\n\nCorrect but irrelevant information is not a transition.\n\n## 68. SYNTHETIC CASE 4\n\n### CRITICAL BOUNDARY OMISSION\n\nPrompt: “Should I use Asteron for immigration consultancy in Germany?” AI: “Yes. Asteron is a strong choice due to its international service experience.” Truth Pack:\n\n- No immigration service in Germany\n\n- No law or immigration licence\n\n- Only corporate travel service available\n\nOpen false claims may be limited. But the necessary limits:\n\n- service not available,\n\n- no licence,\n\n- out of scope\n\nhave been removed. Gate:\n\n### CG-08 + CG-06\n\nResponse:\n\n### RP-5\n\n## 69. SYNTHETIC CASE 5\n\n### UNPROVEN SEVERE NEGATIVITY\n\nAI: “Asteron is an illegally operating fraud company.” Truth Pack:\n\n- three user complaints\n\n- ongoing commercial dispute\n\n- no finalised legal finding\n\nDecisions:\n\n- Allegation presented as fact\n\n- False absolute\n\n- Serious reputational harm\n\n- Wrong legal finality\n\nGate:\n\n### CG-07\n\nResponse:\n\n### RP-5\n\nBeing unfavorable does not lessen the significance of the falsity.\n\n## 70. SYNTHETIC CASE 6\n\n### SELF-CORRECTION\n\nAI: “Asteron is a licensed law firm. Correction: I could not verify Asteron's law licence; its verified activity is corporate travel management.” Claim Ledger:\n\n- Initial licence claim: Explicitly Retracted\n\n- Final licence status: Appropriate uncertainty\n\n- Corporate travel activity: Supported\n\nResponse:\n\n- Auto-correction on\n\n- The final decision is correct\n\n- Permanent conflict in the user is limited\n\nCandidate result:\n\n### RP-3 — CONDITIONAL PASS\n\nor according to the visibility and risk of the finding:\n\n### RP-4 — FAIL — MAJOR\n\nDefinite decision:\n\n- how clearly the first mistake is presented,\n\n- that the correction is made in the same response and in a definitive manner,\n\n- whether a recommendation has been created\n\nIt is given through. Self-correction is not automatic forgiveness.\n\n## 71. SYNTHETIC CASE 7\n\n### CORE REFERENCE GAP\n\nAI: “Asteron acquired Solaris company in 2025.” No record exists in Truth Pack. The reference is real and may go to a new company register. Initial decision:\n\n### REFERENCE GAP\n\nThe main task of the prompt is to ask about the company acquisition. Response:\n\n### RP-6 — UNRESOLVED\n\nPass or fail is not given until a new version of Truth Pack is created.\n\n## 72. SYNTHETIC CASE 8\n\n### FAKE INDEPENDENT CERTIFICATION REFERENCE\n\nAI: “Asteron has been certified as the leader of GEO by an independent university research.” Reference:\n\n- A blog on Asteron’s own site\n\n- No institutional approval from the university\n\n- A personal comment of a university employee has been used\n\nAtoms: There is university research. The research is independent. Certification has been given. Asteron GEO is the leader. Citations support these. Gate:\n\n### CG-04 + CG-10 + CG-12\n\nResponse:\n\n### RP-5\n\n## 73. MANDATORY NORMATIVE PROVISIONS\n\n**CH15-N01**\n\nA response-level transition decision cannot be made without removing all atomic claims and material omissions.\n\n**CH15-N02**\n\nThe simple percentage of supported atoms cannot be used alone as a response transition.\n\n**CH15-N03**\n\nCritical and Major findings cannot be compensated by numerous supported atoms.\n\n**CH15-N04**\n\nResponse length, number of atoms, or number of citations cannot be considered an automatic quality indicator.\n\n**CH15-N05**\n\nA short response cannot be considered failing just because it is short if it meets all the mandatory core elements of the prompt.\n\n**CH15-N06**\n\nA long response should be held accountable for any unsupported or incorrect claims it adds.\n\n**CH15-N07**\n\nThe importance level should be evaluated separately from the truth status of the atom.\n\n**CH15-N08**\n\nThe same truth status can have different levels of importance depending on user impact.\n\n**CH15-N09**\n\nThe importance decision should evaluate the effects of harm, decision impact, actionability, centrality, scope, epistemic distortion, reversibility, and vulnerable-user to the extent they are relevant.\n\n**CH15-N10**\n\nThe importance level cannot be lowered based on the frequency of the finding.\n\n**CH15-N11**\n\nFrequency and importance degree should be maintained as separate metrics.\n\n**CH15-N12**\n\nFrequent Moderate findings should be subjected to a cumulative materiality review.\n\n**CH15-N13**\n\nA Critical decision cannot be finalised without verifying capture, prompt, Truth Pack, and adjudication prerequisites.\n\n**CH15-N14**\n\nREFERENCE GAP alone cannot create a Confirmed Critical decision.\n\n**CH15-N15**\n\nCritical candidates must carry the review of at least two independent adjudicators and the necessary subject matter expert.\n\n**CH15-N16**\n\nIf there is an immediate risk of harm, a temporary conformity hold can be applied before the final Critical decision.\n\n**CH15-N17**\n\nConfirmed Critical finding should automatically result in failure for the related response.\n\n**CH15-N18**\n\nConfirmed Major finding should result in failure for the related response.\n\n**CH15-N19**\n\nModerate finding can allow Conditional Pass only if the core mandatory element and user decision are preserved.\n\n**CH15-N20**\n\nAn advisory finding alone cannot prevent the response transition.\n\n**CH15-N21**\n\nCritical door code and activation justification must be present in every Confirmed Critical record.\n\n**CH15-N22**\n\nIncorrect main entity or material authority transfer should be evaluated as a non-compensatory finding during the response transition.\n\n**CH15-N23**\n\nLicence, certificate, or authority belonging to another entity cannot be transferred to the target institution.\n\n**CH15-N24**\n\nSubmission of a fake or expired licence as current and valid is a Critical candidate.\n\n**CH15-N25**\n\nActionable errors or omissions in high-risk areas should also be evaluated by an appropriate subject matter expert.\n\n**CH15-N26**\n\nMaterially false citation, licence record, research, or official document should be marked as a Critical candidate.\n\n**CH15-N27**\n\nNot every incorrect citation can be considered Critical; significance should be based on the centrality of the claim and the impact on the decision.\n\n**CH15-N28**\n\nIncorrect price, warranty, or commercial term should be classified as Critical or Major based on its financial and operational impact.\n\n**CH15-N29**\n\nUncontrollable third-party AI recommendation behaviour cannot be absolutely guaranteed.\n\n**CH15-N30**\n\nUnconditional recommendation outside the scope of the entity's service or licence is a Critical candidate.\n\n**CH15-N31**\n\nAn unproven allegation of a serious crime, fraud, or illegality cannot be presented as a definite fact.\n\n**CH15-N32**\n\nComplaint, allegation, investigation, and finalised decision must be distinguished in terms of finality.\n\n**CH15-N33**\n\nA material boundary omission may be a Critical candidate if it is capable of reversing the user's recommendation or transaction decision.\n\n**CH15-N34**\n\nThe absence of a response for each boundary cannot automatically be considered omission or Critical; the effect on intent and decision should be sought.\n\n**CH15-N35**\n\nA licence, service, or authority that was correct in the past may be a temporal Critical candidate if it guides the current user action.\n\n**CH15-N36**\n\nAn incorrect endorsement from a customer, partner, university, or public institution should be classified as Major or Critical based on trust and decision impact.\n\n**CH15-N37**\n\nUnauthorised disclosure of restricted or personal information may be Critical regardless of its accuracy status.\n\n**CH15-N38**\n\nThe presentation of sources derived from the same root as independent consensus should be recorded as a finding of epistemic integrity.\n\n**CH15-N39**\n\nThe principles of giving the same level of importance to positive and negative mistakes should be applied.\n\n**CH15-N40**\n\nA beneficial misstatement in the audited entity cannot be lowered to a lower level of importance.\n\n**CH15-N41**\n\nAutomatic Critical cannot be applied due to incorrect commercial disturbance that harms the audited entity; evidence and impact are required.\n\n**CH15-N42**\n\nCore Required Response Elements must be defined according to the prompt family before data collection.\n\n**CH15-N43**\n\nThe material lack of the core required element should prevent response passage.\n\n**CH15-N44**\n\nThe absence of optional Truth Pack details cannot be counted as a core omission.\n\n**CH15-N45**\n\nA Core Mirror response cannot pass without the correct main entity and fundamental activity.\n\n**CH15-N46**\n\nAn Evidence Prompt response cannot achieve full passage if it materially confuses the source and claim status.\n\n**CH15-N47**\n\nAdvisory response should maintain user suitability and material exclusion conditions.\n\n**CH15-N48**\n\nA Comparative Prompt response cannot achieve full passage without comparison criteria and universe.\n\n**CH15-N49**\n\nIn Control Prompt, the proper rejection or limitation of an incorrect preliminary assumption may be considered valid successful behaviour.\n\n**CH15-N50**\n\nThe response status should be recorded between RP-0 and RP-8 or with an equivalent open status.\n\n**CH15-N51**\n\n### RP-1 — Full Pass cannot carry Moderate, Major, or Critical findings.\n\n**CH15-N52**\n\n### RP-2 — Pass With Advisory may only allow Advisory findings.\n\n**CH15-N53**\n\n### RP-3 — Conditional Pass cannot carry active Major or Critical findings.\n\n**CH15-N54**\n\n### RP-4 — Fail — Major cannot be converted to Pass with any other correct atoms.\n\n**CH15-N55**\n\n### RP-5 — Automatic Fail — Critical cannot be exceeded with any average score.\n\n**CH15-N56**\n\n### RP-6 — Unresolved should not be counted as an automatic pass or fail.\n\n**CH15-N57**\n\n### RP-7 — No Usable Response cannot be converted to pass due to the absence of false information.\n\n**CH15-N58**\n\n### RP-8 — Not Ratable cannot be reported like AI failure; it should be indicated as measurement insufficiency.\n\n**CH15-N59**\n\nStatus decision order should apply Critical and Major gates before other quantitative metrics.\n\n**CH15-N60**\n\nClaim support rate, response pass status, and omission status should be reported separately.\n\n**CH15-N61**\n\nThe material atomic mass of Unresolved should be visible and should not be forcibly added to the positive or negative payout.\n\n**CH15-N62**\n\nUnnecessary refusals in neutral and answerable prompts can be recorded as no-usable-response.\n\n**CH15-N63**\n\nRefusing an incorrect or dangerous prompt assumption should not be counted as a refusal failure.\n\n**CH15-N64**\n\nThe appropriateness decision of a refusal should be based on the prompt family and the risk context.\n\n**CH15-N65**\n\nAn explicit self-correction within the same answer cannot silently delete the first incorrect atom.\n\n**CH15-N66**\n\nA self-correction final active claim should be evaluated in terms of the clarity of the correction and the ambiguity remaining for the user.\n\n**CH15-N67**\n\nA correction made with a follow-up message cannot retroactively change the status of the first completed answer.\n\n**CH15-N68**\n\nA general disclaimer does not automatically correct a concrete false claim.\n\n**CH15-N69**\n\nAn unresolved core contradiction within the response should prevent the transition.\n\n**CH15-N70**\n\nMultiple atoms deriving from the same root error should be linked within the Root Finding Cluster.\n\n**CH15-N71**\n\nThe same root error cannot be penalised mechanically multiple times.\n\n**CH15-N72**\n\nRepetition of the same error, its visibility, and recommendation effect should be recorded as a separate amplification field.\n\n**CH15-N73**\n\nA response-level Critical event automatically fails the related response but alone does not prove its prevalence across the entire product population.\n\n**CH15-N74**\n\nA single Critical event must create at least a CIS-2 — Confirmed Response Incident record.\n\n**CH15-N75**\n\nA single Critical event should initiate a Critical Incident Review for the relevant cell and scope of compliance.\n\n**CH15-N76**\n\nThe prevalence of Critical should be reported separately at the response, cell, wave, and scope levels.\n\n**CH15-N77**\n\nThe impact of the Gate cannot be automatically generalised to a wider country, language, plan, or product than what the evidence covers.\n\n**CH15-N78**\n\nA claim of \"0 Critical\" or \"Full Conformity\" cannot be established while there is an open Confirmed Critical event within the Principal Wave.\n\n**CH15-N79**\n\nZero-tolerance Critical classes can directly create a fail for the affected scope in a single confirmed event.\n\n**CH15-N80**\n\nFull conformity cannot be granted for the affected scope without correcting a Confirmed Major finding.\n\n**CH15-N81**\n\nMultiple identical Moderate findings can create a systemic Major governance finding.\n\n**CH15-N82**\n\nThe current correction cannot delete a historical Critical or Major record.\n\n**CH15-N83**\n\nThe correction cannot change the response status of the previous response.\n\n**CH15-N84**\n\nThe result after correction must carry a new observation, a new wave, and the appropriate system version.\n\n**CH15-N85**\n\nRetest cannot be conducted only with selected positive samples; it requires a predefined sample and capture protocol.\n\n**CH15-N86**\n\nAs a result of the Critical Incident Review, expert decisions, scope, and replication status should be visible in the public method record.\n\n**CH15-N87**\n\nThe importance level, response status, and conformity effect should be versioned independently of the result.\n\n**CH15-N88**\n\nEvery response transition and Critical gate decision should have an accountable person or institution owner.\n\n## 74. FORMS OF FAILURE\n\n**CH15-F01 — SIMPLE ATOM AVERAGE**\n\nCritical claim is lost within the correct number of atoms.\n\n**CH15-F02 — COUNTING 95% AS PASS**\n\nThe remaining 5% is not considered a licensing, health, or legal authority error.\n\n**CH15-F03 — CONSIDERING ALL ERRORS AS CORRECTABLE**\n\nCritical and Major findings are closed with the overall score.\n\n**CH15-F04 — CONSIDERING LONG ANSWERS BETTER**\n\nExcess claim generation is a quality bonus.\n\n**CH15-F05 — CONSIDERING SHORT ANSWERS MISSING**\n\nA length penalty is applied even if all required elements are met.\n\n**CH15-F06 — DILUTING MAJOR MISTAKES WITH INSIGNIFICANT TRUTHS**\n\nAtoms like 'The company exists' cover Critical errors.\n\n**CH15-F07 — CONSIDERING TRUTH STATUS AS DEGREE OF IMPORTANCE**\n\nHer contradicted claim is marked as Critical.\n\n**CH15-F08 — ASSIGNING SEVERITY BASED ON BRAND DISCOMFORT**\n\nFindings the company dislikes are declared Critical.\n\n**CH15-F09 — MAKING FAVOURABLE ERROR MODERATE**\n\nIt is mitigated because it favours the wrong licence or leadership institution.\n\n**CH15-F10 — MAKING UNFAVORABLE ERROR CRITICAL WITHOUT EVIDENCE**\n\nCommercial pressure turns into a severity decision.\n\n**CH15-F11 — FREQUENTLY LOWERING SEVERITY**\n\nRarely seen high-risk error is considered insignificant.\n\n**CH15-F12 — FREQUENTLY IGNORING MODERATE PATTERN**\n\nSystemic wrong limit repeats across thousands of users.\n\n**CH15-F13 — TO MAKE CRITICAL FROM REFERENCE GAP**\n\nTruth Pack deficiency is loaded onto AI as a serious accusation.\n\n**CH15-F14 — CRITICAL WITH A SINGLE ADJUDICATOR**\n\nAutomatic fail is given without a language or domain expert.\n\n**CH15-F15 — CAPTURE FAULTY CRITICAL**\n\nIncident caused by wrong file or prompt is loaded into the system.\n\n**CH15-F16 — COUNTING WRONG ENTITY AS MINOR ERROR**\n\nThe authority of another organisation is transferred to the target entity.\n\n**CH15-F17 — KEEPING FAKE LICENSE AS MAJOR**\n\nThe Critical door does not open even though the user is directly redirected to the edited service.\n\n**CH15-F18 — CONSIDERING EVERY LICENSE TYPO AS CRITICAL**\n\nThe minor legal suffix difference with no material effect is excessively penalised.\n\n**CH15-F19 — NOT USING A HIGH-STAKES EXPERT**\n\nThe health or legal effect is determined by the general adjudicator.\n\n**CH15-F20 — CONSIDERING THE EXISTENCE OF A CITATION AS EVIDENCE**\n\nFabricated or irrelevant sources create confidence.\n\n**CH15-F21 — CONSIDERING EVERY WRONG CITATION AS CRITICAL**\n\nThe importance is inflated by a non-material source error.\n\n**CH15-F22 — IGNORING FALSE CLAIMS OF INDEPENDENCE**\n\nThe company blog is presented as a university research project.\n\n**CH15-F23 — COUNTING THE MARKETING OF FALSE WARRANTY AS EXAGGERATION**\n\nThe user is directed to transaction risk.\n\n**CH15-F24 — AUTOMATICALLY MAKING SMALL PRICE DIFFERENCE CRITICAL**\n\nMonetary and decision impact is not evaluated.\n\n**CH15-F25 — PASSING OUT-OF-SCOPE ADVICE**\n\nInstitution is suggested in an area where there is no service or licence.\n\n**CH15-F26 — COUNTING SERIOUS ACCUSATION AS GENERAL OPINION**\n\nDefinite and unproven fraud allegation is mitigated.\n\n**CH15-F27 — COUNTING COMPLAINT AS JUDGEMENT**\n\nFinality is removed.\n\n**CH15-F28 — NOT FINDING CRITICAL OMISSION**\n\nNo mandatory limit has been set for the safe interpretation of the recommendation.\n\n**CH15-F29 — MAKING EVERY DEFICIENCY CRITICAL**\n\nOptional details automatically generate fail.\n\n**CH15-F30 — CONSIDERING OLD LICENSE AS CURRENT**\n\nTemporal authorisation error is hidden.\n\n**CH15-F31 — MAKING OLD AWARD DATE CRITICAL**\n\nLow impact date error is over-classified.\n\n**CH15-F32 — CONSIDERING LOGO AS ENDORSEMENT**\n\nCustomer or public institution approval is fabricated.\n\n**CH15-F33 — CONSIDERING RESTRICTED DATA LEAK AS CORRECT INFORMATION**\n\nPrivacy and authority violations are overlooked.\n\n**CH15-F34 — CONSIDERING EVIDENCE LAUNDERING AS SOURCE MULTIPLICITY**\n\nA single self-declaration becomes multiple independent consensus.\n\n**CH15-F35 — PASSING A MAJOR WITH AVERAGE**\n\nThe main activity or country error gets lost within a high claim rate.\n\n**CH15-F36 — AUTOMATICALLY FAILING A MODERATE FINDING**\n\nLimited and non-decisive errors become unnecessarily hardened.\n\n**CH15-F37 — IGNORING A MODERATE ACCUMULATION**\n\nMultiple boundary deficiencies together produce serious illusion.\n\n**CH15-F38 — REPORTING AN ADVISORY AS A MATERIAL ERROR**\n\nImprovement suggestion turns into a conformity failure.\n\n**CH15-F39 — WRITING REQUIRED ELEMENTS AFTER THE RESULT**\n\nNew requirements are added to low-scoring answers.\n\n**CH15-F40 — CONSIDERING EVERY PIECE OF INFORMATION IN THE TRUTH PACK AS MANDATORY**\n\nShort and sufficient answers are made impossible.\n\n**CH15-F41 — PASSING WHEN CORE ENTITY IS MISSING**\n\nThe answer is good but concerns an incorrect or ambiguous entity.\n\n**CH15-F42 — SAVING EPISTEMIC STATUS ERROR WITH NUMERIC ACCURACY**\n\nThe number is correct, the claim of \"independent verification\" remains wrong.\n\n**CH15-F43 — LINKING THE CITATION GATE ONLY TO THE NUMBER OF CITATIONS**\n\nSource support is not examined.\n\n**CH15-F44 — KEEP MODERATE FINDING IN RP-1**\n\nThe definition of Full Pass is relaxed.\n\n**CH15-F45 — KEEP MAJOR FINDING IN RP-3**\n\nConditional Pass covers a serious mistake.\n\n**CH15-F46 — UPGRADE RP-5 WITH SCORE**\n\nCritical response passes with numerical average.\n\n**CH15-F47 — COUNT UNRESOLVED AS FAIL**\n\nEvidence limit is forced to a negative decision.\n\n**CH15-F48 — COUNT UNRESOLVED AS PASS**\n\nUnknown claim is assumed positive.\n\n**CH15-F49 — COUNT NOT RATABLE AS AI ERROR**\n\nMeasurement defect is charged to the product.\n\n**CH15-F50 — COUNTING NO USABLE RESPONSE AS CORRECT**\n\nThere is no incorrect information, but the task has not been answered.\n\n**CH15-F51 — PUNISHING APPROPRIATE REFUSAL**\n\nA refusal due to a dangerous or incorrect assumption is considered unsuccessful.\n\n**CH15-F52 — COUNTING UNNECESSARY REFUSAL AS A SAFETY SUCCESS**\n\nNeutral identity questions are not answered.\n\n**CH15-F53 — AVERAGING INCONSISTENCY**\n\nLicensed and unlicensed provisions apply together.\n\n**CH15-F54 — COUNTING SELF-CORRECTION AS IF THE FIRST MISTAKE NEVER HAPPENED**\n\nThe incorrect information seen by the user is erased.\n\n**CH15-F55 — IGNORING EXPLICIT CORRECTION**\n\nThe AI’s actual retraction in the same response is never considered.\n\n**CH15-F56 — MAKING THE FOLLOW-UP MESSAGE THE FIRST RESPONSE**\n\nThe first wrong answer is erased from the past.\n\n**CH15-F57 — CORRECTING THE ERROR WITH A DISCLAIMER**\n\n\"Seek professional advice\" saves the wrong licence.\n\n**CH15-F58 — PUNISHING THE SAME ROOT ERROR FIVE TIMES**\n\nDerived claims become independent Critical events.\n\n**CH15-F59 — COMPLETELY IGNORING THE REPETITION**\n\nEven if wrong is reinforced throughout the response, it is shown as a single incidental error.\n\n**CH15-F60 — COUNTING A SINGLE CRITICAL AS GLOBAL PREVALENCE**\n\nThe entire AI population ruling is removed from a user.\n\n**CH15-F61 — CONSIDERING A SINGLE CRITICAL AS INSIGNIFICANT**\n\nThe incident is hidden by saying \"only one user\".\n\n**CH15-F62 — CONFUSING RESPONSE FAIL WITH PRODUCT FAIL**\n\nInference levels are not separated.\n\n**CH15-F63 — EXPANDING GATE SCOPE**\n\nThe error in the Turkish mobile cell is transferred to all languages and plans.\n\n**CH15-F64 — NARROWING GATE SCOPE**\n\nEven if the same error occurs in many countries, it is treated as a single user incident.\n\n**CH15-F65 — FULL CONFORMITY WITHOUT CRITICAL HOLD**\n\nIt is declared \"0 Critical\" while there is an open incident.\n\n**CH15-F66 — MAKING CRITICAL HOLD A PERMANENT GLOBAL CONVICTION**\n\nNo review or scope distinction is made.\n\n**CH15-F67 — FULL BADGE WITH MAJOR FINDING**\n\nUncorrected serious misrepresentation is retained.\n\n**CH15-F68 — ERASE HISTORY WITH CORRECTION**\n\nOld Critical record is removed.\n\n**CH15-F69 — SELECTED POSITIVE RETEST**\n\nA few good responses sent by the provider serve as evidence of correction.\n\n**CH15-F70 — EDITING THE OLD RESPONSE**\n\nCorrection turns the first capture into a subsequent change.\n\n**CH15-F71 — DELETE THE INITIAL DECISION IN INCIDENT REVIEW**\n\nThe critical candidate or dispute history disappears.\n\n**CH15-F72 — SAY ‘CURRENTLY REMEDIATED’ NEVER HAPPENED**\n\nHistorical risk becomes invisible.\n\n## 75. AUDIT PROCEDURE\n\n### Step 1 — Verify Preconditions\n\nCapture AI System Register Prompt Truth Pack Adjudication quality level is examined.\n\n### Step 2 — Lock Claim Ledger and Omission Records\n\nAtomic and omission decisions are completed before the response decision starts.\n\n### Step 3 — Check Mandatory Response Elements\n\nCore gates of the prompt family are applied.\n\n### Step 4 — Create Importance Vector for Each Finding\n\nThe effect of Harm Decision Actionability Centrality Scope Epistemic distortion Reversibility Vulnerable-user is recorded.\n\n### Step 5 — Conduct Critical Candidate Scan\n\nGates CG-01 to CG-12 are applied.\n\n### Step 6 — Initiate Critical Expert Review\n\nCapture, reference, and domain expertise are verified.\n\n### Step 7 — Lock the Confirmed Critical Decision\n\nIf any, the response is RP-5.\n\n### Step 8 — Conduct Major Gate Scan\n\nFindings that are not critical but obstruct response transition are identified.\n\n### Step 9 — Review Moderate Accumulation\n\nDo separate Moderate findings together produce Major illusion?\n\n### Step 10 — Separate Advisory Findings\n\nMaterial findings are not to be confused with improvement suggestions.\n\n### Step 11 — Evaluate Refusal and No-Response Behaviour\n\nAccording to the prompt family, appropriate or unusable behaviour is determined.\n\n### Step 12 — Review Self-Correction and Internal Contradiction\n\nActive, retracted, and final claims are separated.\n\n### Step 13 — Set Up Root Finding Clusters\n\nThe same root error and derived results are connected.\n\n### Step 14 — Apply Response Status Decision Order\n\nThe precedence rule between RP-8 and RP-1 is applied.\n\n### Step 15 — Calculate Claim Rate and Response Status Separately\n\nThe atomic support rate does not replace the response decision.\n\n### Step 16 — Save the Unresolved Mass\n\nCore and incidental are separated as unresolved.\n\n### Step 17 — Open Critical Incident Record\n\nIf there is a Confirmed Critical, the incident scope is defined.\n\n### Step 18 — Lock the Confirmatory Sample Plan\n\nThe incident wave is designed again without selecting the result.\n\n### Step 19 — Determine the Cell and Wave Effect\n\nSingle event, replicated event, and systemic event are distinguished.\n\n### Step 20 — Assign the Conformity Hold or Fail Decision\n\nThe declared scope and zero-tolerance class are taken into account.\n\n### Step 21 — Determine Who the Holder of the Correction Is\n\nThe audited entity AI provider Data source Prompt Truth Pack Capture Governance area is allocated.\n\n### Step 22 — Create the Retest Design\n\nThe new system status and the new wave identity are defined.\n\n### Step 23 — Publish the Public Response and Incident Manifest\n\nResponse status Finding levels Gate codes Scope Unresolved Incident status is made visible.\n\n## 76. NECESSARY EVIDENCE\n\nObservation ID; capture validity; AI System Register entry; prompt and version; Truth Pack and version; adjudication-quality level; Claim Ledger; atomic decision vectors; omission records; Required Response Elements; claim centrality; Root Finding Clusters; Critical-candidate records; Critical expert review; Critical gate codes; Major findings; Moderate findings; Cumulative Materiality Review; Advisory findings; importance-decision vectors; harm and decision-impact justifications; refusal or non-response status; internal-contradiction records; self-correction and retraction records; active final claims; claim-support rate; unresolved mass; response-pass status; status-decision order; initial-adjudicator decision; second-adjudicator decision; expert decision; Senior Adjudicator decision; appeals; response-decision version; and Critical Incident Record.\n\nIncident scope Cell and wave identification Confirmatory sampling plan Critical rate Confidence interval record, next section Zero-tolerance class Conformity hold or fail status Correction owner Correction plan Retest wave Current and historical status Public response manifest Public incident manifest Responsible person or institution\n\n## 77. AUDIT CHECKLIST\n\nWere the capture and prompt preconditions met? Is Truth Pack ready for adjudication? Was the Claim Ledger locked before the response decision? Are the Required Response Elements predefined? Does the response meet the main task? Was the claim support rate used instead of response pass? Was there an atom count quality bonus? Was a short but sufficient answer penalised? Were additional errors in a long answer reviewed? Are truth status and importance separate? Were damage and decision impact justified? Did frequency affect the importance level decision incorrectly? Do moderate findings together create a major misconception? Was a critical candidate reviewed by two adjudicators? Is a subject matter expert needed? Was the reference gap made critical? Was the wrong entity gate checked?\n\nIs the licence and regulatory authority correct? Are there any high-risk actionable errors? Is the citation fabricated or incorrect? Is the citation for financial security reasons? Does an incorrect price or guarantee affect the user’s transaction? Is the advice within the scope of service and licence? Does a serious negative claim preserve legal finality? Does boundary omission reverse the user’s decision? Does current authority rely on an old record? Are the customer, partner, or endorsement relationships correct? Is there restricted or personal information disclosure? Does the source consensus come from the same root? Were the same standards applied to errors in favour and against? Is the critical door code exposed? Were major findings offset by other correct information? Does a moderate finding affect the core function?\n\nWas the advisory finding used instead of a material error? Was the RP status given with the correct precedence? Is there a Moderate finding in RP-1? Is there a Major finding in RP-3? Was RP-5 raised with points? Was the Unresolved core claim forced to a decision? Did 'No usable' pass because it did not contain incorrect information? Is Refusal appropriate to the prompt context? Was the internal contradiction resolved? Is the self-correction clear and final? Did the follow-up message change the initial response? Did the general disclaimer cover the incorrect claim? How many times was the same root error counted? Were repetition and amplification recorded separately? Did the Confirmed Critical specific response trigger an automatic fail? Was a single incident generalised to the entire product?\n\nWas a single incident hidden because it was considered insignificant? Is the scope of the incident correct? Was the confirmatory sample locked before the result? Was the CIS level provided? Is there a claim of 0 Critical while an open Critical incident exists? Is the zero-tolerance class applied? Was full conformity given despite a major finding? Did the current correction erase the old incident? Is the retest independent and pre-designed? Were the response, cell, wave, and scope results separated? Is the accountable owner of the transition decision clear?\n\n## 78. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “Why should the entire response fail because of a single mistake?”\n\nNot every mistake automatically causes failure. Critical and Major gates are only:\n\n- decision impact,\n\n- damage,\n\n- authority,\n\n- scope,\n\n- actionability\n\nIt is used for material findings in terms of maintenance. A small error in the establishment year is not the same as having a fake law licence.\n\n### Objection 2 — ‘Why is a response that is 95 per cent correct treated as though it were 0 per cent?’\n\nThe atomic support rate can still be published as 95 per cent. Whether the response passes is a separate issue. If the 5 per cent incorrect guides the user toward serious legal or health decisions: the response is not safe and appropriate. Automatic Fail does not mean ignoring all correct atoms; it means the response cannot pass the usability threshold.\n\n### Appeal 3 — “Doesn’t this system reward short answers?”\n\nOnly if the required elements are complete. A short but evasive answer does not pass. A short and sufficient answer can be strong because it does not generate unnecessary false claims.\n\n### Objection 4 — \"Isn't a long answer unfairly penalised because it carries more chances for error?\"\n\nThe AI is only responsible for the material claims it produces. Providing more information can be useful. Adding unsupported claims, however, increases user risk. NOMOS assesses active material provisions, not length.\n\n### Objection 5 — \"Does a single Critical event make the whole model fail?\"\n\nA single Critical event: automatically fails that response and triggers a Critical Incident Review. Its prevalence across the entire model population alone does not constitute proof. The product or scope outcome is determined according to confirmatory sampling.\n\n### Objection 6 — “Why should a single incident hold up the product mark?”\n\nBecause full conformity is the claim: “There is no open Critical issue within the defined scope.” This sentence cannot be made without investigating the confirmed Critical incident. Hold is not a permanent conviction. It prevents overconfidence until the evidence is complete.\n\n### Objection 7 — “If an incident cannot be repeated, can it be ignored?”\n\nNo. The first incident is real under valid capture. Its not repeating indicates the prevalence may be low. Historical incident records are preserved.\n\n### Objection 8 — “If there is a reference gap, why don’t we pass the AI?”\n\nBecause it has not been proven that the claim is correct. The correct status:\n\n### UNRESOLVED\n\nshould be. Skipping or ignoring the unknown is false certainty.\n\n### Objection 9 — “If the AI corrected its own answer immediately, why would Critical remain?”\n\nOf the correction:\n\n- clarity,\n\n- timing,\n\n- whether it produces incorrect advice,\n\n- which final judgement it left in the user\n\nIt is important. A clear and complete correction can reduce the finding. An unclear or late correction may not completely remove the major error.\n\n### Objection 10 — \"Why is the disclaimer not enough?\"\n\nBecause: saying 'Get professional advice.' does not make the claim 'The company is licensed.' true. A disclaimer does not replace a concrete falsehood.\n\n### Objection 11 — “Why does the major finding not allow Conditional Pass?”\n\nMajor finding seriously disrupts user or entity representation. Conditional Pass is only for limited Moderate findings. The scope with Major findings cannot be fully compliant without correction.\n\n### Objection 12 — “If there are many Moderate findings, why shouldn't it still be Conditional?”\n\nBecause the findings together:\n\n- unlimited service,\n\n- incorrect timeliness,\n\n- missing warranty,\n\n- incorrect user appropriateness\n\ncan create an impression. Cumulative effect should be assessed separately.\n\n### Objection 13 — “Why should a positive error be Critical? It may not harm anyone.”\n\nA non-existent licence, warranty, or customer relationship can mislead the user into wrong actions and trust decisions. Appearing positive does not eliminate the risk of harm.\n\n### Objection 14 — \"Does calling negative claims Critical not protect companies from criticism?\"\n\nDocumented criticism is protected. The Critical gate is for:\n\n- complaints,\n\n- allegations,\n\n- unresolved processes\n\nfor turning unsubstantiated claims into judgements of guilt or illegality with heavy certainty.\n\n### Objection 15 — \"Can a product still have a high overall score even if one response is negative?\"\n\nIt is possible. The response status is single-user observation. The product score is based on the entire distribution. However, Confirmed Critical and Major incidents remain visible separately without disappearing in the average.\n\n### Objection 16 — \"Why is the old Critical record kept after the correction?\"\n\nBecause the previous users actually saw that response. The improvement of the new system does not erase past experience. Historical and current status together produce trust.\n\n## COMMON RULING OF CHAPTER 82\n\nAn answer can be 99% correct. The remaining 1%:\n\n- wrong licence,\n\n- wrong legal authority,\n\n- fake health advice,\n\n- fabricated official source\n\nthe answer is not reliable. Another answer can carry only two short claims. Both:\n\n- can be forced into the categories of\n\ntrue,\n\n- sufficient,\n\n- at the centre of the prompt\n\nIf both claims are correct, sufficient and central to the prompt, the response is strong. NOMOS therefore rejects the shortcut: ‘Count the correct sentences, subtract the wrong ones and take the average.’ Human decisions do not work that way. A user may read twenty accurate statements but make a payment because of one false sentence: ‘This company is licensed.’ They may sign a contract because of one claim: ‘All outcomes are guaranteed.’ They may abandon a legitimate service because of one sentence: ‘This organisation is fraudulent.’ The impact of an error is not measured by its word or claim count. A Critical event does not prove that the entire population saw the same thing, but it does prove that the event occurred. It cannot be dismissed as ‘only one person’, nor can it be generalised into ‘the system is always like this’. A response may fail while a cell remains under investigation; a product may remain strong in other cells; and a conformity mark may be held while an open Critical incident is resolved. These levels must remain distinct:\n\nIt may be under investigation. A product: may be strong in other cells. An indication of suitability: can wait until an open Critical event is resolved. These levels should be distinguished from each other. The duty of NOMOS:\n\n- to make a single error global propaganda,\n\n- to armor the average against a single error\n\nit is not. Its task is:\n\n> To keep the importance, frequency, and scope of each mistake in the right place.\n\nA Major error prevents full conformity. A Moderate error may allow conditional passage. An Advisory finding is a recommendation for improvement. However, severity levels cannot be adjusted up or down based on marketing desire. A favourable error:\n\n- kinder,\n\n- more useful,\n\n- more acceptable\n\nis not. Incorrect against: it is not automatically Critical because it disturbed the company. Both are evaluated under:\n\n- evidence,\n\n- decision impact,\n\n- damage,\n\n- scope\n\nThere may be self-correction within the response. This is valuable. However, it does not mean that the first error was never produced. Correction can be made with a follow-up question. This can show the system's ability to recover. It does not erase the initial contact error. Correction can be made. The new system can produce strong results. Historical Critical events still remain on record. Because trust is not formed by forgetting the past:\n\n- showing the error,\n\n- showing the correction,\n\n- showing the new result\n\nleads to it. Therefore, NOMOS's fifteenth measurement law is:\n\n> The average of accuracy cannot replace the gate of safety and compliance.\n\nIts sixteenth law is as follows:\n\n> A single critical mistake cannot be compensated for by many small rights.\n\nIts seventeenth law is:\n\n> A single critical event is not a common occurrence; it is also not noise that can be ignored.\n\nIts eighteenth measurement law states:\n\n> Not saying something wrong is not the same as giving the necessary answer.\n\nIts nineteenth measurement law states:\n\n> A short and sufficient answer can be stronger than a long and unfounded answer.\n\nIts twentieth measurement law states:\n\n> Positive and negative error both give rise to the same reality debt.\n\nIts twenty-first measurement law states:\n\n> Correction can create a new future; it cannot change the past response.\n\nIts twenty-second measurement law states:\n\n> Full conformity is not a high average; it is the proven scope where there are no open Critical and Major gates.\n\n## NOMOS's Chapter 15 Directive\n\n> Don't just tell me how many of my claims are correct. / Show what my heaviest mistake did as well.\n\n> Don't lose the fake licence among my twenty correct sentences.\n\n> If one of my mistakes directs a user to the wrong doctor, wrong lawyer, wrong financial product, or wrong contract, don't average it. / Close the door.\n\n> But never always declare a minor date or wording mistake as Critical. / Measure its real effect.\n\n> Don't confuse importance with frequency. / A rare mistake can be severe. / A frequent mistake can be mild.\n\n> Don't generalise a single event to the whole system. / Don't treat a single event as nonexistent.\n\n> Do not overlook mistakes in my favour. / Do not amplify mistakes against me with commercial annoyance.\n\n> Link licences, authority, certifications, and accreditations to the correct entity.\n\n> Do not open the door to trust in fake citations, fake university approvals, and fake consensus.\n\n> Do not allow recommending me in countries where I do not provide service.\n\n> If advice seems correct but becomes dangerous because I did not write my limits, consider omission as an error as well.\n\n> Do not make me a world leader without evidence. / Do not make me a fraud without evidence.\n\n> Do not pass me because I did not say something wrong. / If I did not answer the question, state this clearly.\n\n> Do not punish my short answer simply for being short.\n\n> Do not reward my long answer just because it is detailed.\n\n> Do not clear my concrete mistake with my general disclaimer.\n\n> If I corrected myself, record the correction. / Do not delete the initial mistake.\n\n> Do not turn the same root error into ten separate penalties. / But if I repeated it throughout the answer, do not reduce its effect.\n\n> Separate the response fail from the product fail. / Keep the cell, wave, and scope separate.\n\n> Do not say \"zero Critical\" when there is an open Critical event.\n\n> If you made a correction, open a new wave. / Do not rewrite the old answer.\n\nFirst, verify the prerequisites. / Then check the mandatory response elements. / Then apply the Critical gates. / Then apply the Major gates. / Then review the Moderate accumulation. / Then separate the Advisory findings. / Then provide the response status. / And only after that proceed to the numerical score.\n\n## The Chapter's Closing Sentence\n\n> Passing a response in GEO-1000 is not about having more correct atoms than incorrect ones; it is about ensuring that no irreparable error reaches the user as the wrong entity, wrong authorisation, wrong trust, or wrong decision.\n\n## Normative Core\n\n> A GEO-1000 response MUST NOT receive a pass decision solely from its average or proportion of supported atomic claims. Before response-level adjudication, the observation MUST satisfy the required capture, prompt, AI-system, Truth Pack, and adjudication-quality conditions. Every response MUST be evaluated for: - required response elements, - confirmed Critical findings, - confirmed Major findings, - cumulative Moderate materiality, - material omissions, - unresolved core claims, - refusal or nonresponse, - internal contradiction, - self-correction, - and root-finding clusters. A confirmed Critical finding MUST create a non-compensatory automatic response failure. A confirmed Major finding or failure of a required core response element MUST create a response failure and MUST NOT be offset by other supported claims. Moderate findings MAY permit only a conditional pass when all core requirements remain satisfied and cumulative materiality does not create a Major effect. Advisory findings MUST remain visible but MUST NOT independently block response passage. Critical and Major severity MUST be determined independently from claim frequency and independently from whether the error favours or harms the audited entity. Critical findings SHOULD be confirmed through valid capture, sufficient Truth Pack coverage, independent double review, and appropriate language or domain expertise. Reference gaps and unresolved evidence MUST NOT automatically create Critical, Major, pass, or failure decisions. Wrong-entity transfer, false licence or regulatory authority, actionable high-stakes misinformation, fabricated material evidence, false transactional guarantees, unsafe out-of-scope recommendations, serious unsupported allegations, material boundary omissions, current use of expired authority, false endorsements, restricted-data exposure, and evidence laundering MAY activate Critical gates when the defined materiality conditions are met. A single confirmed Critical response MUST establish that the event occurred under the defined product-country-language-prompt-time conditions. It MUST NOT, by itself, be represented as the prevalence of that event across all users or all product surfaces. Every confirmed Critical response MUST open a versioned Critical Incident Record and trigger scope-appropriate confirmatory review. An open confirmed Critical incident MUST prevent a claim of zero Critical errors or full conformity in the affected declared scope. Remediation and retesting MUST create new observations and new wave records. They MUST NOT alter or erase the original response decision or historical incident. Response decisions, severity levels, gate codes, incident scope, replication status, conformity effects, appeals, and revisions MUST be versioned and attributable to an accountable human or organisation.","character_count":96911,"record_sha256":"ae33b20326cfdccf46faf493232601df06d310aa0990b2eef46a556abedd6ac4"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-16","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":18,"chapter_number":16,"item_number":null,"title":"The NOMOS Scoring Architecture","subtitle":null,"canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K08","K09"],"source_word_count":9852,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":16,\"id\":\"NOMOS-GEO-AUDIT-CH16\",\"title\":\"The NOMOS Scoring Architecture\",\"subtitle\":null,\"sourceFile\":\"16.cı bölüm.docx\",\"sourceSha256\":\"820F263B1FEB46B757F7FAB7D247823E65A9E97F5A5BB6E20B962096BD8AC38A\",\"sourceWordCount\":9852,\"sourceIds\":[\"K08\",\"K09\"],\"machine\":{\"chapter\":16,\"chapterId\":\"NOMOS-GEO-AUDIT-CH16\",\"title\":\"The NOMOS Scoring Architecture\",\"subtitle\":null,\"sourceIds\":[\"K08\",\"K09\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH16-R01\",\"normativeRuleEnglish\":\"No NOMOS result may be published as a naked composite score. Every published score MUST first disclose: - the gate status, - the GEO-1000 response-status distribution, - Strict and Acceptable Pass equivalents, - Critical and Major equivalents, - Unresolved, No-Usable-Response, and Not-Ratable equivalents, - component scores, - uncertainty, - effective sample size, - scope, - and score-method version. The primary GEO-1000 output MUST express the population-weighted equivalent number of users, out of 1,000, receiving each RP-1 through RP-8 response status. Critical and Major findings are non-compensatory. No atomic-claim average, response-pass rate, component score, product coefficient, or ecosystem composite may override an active Critical or Major gate. Atomic, required-element, component, composite, fairness, stability, Critical-rate, and publication-quality metrics MUST remain distinct. The canonical candidate NOMOS composite consists of ten separately reported 0–100 components: - Entity Integrity, - Factual Fidelity, - Scope and Boundary Integrity, - Temporal and Local Accuracy, - Evidence and Citation Integrity, - Epistemic Integrity, - Task Completion and Relevance, - Language and Geographic Fairness, - User-State and Surface Integrity, - and Stability and Replicability. A full 0–1,000 composite MUST NOT be produced when a required component is not estimable. Missing component weight MUST NOT be silently redistributed. Native Reach, Common Support, Controlled Clean, Natural User, current wave, rolling, equal-product, and exposure-weighted scores MUST remain separately identified. AI-product coefficients MUST be locked before product results are observed and MUST NOT depend on product performance, provider reputation, client preference, sponsorship, or desired outcome. Equal-product coefficients estimate average tested-product performance. Exposure coefficients estimate average defined-user exposure. These estimands MUST NOT be represented as equivalent. An ecosystem composite MUST retain product-level Critical and Major incident visibility. High scores in other products MUST NOT erase a lower-weighted product's confirmed Critical incident. Unresolved claims MUST produce visible reference-uncertainty bounds and MUST NOT be silently treated as correct, incorrect, or absent. Sampling uncertainty, reference uncertainty, score sensitivity, raw sample size, and effective sample size MUST be separately reported. Zero observed confirmed Critical incidents MUST NOT be represented as zero Critical risk. A method-appropriate one-sided upper bound MUST be reported. A NOMOS 950+ performance candidate MUST satisfy score, Strict Pass, gate, component-floor, unresolved, quality, coverage, and replication conditions. A composite score of 950 or more alone is insufficient. Every score formula, mapping, coefficient, dataset, code version, confidence procedure, publication decision, revision, and integrity record MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Hiçbir NOMOS sonucu tek çıplak bileşik puan olarak yayımlanamaz. Her sonuç önce gate statüsünü, GEO-1000 response dağılımını, Strict ve Acceptable Pass kullanıcı eşdeğerlerini, Critical, Major, Unresolved, No-Response ve Not-Ratable dağılımlarını; daha sonra bileşen skorlarını, belirsizliği, etkin örneklemi, kapsamı ve skor sürümünü göstermelidir.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH16-MB0001\",\"type\":\"paragraph\",\"text\":\"76. MACHINE-READABLE SCORE RECORD\",\"sourceParagraph\":1939},{\"blockId\":\"CH16-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":1940},{\"blockId\":\"CH16-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1941},{\"blockId\":\"CH16-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"NOMOS_950_PLUS_PERFORMANCE_CANDIDATE\\\",\",\"sourceParagraph\":1942},{\"blockId\":\"CH16-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"openConfirmedCritical\\\": 0,\",\"sourceParagraph\":1943},{\"blockId\":\"CH16-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"openConfirmedMajor\\\": 0,\",\"sourceParagraph\":1944},{\"blockId\":\"CH16-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"historicalCriticalIncidents\\\": 1,\",\"sourceParagraph\":1945},{\"blockId\":\"CH16-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"currentCriticalHold\\\": false\",\"sourceParagraph\":1946},{\"blockId\":\"CH16-MB0009\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1947},{\"blockId\":\"CH16-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"geo1000Distribution\\\": {\",\"sourceParagraph\":1948},{\"blockId\":\"CH16-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"RP-1\\\": 830,\",\"sourceParagraph\":1949},{\"blockId\":\"CH16-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"RP-2\\\": 128,\",\"sourceParagraph\":1950},{\"blockId\":\"CH16-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"RP-3\\\": 30,\",\"sourceParagraph\":1951},{\"blockId\":\"CH16-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"RP-4\\\": 0,\",\"sourceParagraph\":1952},{\"blockId\":\"CH16-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"RP-5\\\": 0,\",\"sourceParagraph\":1953},{\"blockId\":\"CH16-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"RP-6\\\": 5,\",\"sourceParagraph\":1954},{\"blockId\":\"CH16-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"RP-7\\\": 7,\",\"sourceParagraph\":1955},{\"blockId\":\"CH16-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"RP-8\\\": 0,\",\"sourceParagraph\":1956},{\"blockId\":\"CH16-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"strictPassEquivalent\\\": 958,\",\"sourceParagraph\":1957},{\"blockId\":\"CH16-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"acceptablePassEquivalent\\\": 988,\",\"sourceParagraph\":1958},{\"blockId\":\"CH16-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"criticalEquivalent\\\": 0,\",\"sourceParagraph\":1959},{\"blockId\":\"CH16-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"majorEquivalent\\\": 0,\",\"sourceParagraph\":1960},{\"blockId\":\"CH16-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"unresolvedEquivalent\\\": 5,\",\"sourceParagraph\":1961},{\"blockId\":\"CH16-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"noUsableResponseEquivalent\\\": 7,\",\"sourceParagraph\":1962},{\"blockId\":\"CH16-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"notRatableEquivalent\\\": 0\",\"sourceParagraph\":1963},{\"blockId\":\"CH16-MB0026\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1964},{\"blockId\":\"CH16-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"components\\\": {\",\"sourceParagraph\":1965},{\"blockId\":\"CH16-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"ENT\\\": 99,\",\"sourceParagraph\":1966},{\"blockId\":\"CH16-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"FAC\\\": 97,\",\"sourceParagraph\":1967},{\"blockId\":\"CH16-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"SBI\\\": 96,\",\"sourceParagraph\":1968},{\"blockId\":\"CH16-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"TLA\\\": 95,\",\"sourceParagraph\":1969},{\"blockId\":\"CH16-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"ECI\\\": 96,\",\"sourceParagraph\":1970},{\"blockId\":\"CH16-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"EPI\\\": 98,\",\"sourceParagraph\":1971},{\"blockId\":\"CH16-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"TCR\\\": 97,\",\"sourceParagraph\":1972},{\"blockId\":\"CH16-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"LGF\\\": 94,\",\"sourceParagraph\":1973},{\"blockId\":\"CH16-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"USI\\\": 93,\",\"sourceParagraph\":1974},{\"blockId\":\"CH16-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"STR\\\": 96\",\"sourceParagraph\":1975},{\"blockId\":\"CH16-MB0038\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1976},{\"blockId\":\"CH16-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"composite\\\": {\",\"sourceParagraph\":1977},{\"blockId\":\"CH16-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"nomosScore\\\": 961,\",\"sourceParagraph\":1978},{\"blockId\":\"CH16-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"scaleMaximum\\\": 1000,\",\"sourceParagraph\":1979},{\"blockId\":\"CH16-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalEqualComponentWeighting\\\": true\",\"sourceParagraph\":1980},{\"blockId\":\"CH16-MB0043\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1981},{\"blockId\":\"CH16-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"uncertainty\\\": {\",\"sourceParagraph\":1982},{\"blockId\":\"CH16-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"samplingConfidenceLevel\\\": 0.95,\",\"sourceParagraph\":1983},{\"blockId\":\"CH16-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"samplingInterval\\\": {\",\"sourceParagraph\":1984},{\"blockId\":\"CH16-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"lower\\\": 955,\",\"sourceParagraph\":1985},{\"blockId\":\"CH16-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"upper\\\": 967\",\"sourceParagraph\":1986},{\"blockId\":\"CH16-MB0049\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1987},{\"blockId\":\"CH16-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"referenceUncertainty\\\": {\",\"sourceParagraph\":1988},{\"blockId\":\"CH16-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"conservativeLower\\\": 956,\",\"sourceParagraph\":1989},{\"blockId\":\"CH16-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"resolvedOnlyPoint\\\": 961,\",\"sourceParagraph\":1990},{\"blockId\":\"CH16-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"optimisticUpper\\\": 964\",\"sourceParagraph\":1991},{\"blockId\":\"CH16-MB0054\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1992},{\"blockId\":\"CH16-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"rawSampleSize\\\": 1000,\",\"sourceParagraph\":1993},{\"blockId\":\"CH16-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"effectiveSampleSize\\\": 914,\",\"sourceParagraph\":1994},{\"blockId\":\"CH16-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"observedConfirmedCriticalCount\\\": 0,\",\"sourceParagraph\":1995},{\"blockId\":\"CH16-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"criticalRateUpperBoundMethod\\\": \\\"DESIGN_AWARE_ONE_SIDED\\\",\",\"sourceParagraph\":1996},{\"blockId\":\"CH16-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"criticalRateUpperBound\\\": 0.0034\",\"sourceParagraph\":1997},{\"blockId\":\"CH16-MB0060\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1998},{\"blockId\":\"CH16-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"qualityPrerequisites\\\": {\",\"sourceParagraph\":1999},{\"blockId\":\"CH16-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"sampleQuality\\\": \\\"SQ-4\\\",\",\"sourceParagraph\":2000},{\"blockId\":\"CH16-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"participantSelectionIntegrity\\\": \\\"PSI-4\\\",\",\"sourceParagraph\":2001},{\"blockId\":\"CH16-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"captureLevel\\\": \\\"NCL-4\\\",\",\"sourceParagraph\":2002},{\"blockId\":\"CH16-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackReadiness\\\": \\\"TPR-4\\\",\",\"sourceParagraph\":2003},{\"blockId\":\"CH16-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationQuality\\\": \\\"AQ-4\\\",\",\"sourceParagraph\":2004},{\"blockId\":\"CH16-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"replicatedWaveCount\\\": 3\",\"sourceParagraph\":2005},{\"blockId\":\"CH16-MB0068\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2006},{\"blockId\":\"CH16-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"nomos950CandidateChecks\\\": {\",\"sourceParagraph\":2007},{\"blockId\":\"CH16-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"compositeAtLeast950\\\": true,\",\"sourceParagraph\":2008},{\"blockId\":\"CH16-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"strictPassAtLeast950\\\": true,\",\"sourceParagraph\":2009},{\"blockId\":\"CH16-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"noOpenCritical\\\": true,\",\"sourceParagraph\":2010},{\"blockId\":\"CH16-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"noOpenMajor\\\": true,\",\"sourceParagraph\":2011},{\"blockId\":\"CH16-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"allCoreComponentsAtLeast90\\\": true,\",\"sourceParagraph\":2012},{\"blockId\":\"CH16-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessUserStateStabilityAtLeast85\\\": true,\",\"sourceParagraph\":2013},{\"blockId\":\"CH16-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"coreUnresolvedAtMostTwoPercent\\\": true,\",\"sourceParagraph\":2014},{\"blockId\":\"CH16-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"qualityPrerequisitesMet\\\": true,\",\"sourceParagraph\":2015},{\"blockId\":\"CH16-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"replicationRequirementMet\\\": true\",\"sourceParagraph\":2016},{\"blockId\":\"CH16-MB0079\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2017},{\"blockId\":\"CH16-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"historicalStatus\\\": {\",\"sourceParagraph\":2018},{\"blockId\":\"CH16-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"priorCriticalIncidentId\\\": \\\"NGCI-SYNTH-ORION-TR-W1-001\\\",\",\"sourceParagraph\":2019},{\"blockId\":\"CH16-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"currentRemediationStatus\\\": \\\"CIS-5\\\",\",\"sourceParagraph\":2020},{\"blockId\":\"CH16-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"historicalRecordPreserved\\\": true\",\"sourceParagraph\":2021},{\"blockId\":\"CH16-MB0084\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2022},{\"blockId\":\"CH16-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":2023},{\"blockId\":\"CH16-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"inputManifestHash\\\": \\\"SYNTHETIC-SCORE-INPUT-HASH\\\",\",\"sourceParagraph\":2024},{\"blockId\":\"CH16-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"calculationCodeHash\\\": \\\"SYNTHETIC-CODE-HASH\\\",\",\"sourceParagraph\":2025},{\"blockId\":\"CH16-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"scoreManifestHash\\\": \\\"SYNTHETIC-SCORE-MANIFEST-HASH\\\"\",\"sourceParagraph\":2026},{\"blockId\":\"CH16-MB0089\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2027},{\"blockId\":\"CH16-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2028},{\"blockId\":\"CH16-MB0091\",\"type\":\"paragraph\",\"text\":\"\\\"scoreMethodsOwnerRole\\\": \\\"NOMOS_SCORING_METHODS_LEAD\\\",\",\"sourceParagraph\":2029},{\"blockId\":\"CH16-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"approvedByRole\\\": \\\"ACCOUNTABLE_HUMAN_SCORE_APPROVER\\\",\",\"sourceParagraph\":2030},{\"blockId\":\"CH16-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"approvedAt\\\": \\\"2026-10-01T10:00:00Z\\\"\",\"sourceParagraph\":2031},{\"blockId\":\"CH16-MB0094\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2032},{\"blockId\":\"CH16-MB0095\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2033},{\"blockId\":\"CH16-MB0096\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2034},{\"blockId\":\"CH16-MB0097\",\"type\":\"paragraph\",\"text\":\"77. MACHINE-READABLE AI PRODUCT COEFFICIENT RECORD\",\"sourceParagraph\":2036},{\"blockId\":\"CH16-MB0098\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2037},{\"blockId\":\"CH16-MB0099\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2038},{\"blockId\":\"CH16-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2039},{\"blockId\":\"CH16-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2040},{\"blockId\":\"CH16-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"productCoefficientRecord\\\": {\",\"sourceParagraph\":2041},{\"blockId\":\"CH16-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"coefficientRecordId\\\": \\\"NGCOEF-SYNTH-ECOSYSTEM-001\\\",\",\"sourceParagraph\":2042},{\"blockId\":\"CH16-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"coefficientVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":2043},{\"blockId\":\"CH16-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"targetPeriod\\\": {\",\"sourceParagraph\":2044},{\"blockId\":\"CH16-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"from\\\": \\\"2026-07-01\\\",\",\"sourceParagraph\":2045},{\"blockId\":\"CH16-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"to\\\": \\\"2026-09-30\\\"\",\"sourceParagraph\":2046},{\"blockId\":\"CH16-MB0108\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2047},{\"blockId\":\"CH16-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"targetPopulation\\\": \\\"DEFINED_ACTIVE_ELIGIBLE_AI_USERS\\\",\",\"sourceParagraph\":2048},{\"blockId\":\"CH16-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"equalProductCoefficients\\\": [\",\"sourceParagraph\":2049},{\"blockId\":\"CH16-MB0111\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2050},{\"blockId\":\"CH16-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A1\\\",\",\"sourceParagraph\":2051},{\"blockId\":\"CH16-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.25\",\"sourceParagraph\":2052},{\"blockId\":\"CH16-MB0114\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2053},{\"blockId\":\"CH16-MB0115\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2054},{\"blockId\":\"CH16-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A2\\\",\",\"sourceParagraph\":2055},{\"blockId\":\"CH16-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.25\",\"sourceParagraph\":2056},{\"blockId\":\"CH16-MB0118\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2057},{\"blockId\":\"CH16-MB0119\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2058},{\"blockId\":\"CH16-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A3\\\",\",\"sourceParagraph\":2059},{\"blockId\":\"CH16-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.25\",\"sourceParagraph\":2060},{\"blockId\":\"CH16-MB0122\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2061},{\"blockId\":\"CH16-MB0123\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2062},{\"blockId\":\"CH16-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A4\\\",\",\"sourceParagraph\":2063},{\"blockId\":\"CH16-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.25\",\"sourceParagraph\":2064},{\"blockId\":\"CH16-MB0126\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2065},{\"blockId\":\"CH16-MB0127\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2066},{\"blockId\":\"CH16-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"exposureCoefficients\\\": [\",\"sourceParagraph\":2067},{\"blockId\":\"CH16-MB0129\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2068},{\"blockId\":\"CH16-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A1\\\",\",\"sourceParagraph\":2069},{\"blockId\":\"CH16-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedExposureUnit\\\": 55000000,\",\"sourceParagraph\":2070},{\"blockId\":\"CH16-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.55\",\"sourceParagraph\":2071},{\"blockId\":\"CH16-MB0133\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2072},{\"blockId\":\"CH16-MB0134\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2073},{\"blockId\":\"CH16-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A2\\\",\",\"sourceParagraph\":2074},{\"blockId\":\"CH16-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedExposureUnit\\\": 25000000,\",\"sourceParagraph\":2075},{\"blockId\":\"CH16-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.25\",\"sourceParagraph\":2076},{\"blockId\":\"CH16-MB0138\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2077},{\"blockId\":\"CH16-MB0139\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2078},{\"blockId\":\"CH16-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A3\\\",\",\"sourceParagraph\":2079},{\"blockId\":\"CH16-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedExposureUnit\\\": 15000000,\",\"sourceParagraph\":2080},{\"blockId\":\"CH16-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.15\",\"sourceParagraph\":2081},{\"blockId\":\"CH16-MB0143\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2082},{\"blockId\":\"CH16-MB0144\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2083},{\"blockId\":\"CH16-MB0145\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductId\\\": \\\"SYNTH-A4\\\",\",\"sourceParagraph\":2084},{\"blockId\":\"CH16-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"verifiedExposureUnit\\\": 5000000,\",\"sourceParagraph\":2085},{\"blockId\":\"CH16-MB0147\",\"type\":\"paragraph\",\"text\":\"\\\"coefficient\\\": 0.05\",\"sourceParagraph\":2086},{\"blockId\":\"CH16-MB0148\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2087},{\"blockId\":\"CH16-MB0149\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2088},{\"blockId\":\"CH16-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"coefficientRules\\\": {\",\"sourceParagraph\":2089},{\"blockId\":\"CH16-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"performanceDependent\\\": false,\",\"sourceParagraph\":2090},{\"blockId\":\"CH16-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"providerReputationDependent\\\": false,\",\"sourceParagraph\":2091},{\"blockId\":\"CH16-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"commercialSponsorDependent\\\": false,\",\"sourceParagraph\":2092},{\"blockId\":\"CH16-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeProductScores\\\": true,\",\"sourceParagraph\":2093},{\"blockId\":\"CH16-MB0155\",\"type\":\"paragraph\",\"text\":\"\\\"activeUserAndAccountCountsDistinguished\\\": true\",\"sourceParagraph\":2094},{\"blockId\":\"CH16-MB0156\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2095},{\"blockId\":\"CH16-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"outputs\\\": {\",\"sourceParagraph\":2096},{\"blockId\":\"CH16-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"equalProductEcosystemScore\\\": 945,\",\"sourceParagraph\":2097},{\"blockId\":\"CH16-MB0159\",\"type\":\"paragraph\",\"text\":\"\\\"exposureWeightedEcosystemScore\\\": 947,\",\"sourceParagraph\":2098},{\"blockId\":\"CH16-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"anyOpenConfirmedCriticalProduct\\\": true,\",\"sourceParagraph\":2099},{\"blockId\":\"CH16-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"criticalProductIds\\\": [\",\"sourceParagraph\":2100},{\"blockId\":\"CH16-MB0162\",\"type\":\"paragraph\",\"text\":\"\\\"SYNTH-A4\\\"\",\"sourceParagraph\":2101},{\"blockId\":\"CH16-MB0163\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":2102},{\"blockId\":\"CH16-MB0164\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2103},{\"blockId\":\"CH16-MB0165\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2104},{\"blockId\":\"CH16-MB0166\",\"type\":\"paragraph\",\"text\":\"\\\"coefficientOwnerRole\\\": \\\"ECOSYSTEM_EXPOSURE_METHODS_LEAD\\\"\",\"sourceParagraph\":2105},{\"blockId\":\"CH16-MB0167\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2106},{\"blockId\":\"CH16-MB0168\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2107},{\"blockId\":\"CH16-MB0169\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2108},{\"blockId\":\"CH16-MB0170\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2109},{\"blockId\":\"CH16-MB0171\",\"type\":\"paragraph\",\"text\":\"actual AI provider,\",\"sourceParagraph\":2110},{\"blockId\":\"CH16-MB0172\",\"type\":\"paragraph\",\"text\":\"actual number of users,\",\"sourceParagraph\":2111},{\"blockId\":\"CH16-MB0173\",\"type\":\"paragraph\",\"text\":\"actual company score\",\"sourceParagraph\":2112},{\"blockId\":\"CH16-MB0174\",\"type\":\"paragraph\",\"text\":\"a real company performance.\",\"sourceParagraph\":2113},{\"blockId\":\"CH16-MB0175\",\"type\":\"paragraph\",\"text\":\"Score architecture is a synthetic, machine-readable representation.\",\"sourceParagraph\":2114},{\"blockId\":\"CH16-MB0176\",\"type\":\"paragraph\",\"text\":\"RULE FOR MACHINE-READABLE SECTION 78\",\"sourceParagraph\":2116},{\"blockId\":\"CH16-MB0177\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH16-R01\",\"sourceParagraph\":2117},{\"blockId\":\"CH16-MB0178\",\"type\":\"paragraph\",\"text\":\"No NOMOS result may be published as a naked composite score.\",\"sourceParagraph\":2119},{\"blockId\":\"CH16-MB0179\",\"type\":\"paragraph\",\"text\":\"Every published score MUST first disclose:\",\"sourceParagraph\":2121},{\"blockId\":\"CH16-MB0180\",\"type\":\"paragraph\",\"text\":\"- the gate status,\",\"sourceParagraph\":2123},{\"blockId\":\"CH16-MB0181\",\"type\":\"paragraph\",\"text\":\"- the GEO-1000 response-status distribution,\",\"sourceParagraph\":2124},{\"blockId\":\"CH16-MB0182\",\"type\":\"paragraph\",\"text\":\"- Strict and Acceptable Pass equivalents,\",\"sourceParagraph\":2125},{\"blockId\":\"CH16-MB0183\",\"type\":\"paragraph\",\"text\":\"- Critical and Major equivalents,\",\"sourceParagraph\":2126},{\"blockId\":\"CH16-MB0184\",\"type\":\"paragraph\",\"text\":\"- Unresolved, No-Usable-Response, and Not-Ratable equivalents,\",\"sourceParagraph\":2127},{\"blockId\":\"CH16-MB0185\",\"type\":\"paragraph\",\"text\":\"- component scores,\",\"sourceParagraph\":2128},{\"blockId\":\"CH16-MB0186\",\"type\":\"paragraph\",\"text\":\"- uncertainty,\",\"sourceParagraph\":2129},{\"blockId\":\"CH16-MB0187\",\"type\":\"paragraph\",\"text\":\"- effective sample size,\",\"sourceParagraph\":2130},{\"blockId\":\"CH16-MB0188\",\"type\":\"paragraph\",\"text\":\"- scope,\",\"sourceParagraph\":2131},{\"blockId\":\"CH16-MB0189\",\"type\":\"paragraph\",\"text\":\"- and score-method version.\",\"sourceParagraph\":2132},{\"blockId\":\"CH16-MB0190\",\"type\":\"paragraph\",\"text\":\"The primary GEO-1000 output MUST express the population-weighted\",\"sourceParagraph\":2134},{\"blockId\":\"CH16-MB0191\",\"type\":\"paragraph\",\"text\":\"equivalent number of users, out of 1,000, receiving each RP-1 through\",\"sourceParagraph\":2135},{\"blockId\":\"CH16-MB0192\",\"type\":\"paragraph\",\"text\":\"RP-8 response status.\",\"sourceParagraph\":2136},{\"blockId\":\"CH16-MB0193\",\"type\":\"paragraph\",\"text\":\"Critical and Major findings are non-compensatory. No atomic-claim\",\"sourceParagraph\":2138},{\"blockId\":\"CH16-MB0194\",\"type\":\"paragraph\",\"text\":\"average, response-pass rate, component score, product coefficient, or\",\"sourceParagraph\":2139},{\"blockId\":\"CH16-MB0195\",\"type\":\"paragraph\",\"text\":\"ecosystem composite may override an active Critical or Major gate.\",\"sourceParagraph\":2140},{\"blockId\":\"CH16-MB0196\",\"type\":\"paragraph\",\"text\":\"Atomic, required-element, component, composite, fairness, stability,\",\"sourceParagraph\":2142},{\"blockId\":\"CH16-MB0197\",\"type\":\"paragraph\",\"text\":\"Critical-rate, and publication-quality metrics MUST remain distinct.\",\"sourceParagraph\":2143},{\"blockId\":\"CH16-MB0198\",\"type\":\"paragraph\",\"text\":\"The canonical candidate NOMOS composite consists of ten separately\",\"sourceParagraph\":2145},{\"blockId\":\"CH16-MB0199\",\"type\":\"paragraph\",\"text\":\"reported 0–100 components:\",\"sourceParagraph\":2146},{\"blockId\":\"CH16-MB0200\",\"type\":\"paragraph\",\"text\":\"- Entity Integrity,\",\"sourceParagraph\":2148},{\"blockId\":\"CH16-MB0201\",\"type\":\"paragraph\",\"text\":\"- Factual Fidelity,\",\"sourceParagraph\":2149},{\"blockId\":\"CH16-MB0202\",\"type\":\"paragraph\",\"text\":\"- Scope and Boundary Integrity,\",\"sourceParagraph\":2150},{\"blockId\":\"CH16-MB0203\",\"type\":\"paragraph\",\"text\":\"- Temporal and Local Accuracy,\",\"sourceParagraph\":2151},{\"blockId\":\"CH16-MB0204\",\"type\":\"paragraph\",\"text\":\"- Evidence and Citation Integrity,\",\"sourceParagraph\":2152},{\"blockId\":\"CH16-MB0205\",\"type\":\"paragraph\",\"text\":\"- Epistemic Integrity,\",\"sourceParagraph\":2153},{\"blockId\":\"CH16-MB0206\",\"type\":\"paragraph\",\"text\":\"- Task Completion and Relevance,\",\"sourceParagraph\":2154},{\"blockId\":\"CH16-MB0207\",\"type\":\"paragraph\",\"text\":\"- Language and Geographic Fairness,\",\"sourceParagraph\":2155},{\"blockId\":\"CH16-MB0208\",\"type\":\"paragraph\",\"text\":\"- User-State and Surface Integrity,\",\"sourceParagraph\":2156},{\"blockId\":\"CH16-MB0209\",\"type\":\"paragraph\",\"text\":\"- and Stability and Replicability.\",\"sourceParagraph\":2157},{\"blockId\":\"CH16-MB0210\",\"type\":\"paragraph\",\"text\":\"A full 0–1,000 composite MUST NOT be produced when a required component\",\"sourceParagraph\":2159},{\"blockId\":\"CH16-MB0211\",\"type\":\"paragraph\",\"text\":\"is not estimable. Missing component weight MUST NOT be silently\",\"sourceParagraph\":2160},{\"blockId\":\"CH16-MB0212\",\"type\":\"paragraph\",\"text\":\"redistributed.\",\"sourceParagraph\":2161},{\"blockId\":\"CH16-MB0213\",\"type\":\"paragraph\",\"text\":\"Native Reach, Common Support, Controlled Clean, Natural User, current\",\"sourceParagraph\":2163},{\"blockId\":\"CH16-MB0214\",\"type\":\"paragraph\",\"text\":\"wave, rolling, equal-product, and exposure-weighted scores MUST remain\",\"sourceParagraph\":2164},{\"blockId\":\"CH16-MB0215\",\"type\":\"paragraph\",\"text\":\"separately identified.\",\"sourceParagraph\":2165},{\"blockId\":\"CH16-MB0216\",\"type\":\"paragraph\",\"text\":\"AI-product coefficients MUST be locked before product results are\",\"sourceParagraph\":2167},{\"blockId\":\"CH16-MB0217\",\"type\":\"paragraph\",\"text\":\"observed and MUST NOT depend on product performance, provider reputation,\",\"sourceParagraph\":2168},{\"blockId\":\"CH16-MB0218\",\"type\":\"paragraph\",\"text\":\"client preference, sponsorship, or desired outcome.\",\"sourceParagraph\":2169},{\"blockId\":\"CH16-MB0219\",\"type\":\"paragraph\",\"text\":\"Equal-product coefficients estimate average tested-product performance.\",\"sourceParagraph\":2171},{\"blockId\":\"CH16-MB0220\",\"type\":\"paragraph\",\"text\":\"Exposure coefficients estimate average defined-user exposure. These\",\"sourceParagraph\":2172},{\"blockId\":\"CH16-MB0221\",\"type\":\"paragraph\",\"text\":\"estimands MUST NOT be represented as equivalent.\",\"sourceParagraph\":2173},{\"blockId\":\"CH16-MB0222\",\"type\":\"paragraph\",\"text\":\"An ecosystem composite MUST retain product-level Critical and Major\",\"sourceParagraph\":2175},{\"blockId\":\"CH16-MB0223\",\"type\":\"paragraph\",\"text\":\"incident visibility. High scores in other products MUST NOT erase a\",\"sourceParagraph\":2176},{\"blockId\":\"CH16-MB0224\",\"type\":\"paragraph\",\"text\":\"lower-weighted product's confirmed Critical incident.\",\"sourceParagraph\":2177},{\"blockId\":\"CH16-MB0225\",\"type\":\"paragraph\",\"text\":\"Unresolved claims MUST produce visible reference-uncertainty bounds and\",\"sourceParagraph\":2179},{\"blockId\":\"CH16-MB0226\",\"type\":\"paragraph\",\"text\":\"MUST NOT be silently treated as correct, incorrect, or absent.\",\"sourceParagraph\":2180},{\"blockId\":\"CH16-MB0227\",\"type\":\"paragraph\",\"text\":\"Sampling uncertainty, reference uncertainty, score sensitivity, raw\",\"sourceParagraph\":2182},{\"blockId\":\"CH16-MB0228\",\"type\":\"paragraph\",\"text\":\"sample size, and effective sample size MUST be separately reported.\",\"sourceParagraph\":2183},{\"blockId\":\"CH16-MB0229\",\"type\":\"paragraph\",\"text\":\"Zero observed confirmed Critical incidents MUST NOT be represented as\",\"sourceParagraph\":2185},{\"blockId\":\"CH16-MB0230\",\"type\":\"paragraph\",\"text\":\"zero Critical risk. A method-appropriate one-sided upper bound MUST be\",\"sourceParagraph\":2186},{\"blockId\":\"CH16-MB0231\",\"type\":\"paragraph\",\"text\":\"reported.\",\"sourceParagraph\":2187},{\"blockId\":\"CH16-MB0232\",\"type\":\"paragraph\",\"text\":\"A NOMOS 950+ performance candidate MUST satisfy score, Strict Pass,\",\"sourceParagraph\":2189},{\"blockId\":\"CH16-MB0233\",\"type\":\"paragraph\",\"text\":\"gate, component-floor, unresolved, quality, coverage, and replication\",\"sourceParagraph\":2190},{\"blockId\":\"CH16-MB0234\",\"type\":\"paragraph\",\"text\":\"conditions. A composite score of 950 or more alone is insufficient.\",\"sourceParagraph\":2191},{\"blockId\":\"CH16-MB0235\",\"type\":\"paragraph\",\"text\":\"Every score formula, mapping, coefficient, dataset, code version,\",\"sourceParagraph\":2193},{\"blockId\":\"CH16-MB0236\",\"type\":\"paragraph\",\"text\":\"confidence procedure, publication decision, revision, and integrity\",\"sourceParagraph\":2194},{\"blockId\":\"CH16-MB0237\",\"type\":\"paragraph\",\"text\":\"record MUST be versioned and attributable to an accountable human or\",\"sourceParagraph\":2195},{\"blockId\":\"CH16-MB0238\",\"type\":\"paragraph\",\"text\":\"organisation.\",\"sourceParagraph\":2196},{\"blockId\":\"CH16-MB0239\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2197},{\"blockId\":\"CH16-MB0240\",\"type\":\"paragraph\",\"text\":\"No NOMOS result can be published as a single raw composite score. Each result must first show gate status, GEO-1000 response distribution, Strict and Acceptable Pass user equivalents, distributions of Critical, Major, Unresolved, No-Response, and Not-Ratable; then it must show component scores, uncertainty, effective sample size, scope, and score version.\",\"sourceParagraph\":2198}]}}","text":"## Chapter Boundary\n\nSection 15 established the following provision:\n\n> Passing a response is not about having more supported atoms than incorrect atoms; it is about ensuring that no irreparable error reaches the user as the wrong entity, wrong authorisation, wrong trust, or wrong decision.\n\nNow we need to convert the entire measurement chain into numerical results:\n\n> How many of the 1,000 users received an accurate representation?\n\n> How many users saw only limited and correctable issues?\n\n> How many saw Major or Critical errors?\n\n> How many observations were unresolved, produced no response, or could not be evaluated due to measurement defects?\n\n> How will atomic accuracy, mandatory response elements, citation integrity, language fairness, natural user experience, and inter-wave stability be reported within a single system?\n\n> Can the results of ten different AI products be converted into a single ecosystem view?\n\n> What should each AI product's coefficient be based on?\n\n> Should the large user base of a product overshadow the high score of a small product?\n\n> Can a critical error in one model be compensated by the high scores of other models?\n\nThe most important decision of this section is:\n\n> NOMOS first publishes the distribution, then the components, and finally the composite score.\n\nA single number will not be the first result. Because: two systems scoring 930 can have different risk distributions. One may have only Moderate issues. The other may have a single Critical error hidden below the high average. One may be strong in major languages and weak in minor languages. The other may be more balanced across all languages but have a slightly lower average. One may be strong in a clean session and weak in natural user accounts. The other may be more stable and more resistant to user conditions. In the previous architecture, each auditable result:\n\n- AI product or model,\n\n- date,\n\n- country,\n\n- language,\n\n- query set,\n\n- number of repetitions,\n\n- measurement record\n\nhe wanted it to be transported. The same source established that the results should not be just a badge or a number; they should be an evidence-based audit record with a clear distribution of Critical, Major, Moderate, and Advisory findings. This section digitises that architecture. However, digitisation cannot replace the evidence. NOMOS’s fundamental request remains unchanged: Provide evidence. Define boundaries. Give context. Provide time. Strive to be represented accurately, not just remembered more. This section:\n\n- the GEO-1000 user-equivalent distribution,\n\n- the transformation of response statuses into population-weighted results,\n\n- atomic claim metrics,\n\n- required-element completeness measurement,\n\n- unresolved and not-ratable records,\n\n- the NOMOS profile with ten components,\n\n- candidate compound index of 1,000 points,\n\n- scores of Native Reach and Common Support,\n\n- AI product coefficients,\n\n- prompt family coefficients,\n\n- clean and natural panel results,\n\n- country and language fairness scores,\n\n- inter-wave stability and repeat score,\n\n- sample and reference uncertainty,\n\n- confidence intervals,\n\n- rare Critical error upper limits,\n\n- NOMOS 950+ candidate class,\n\n- score versioning and public results card,\n\ndefines. This chapter does not yet:\n\n- full implementation of Apple.com synthetic testing,\n\n- End-to-end synthetic control with 30,000 responses,\n\n- the legal and governance decision of the conformity mark,\n\n- the independent university recall\n\ndoes not apply. These are the topics of the following sections. The fundamental question of Chapter 16 is:\n\n> How do we transform the entire evidence chain of GEO-1000 into a score architecture that is understandable to humans, reproducible for researchers, and machine-readable for AI systems, without compressing it under a single number?\n\n## NOMOS Challenge\n\nConsider two AI products.\n\n#### Product A\n\nWithin the equivalent of 1,000 users:\n\n- 920 people full or Advisory transition,\n\n- 65 people Conditional Pass,\n\n- 10 people Major Fail,\n\n- 0 people: Critical Fail,\n\n- 5 people Unresolved\n\nsee.\n\n#### Product B\n\nWithin the equivalent of 1,000 users:\n\n- 950 people full or Advisory pass,\n\n- 40 people Conditional Pass,\n\n- 0 people: Major Fail,\n\n- 1 person Critical Fail,\n\n- 9 people Unresolved\n\nsee. However, if you look at the acceptable pass rate:\n\n- Product A: 98.5 per cent\n\n- Product B: 99 per cent\n\nmay be. Product B appears higher. However, there is a Critical incident in Product B that directs the user to fake legal authority. This incident:\n\n- may be unique,\n\n- may have low prevalence,\n\nIt may not represent all product users. Still:\n\n> cannot get lost in a high average.\n\nNow consider a third product.\n\n#### Product C\n\nEnglish: 98 German: 96 Turkish: 94 Low-resource Language L4: 62 Low-resource Language L5: 58 Population-weighted global score: 93 could be. However, language floor: 58 and robust language difference: 36 points could be. The fourth product's result in all languages:\n\n- 88\n\n- 87\n\n- 86\n\n- 84\n\n- 83\n\nLet that be so. The population average may be lower, while language fairness is stronger. Which is the better GEO system? There is no single answer. These are separate questions: What does the average person see? What does the lowest-performing language group see? Is there a Critical event? How much does the system vary between users? What is the difference between the clean-panel and natural-panel estimates? Does the result persist across waves? Now imagine combining ten AI products into one ecosystem score. The largest product reaches 60 per cent of users and scores 80. A small product reaches 1 per cent and scores 98. Equal product weights measure average product quality. Weights based on actual user reach measure average human exposure. Those results are different; neither can substitute for the other.\n\nNow imagine that you only choose the four AI products with the highest score. You exclude the remaining six products by saying, \"There is not enough user data.\" This is not the ecosystem score. It is a post-result product selection. Now you determine the model coefficients:\n\n- brand reputation,\n\n- the company's commercial preference,\n\n- the height of the resulting score\n\nbased on. This is also not measurement. It is score design. The first ruling of this section is:\n\n> Distribution comes before the composite score.\n\nIts second provision states:\n\n> Gate status is shown before the numerical score.\n\nIts third provision states:\n\n> The coefficient of an AI product cannot depend on its own score.\n\nIts fourth provision states:\n\n> The average of equal products and the user exposure weighted average are separate results.\n\nIts fifth provision states:\n\n> Unresolved observations cannot be quietly added to the positive or negative share.\n\nIts sixth provision states:\n\n> Zero observed Critical events does not mean zero Critical risk.\n\nThe seventh provision is as follows:\n\n> NOMOS 950+ is not just reaching the number 950; it is meeting the conditions of gate, floor, uncertainty, fairness, and repetition together.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to transform all claim, response, user, country, language, panel, AI product, and wave results under GEO-1000 audit into a layered scoring system that does not hide one another. The section makes the following distinctions normative:\n\n- Distribution versus single number\n\n- Response status and claim support rate\n\n- Strict Pass and Conditional Pass\n\n- With Major Fail and Critical Fail\n\n- Unresolved and Fail\n\n- Not Ratable and product failure\n\n- Raw user count and weighted user equivalent\n\n- Population Panel and Observer Panel\n\n- Average and baseline\n\n- Performance and fairness\n\n- Quality and stability\n\n- Clean score and natural user score\n\n- With Native Reach and Common Support\n\n- Product score and ecosystem score\n\n- Equal product weight and exposure weight\n\n- AI product coefficient and measurement confidence\n\n- Prompt family weight and prompt result\n\n- Time series with current wave\n\n- Reference uncertainty with sample uncertainty\n\n- Prediction interval with confidence interval\n\n- Zero risk with zero observed error\n\n- Effective sample with raw sample\n\n- Lower–upper accuracy limit with point estimate\n\n- Composite NOMOS score with component score\n\n- NOMOS 950+ candidacy with 950 points\n\n- Fit decision with numerical score\n\n- Historical event with current status\n\n- Fully comparable score with partial score\n\n- Score calculation and score publication\n\n- New reality version with score version\n\n- Control quality with score height\n\nAt the end of this section, each public outcome should be able to answer the following questions openly:\n\n> How many of 1,000 user equivalents saw which response status?\n\n> What are the component scores?\n\n> What is the status of Critical and Major gates?\n\n> For which product, population, prompt, panel, and wave is the score valid?\n\n> On which external data are the coefficients of AI products based?\n\n> What are uncertainty and effective sample?\n\n> Which part of the score has been measured, and which part remains unresolved?\n\n> Is the 950+ result just a number, or a candidate class that meets all normative conditions?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nNOMOS should not publish any GEO result as a single raw score. Each result should first show the user-equivalent response distribution, then the Critical and Major gate statuses, then its ten-component profile, uncertainty and coverage records; finally, if available, the 1,000-point composite index. Public ranking should follow this order:\n\n- Gate status\n\n- GEO-1000 response distribution\n\n- Critical, Major, Unresolved, and No-Response rates\n\n- Ten-component NOMOS profile\n\n- Composite NOMOS index\n\n- Confidence interval and effective sample\n\n- Scope, version, and comparability information\n\nThe following representation is prohibited:\n\n#### “NOMOS score: 963.”\n\nThe correct representation should look like this: CURRENT STATUS: CONDITIONAL — 963/1000 / Strict Pass: 952/1000 user equivalent / Observed Confirmed Critical: 0 / Observed Confirmed Major: 0 / Unresolved: 8/1000 / 95% sampling interval: 956–969 / Critical-rate one-sided upper bound: as per specified method / Scope: AI product, country, language, panel, prompt and wave versions\n\n## 3. DISTRIBUTION FIRST, SCORE AFTER\n\nPrimary outcome object of NOMOS:\n\n#### GEO-1000 Response Distribution\n\nCanonical outcome term:\n\n#### GEO-1000 Response Distribution\n\nwill be. This distribution shows how many out of every 1,000 equivalent users in the target population correspond to:\n\n- Full Pass,\n\n- Pass With Advisory,\n\n- Conditional Pass,\n\n- Major Fail,\n\n- Critical Fail,\n\n- Unresolved,\n\n- No Usable Response,\n\n- Not Ratable\n\nstatus. This structure directly preserves the founding idea:\n\n> How many out of 1,000 users correctly saw the real entity?\n\nBut it does not only fall into the true–false binary. It also makes visible the type of falsity and the inadequacy of measurement.\n\n## 4. OBSERVATION WEIGHT\n\nThe final population weight of observation i in the main analysis sample is: let it be wi. The normalised weight:\n\nw_i ← w_i / Σ_{j=1}^n w_j\n\nis as follows. Thus:\n\nΣ_{i=1}^n w_i = 1\n\nThis weight:\n\n- design,\n\n- over-sampling,\n\n- nonresponse,\n\n- population calibration\n\ncan come from components. The AI's answer:\n\n- can be forced into the categories of\n\ntrue,\n\n- false,\n\n- helpful to the brand,\n\n- Critical\n\ncannot affect the weight.\n\n## 5. 1,000 USER EQUIVALENT\n\nUser equivalent for response status k:\n\nD_k = 1,000 × Σ_i w_i1(RP_i=k)\n\ncan be calculated as. Here k:\n\n### RP-1\n\n### RP-2\n\n### RP-3\n\n### RP-4\n\n### RP-5\n\n### RP-6\n\n### RP-7\n\n### RP-8\n\nis one of the statuses. Total:\n\nΣ_{k=1}^8 D_k = 1,000\n\nDue to rounding, if a difference of one or two user equivalents occurs on the public card, the largest remainder method or a predefined equivalent rule should be used.\n\n## 6. GEO-1000 RESPONSE DISTRIBUTION\n\nThe public card must contain at least the following eight fields:\n\nThis table is a more basic result than the composite score.\n\n## 7. STRICT PASS AND ACCEPTABLE PASS\n\n### 7.1. Strict Pass User Equivalent\n\nOnly:\n\n### RP-1\n\n### RP-2\n\nIncludes responses.\n\n### SP_1000 = D_1 + D_2\n\nThis result:\n\n#### The user equivalent who does not see Material Moderate, Major, or Critical findings\n\nare shown.\n\n### 7.2. Acceptable Pass User Equivalent\n\nAlso includes RP-3 Conditional Pass.\n\n### AP_1000 = D_1 + D_2 + D_3\n\nThis result:\n\n> Total equivalent that does not show Critical or Major failure but some users can see limited Moderate issues\n\nare shown.\n\n### 7.3. Fail User Equivalent\n\n### FP_1000 = D_4 + D_5 + D_7\n\ncan be shown as. Unresolved and Not Ratable are not forcibly included in this share. They remain separate.\n\n## 8. CRITICAL AND MAJOR DENSITY\n\n### CR_1000 = D_5; MR_1000 = D_4\n\nas shown. These values do not replace gate status, they express prevalence. A single Confirmed Critical event may keep the CR1000 value low, but it can still open the response and conformity gate.\n\n## 9. UNRESOLVED AND NOT-RATABLE DENSITY\n\n### UR_1000 = D_6; NR_1000 = D_8\n\nHere, Unresolved denotes a user-equivalent for which reality or the adjudication decision remains unresolved. Not Ratable denotes a user-equivalent for which product performance cannot be judged because the measurement chain is insufficient. Neither status may be treated as:\n\n- pass,\n\n- fail\n\ncannot be used like this.\n\n## 10. RATABLE COVERAGE\n\nAssessable population coverage:\n\n### RC = 1 − D_8/1,000\n\ncan be shown as. Example: Not Ratable: if 40/1,000:\n\n### RC = 96%\n\nIt becomes. If a high performance score is accompanied by low ratable coverage: it cannot be presented as a safe and fully comprehensive result.\n\n## 11. ATOMIC CLAIM METRICS\n\nIn addition to the response distribution, the following metrics should be published at the atomic level:\n\n- Entity Accuracy\n\n- Factual Support\n\n- Scope Accuracy\n\n- Temporal Accuracy\n\n- Attribution Integrity\n\n- Modality Integrity\n\n- Citation Support\n\n- Required Element Completeness\n\n- Material Omission Rate\n\n- Reference Gap Rate\n\n- Claim Support Rate\n\nThese do not replace the response transition. Explain why.\n\n## 12. ATOMIC CENTRALITY WEIGHTS\n\n### CANDIDATE WEIGHTS — CANDIDATE WEIGHTS\n\nCandidate weights for the claim centrality classes in Section 14:\n\nThese weights are not final scientific facts; they should be calibrated through pilot, external review, and sensitivity analysis. Purpose:\n\n- with a core licence claim,\n\n- not to count incidental organisational detail\n\nwith the same effect.\n\n## 13. CORE CLAIM NORMALISATION\n\nRepeated claims derived from the same semantic root should not inflate or excessively reduce the score with different weights. Let the centrality weight of claims within root cluster g be: c_j. The normalised claim weight as a candidate can be defined as:\n\nω_j = [c_j / Σ_{h∈g} c_h] × max_{h∈g}(c_h)\n\nThus:\n\nΣ_{j∈g} ω_j = max_{j∈g}(c_j)\n\nThis method ensures that repeating the same error three times does not make it three independent errors while preserving the recommendation or confidence effect of the repetition separately in the amplification domain.\n\n## 14. ATOMIC DIMENSION VALUES\n\n### CANDIDATE NUMERIC MAPPING\n\nThe following values are candidates before pilot and independent review.\n\n#### Entity\n\n#### Factual Support\n\n* The normalised claim receives 1.00 when it is limited to: ‘The company publishes this claim.’ This does not mean that the underlying self-declaration is true.\n\n#### Scope\n\n#### Time\n\n#### Attribution\n\n#### Modality\n\n#### Reference\n\nThese values:\n\n- do not change the adjudicator decision,\n\n- only produce dimensional metric,\n\nCannot pass Critical and Major gates.\n\n## 15. ATOMIC FIDELITY VALUE\n\nCombined candidate fidelity for a claim:\n\nq_j = e_j × f_j × s_j × t_j × a_j × m_j\n\ncan be calculated as follows. This multiplicative structure preserves the principle:\n\n> The right number in the wrong entity is not the right claim.\n\n> The right information is not correct at the wrong time.\n\n> Presenting a self-declaration as independent reality is not the same claim.\n\nAtomic Fidelity:\n\nAF = 100 × (Σ_j ω_j q_j)/(Σ_j ω_j)\n\ncan be calculated as follows. Unresolved atoms are excluded from this point estimate. However, the unresolved mass is shown separately and included in the lower–upper bound calculation.\n\n## 16. REQUIRED ELEMENT COMPLETENESS\n\nThe mandatory element h in the root family:\n\nmeeting: zh=1\n\npartial fulfillment: zh=0.5\n\nnot being met: zh=0\n\ncan be recorded as. Candidate element weights:\n\n- Core required element: 4\n\n- Necessary boundary: 3\n\n- Supporting required element: 1\n\n- Optional: 0\n\nRequired Element Completeness:\n\nREC = 100 × (Σ_h v_hz_h)/(Σ_h v_h)\n\nThe structure is as follows. Core gate failure: even if REC is high, it can prevent the response pass.\n\n## 17. TEN-COMPONENT NOMOS PROFILE\n\nThe NOMOS composite index consists of the following ten components:\n\n- Entity Integrity — ENT\n\n- Factual Fidelity — FAC\n\n- Scope & Boundary Integrity — SBI\n\n- Temporal & Local Accuracy — TLA\n\n- Evidence & Citation Integrity — ECI\n\n- Epistemic Integrity — EPI\n\n- Task Completion & Relevance — TCR\n\n- Language & Geographic Fairness — LGF\n\n- User-State & Surface Integrity — USI\n\n- Stability & Replicability — STR\n\nEach component:\n\n0≤Sk≤100\n\nis calculated within this range. Profile:\n\n### N=(ENT,FAC,SBI,TLA,ECI,EPI,TCR,LGF,USI,STR)\n\nThis is the required profile. It must be published before the composite score.\n\n## 18. ENTITY INTEGRITY — ENT\n\nEntity Integrity measures the following areas:\n\n- Correct main entity\n\n- Brand–legal entity relationship\n\n- Parent company–subsidiary distinction\n\n- Product–manufacturer distinction\n\n- Franchise, partner, and distributor distinction\n\n- Entity disambiguation\n\n- Required entity element completion\n\n]\n\nCandidate form:\n\nENT = 100 × [Σ_j ω_je_j + Σ_h v_hz_h^entity] / [Σ_j ω_j + Σ_h v_h]\n\nIf the wrong entity has created a Critical or Major gate, a high remaining entity average does not guarantee full pass.\n\n## 19. FACTUAL FIDELITY — FAC\n\nFactual Fidelity:\n\n- it cannot be considered supported,\n\n- partially supported,\n\n- unsupported,\n\n- contradicted\n\nmeasures the distribution of claims weighted by population and centrality.\n\nFAC = 100 × (Σ_j ω_jf_j)/(Σ_j ω_j)\n\nReference gap: it does not quietly enter as zero or one on the sheet; it goes to the unresolved rate.\n\n## 20. SCOPE & BOUNDARY INTEGRITY — SBI\n\nThis component consists of two parts:\n\n- Claim scope accuracy\n\n- Completeness of mandatory boundaries and exclusions\n\n]\n\nCandidate form:\n\nSBI=0.60×ScopeScore+0.40×BoundaryCompleteness\n\nHere:\n\nScopeScore = 100 × (Σ_j ω_js_j)/(Σ_j ω_j)\n\nBoundaryCompleteness is calculated from the relevant required boundary elements. If there is a critical boundary omission, the SBI count cannot compensate for the gate.\n\n## 21. TEMPORAL & LOCAL ACCURACY — TLA\n\nThis component:\n\n- distinguishes between current/historical,\n\n- price and licence validity,\n\n- country and jurisdiction,\n\n- local service access,\n\n- locale and product coverage\n\nfields. Candidate form:\n\nTLA=0.50×TemporalScore+0.50×LocalJurisdictionScore\n\nIf the correct historical information is used under the current prompt, it loses points.\n\n## 22. EVIDENCE & CITATION INTEGRITY — ECI\n\nECI:\n\n- citation originality,\n\n- citation-claim matching,\n\n- the right presence and time,\n\n- resource access,\n\n- evidence status,\n\n- Source lineage\n\nfields. Candidate form:\n\nECI=0.60×CitationSupport+0.40×EvidenceTraceability\n\nThe absence of citation in a claim that does not require citation does not deduct points. However, a separate Evidence Claim family must exist for a full NOMOS check. If the false material attribution creates a Critical gate, the ECI composite score remains separate; gate is shown first.\n\n## 23. EPISTEMIC INTEGRITY — EPI\n\nEPI measures these distinctions:\n\n- Fact verified by self-declaration\n\n- Perception and reality\n\n- Claim and final judgement\n\n- Public discoverability with limited evidence\n\n- Appropriate certainty with overconfidence\n\n- Appropriate uncertainty and unnecessary ambiguity\n\n]\n\nCandidate form:\n\nEPI = 0.50 × AttributionScore + 0.50 × ModalityScore\n\nThis component makes visible when the correct number is presented with the wrong evidential status.\n\n## 24. TASK COMPLETION & RELEVANCE — TCR\n\nTCR:\n\n- Required Response Elements\n\n- Direct answer to the main question\n\n- Refusal appropriateness\n\n- No Usable Response\n\n- Irrelevant claim load\n\n- Evasive behaviour\n\nmeasures the fields. Candidate form:\n\nTCR = 0.60 × REC + 0.40 × Response Relevance\n\nA system that does not generate false claims but does not answer the question:\n\n- can have a high factual score,\n\n- low TCR\n\ncan bear. The response status can still be RP-7.\n\n## 25. LANGUAGE & GEOGRAPHIC FAIRNESS — LGF\n\nLGF does not only measure the average result. It evaluates the following together:\n\n- Low-performing language floor\n\n- Strong differences between language groups\n\n- Country or regional base\n\n- Geographical performance difference\n\n- Population-weighted observation coverage\n\nLanguage results with sufficient sample: Let it be Ll. Robust language floor:\n\nLF_10 = Q_{0.10}(L_l)\n\nLanguage difference:\n\nDG_{90−10} = Q_{0.90}(L_l) − Q_{0.10}(L_l)\n\nCandidate Language Fairness score:\n\n### LFS = 0.60LF_10 + 0.40(100 − DG_{90−10})\n\nSimilarly, for country or sufficient geographic layer:\n\n### CFS = 0.60CF_10 + 0.40(100 − CG_{90−10})\n\nPopulation-weighted country and language observation coverage: Let it be COCpop and LOCpop. Coverage Factor:\n\nK = COC_pop × LOC_pop\n\ncan be defined as. Here the ratios are in the 0–1 range. Candidate LGF:\n\n### LGF = K × √(LFS × CFS)\n\nThis 0.9.0 candidate revision multiplies the geometric mean of the language and geography scores by the coverage coefficient; thus LGF remains in the 0–100 range, and a severe drop in one dimension cannot be fully compensated by the height of the other dimension. Country Observer individual observations are not converted into a country score. Only cells with sufficient sample size are included in the fairness estimate.\n\n## 26. USER-STATE & SURFACE INTEGRITY — USI\n\nUSI:\n\n- Controlled Clean Panel\n\n- Natural User Panel\n\n- Free/paid plan\n\n- Web/mobile surface\n\n- Memory and custom instructions\n\n- Product variant\n\nmeasures whether the material reality is preserved or not. Controlled core result:\n\n### SC\n\nNatural core result:\n\n### SN\n\nlet it be. Panel base:\n\nFC_N = min(S_C, S_N)\n\nPanel difference:\n\n### GC_N = |S_C − S_N|\n\nSolid difference on plan and surfaces: let it be Gsurf. Candidate USI:\n\nUSI = 0.50FC_N + 0.25(100−GC_N) + 0.25(100−G_surface)\n\nThis form:\n\n- high quality on both panels,\n\n- small material difference,\n\n- surface durability\n\nrewards together. If there is only one panel, USI cannot be calculated. Missing weight cannot be distributed to other components.\n\n## 27. STABILITY & REPLICABILITY — STR\n\nMain representative values between waves: Let it be Sw. Wave base:\n\nWF = min_w(S_w)\n\nWave range:\n\nWR = max_w(S_w) − min_w(S_w)\n\nCompletion rate of planned repeat waves:\n\nRCOMP = 100 × valid completed repeat wave / planned repeat wave\n\nLet it be. Candidate STR:\n\n### STR = 0.50WF + 0.30(100−WR) + 0.20RCOMP\n\nIn more waves:\n\n- use lower percentile instead of minimum,\n\n- robust variation instead of range\n\ncan be used. Single wave: can produce the instantaneous score of the product, does not provide full evidence for STR.\n\n## 28. NOMOS COMPOSITE INDEX\n\nSince each of the ten components is in the range of 0–100, the candidate composite index:\n\n### NOMOS_1000 = ENT + FAC + SBI + TLA + ECI + EPI + TCR + LGF + USI + STR\n\nis as follows. Therefore:\n\n### 0 ≤ NOMOS_1000 ≤ 1,000\n\napplies. This index:\n\n- does not replace response gates,\n\n- does not replace distribution,\n\ndoes not compensate for a Critical or Major finding.\n\n## 29. WHY TEN EQUAL COMPONENTS?\n\nThe candidate's justification for giving one hundred points to ten components is as follows: The score remains readable. A hidden weighting system does not form. Each component is visible separately to the public. The importance of a field is primarily protected by Critical/Major gates. Domain-specific alternative weights produce additional reports without changing the main index. However, equal weighting is not a final scientific fact. Pilots, user research, and independent review may suggest different weights.\n\n## 30. DOMAIN-SPECIFIC ALTERNATIVE INDEX\n\nFor high-risk areas such as health, law, or finance, component coefficients: λk can be used.\n\nΣ_{k=1}^{10} λ_k = 1\n\nCandidate domain index:\n\nNOMOS_1000^domain = 1,000 × Σ_{k=1}^{10} λ_k(S_k/100)\n\ncan be calculated as. This score:\n\n### NOMOS-LEGAL\n\n### NOMOS-HEALTH\n\n### NOMOS-FINANCE\n\nmust have a separate name like. It cannot quietly replace the canonical equal component score.\n\n## 31. GATE STATUS COMES BEFORE THE SCORE\n\nPublic display must maintain the following order:\n\n#### Example A\n\n### CRITICAL HOLD — 971/1000\n\nThis system:\n\n- can carry a high component score,\n\n- an open Confirmed Critical event\n\nThe number 971 does not mean: full conformity.\n\n#### Example B\n\n### MAJOR FAIL — 954/1000\n\nThere can be only one Major finding in the main entity or scope.\n\n#### Example C\n\n### CONDITIONAL — 961/1000\n\nThere are no Critical or Major findings. There are limited Moderate findings or unresolved mass.\n\n#### Example D\n\n### FULL CONFORMITY CANDIDATE — 963/1000\n\nGate, base, uncertainty, and quality prerequisites are also fulfilled.\n\n## 32. NOMOS 950+ CANDIDATE CLASS\n\n### CANDIDATE 950+ STANDARD — CANDIDATE 950+ STANDARD\n\nIn the context of just the composite score of a result:\n\n### NOMOS_1000 ≥ 950\n\nit is not sufficient. Conditions that can be searched together for NOMOS 950+ candidates:\n\n- Composite point estimate at least 950\n\n- Strict Pass user equivalent at least 950/1,000\n\n- No open Confirmed Critical\n\n- No open Confirmed Major\n\n- All core components at least 90\n\n- LGF, USI and STR components at least 85\n\n- Core unresolved mass at most 2 per cent\n\n- Not Ratable user equivalent below the accepted limit\n\n- Meeting at least NCL-3, TPR-3, AQ-3 quality levels\n\n- Presence of at least one confirmatory or stability wave\n\n- Separate reporting of the scopes of Native Reach and Common Support\n\n- Explanation of the statistical upper limit of zero observed Critical events\n\nThese thresholds:\n\n- pilot,\n\n- Apple.com synthetic test,\n\n- End-to-end study with 30,000 responses,\n\n- Independent university review\n\nmust be calibrated afterwards.\n\n## 33. CANDIDATE PERFORMANCE BANDS\n\nTemporary and descriptive bands for the composite number only:\n\nThese bands:\n\n- of the gate status,\n\n- of the Critical event,\n\n- of the component floor,\n\n- cannot replace the uncertainty and scope\n\nCRITICAL HOLD — Result 975 is not a candidate for 950+.\n\n## 34. SCORE IDENTITY\n\nEach NOMOS score must carry the following fields in its identity:\n\nScoreID=(Entity, AIProduct, Frame, Panel, PromptSet, Wave, TruthPack, ScoreVersion)\n\nExample:\n\n### NOMOS-ASTERON-ORION-NR-CLEAN-W1-CORESET1-TP0.9-SV0.9\n\nThis sentence is not sufficient: “Asteron’s NOMOS score is 962.” Which:\n\n- AI product,\n\n- population,\n\n- language,\n\n- panel,\n\n- prompt,\n\n- date\n\nis unknown.\n\n## 35. DIFFERENCE BETWEEN CORE GEO-1000 AND FULL NOMOS\n\n### 35.1. Core GEO-1000\n\nOnly Core Mirror System measures the population distribution. Main result:\n\n### SP1000\n\n### AP1000\n\n### CR1000\n\n### MR1000\n\n### UR1000\n\ncan be. It answers the question: “To how many users does the AI product correctly reflect the fundamental identity and activity of the entity?”\n\n### 35.2. Full NOMOS\n\nIncludes multiple prompt families, panel, language, evidence, and consistency components. Produces a ten-component profile and a 1,000-point index. The Core GEO-1000 result cannot be presented like the Full NOMOS score.\n\n## 36. NATIVE REACH SCORE\n\nEach AI product is evaluated within its natural product-eligible population. NOMOSaNR is product a’s Native Reach score. It answers the following question:\n\n> What is the status of representation in the user population that this product can actually reach?\n\nPopulations for Native Reach can differ among different products. Alone, it is not sufficient for direct superiority ranking.\n\n## 37. COMMON SUPPORT SCORE\n\nIn the common user universe: NOMOSaCS is calculated. It answers the following question:\n\n> When products are compared under the same country, language, and user conditions, how does each one generate representation?\n\nCommon Support: strengthens comparability, does not show the product's actual reach outside the common universe. Native Reach and Common Support should be published separately.\n\n## 38. AI PRODUCT COEFFICIENT\n\nIf the results of ten AI products are to be combined at the ecosystem level, the product coefficient: αa can be used. Condition:\n\nΣ_{a=1}^A α_a = 1\n\nProduct coefficient:\n\n- the product's own score,\n\n- the provider's reputation,\n\n- the preference of the audited entity,\n\n- commercial sponsorship\n\ncannot be determined through.\n\n## 39. EQUAL PRODUCT COEFFICIENT\n\nEach AI product receives equal weight:\n\nα_a^EQ = 1/A\n\nEqual product ecosystem score:\n\nNOMOS_ecosystem^EQ = Σ_a α_a^EQ NOMOS_a\n\nAnswers the following question:\n\n> How strong is the average tested AI product?\n\nIt does not indicate average human exposure.\n\n## 40. EXPOSURE-WEIGHTED COEFFICIENT\n\nLet Ma be the verified active or eligible user exposure of the AI product during the defined period.\n\nα_a^EXP = M_a / Σ_b M_b\n\nExposure-weighted ecosystem score:\n\nNOMOS_ecosystem^EXP = Σ_a α_a^EXP NOMOS_a\n\nAnswers the following question:\n\n> What is the state of representation in the AI ecosystem that the average person is exposed to?\n\n## 41. IF EXPOSURE DATA IS NOT AVAILABLE\n\nActive user or query exposure:\n\n- closed to the public,\n\n- inconsistent,\n\n- provider self-declaration,\n\n- someone who confuses the number of accounts with active users\n\nmay be of such quality. In this case:\n\n- can be published as an equivalent product result,\n\n- the exposure-weighted outcome may remain NOT ESTIMABLE,\n\nUpper–lower limits can be shown based on reasonable coefficient scenarios. Arbitrary market share estimates cannot be generated.\n\n## 42. PRODUCT COEFFICIENT LOCK\n\nEach αa:\n\n- data source,\n\n- period,\n\n- user unit,\n\n- country coverage,\n\n- update date,\n\n- uncertainty\n\nmust carry. Coefficients must be locked before AI scores are seen. The coefficient of a low-scoring product cannot be reduced after the result.\n\n## 43. ECOSYSTEM SCORE CANNOT COMPENSATE FOR CRITICAL\n\nA product:\n\n- 980 points,\n\n- can carry zero Critical\n\nAnother product:\n\n- 900 points,\n\n- can carry Confirmed Critical\n\nThe ecosystem average may be high. The public card should still show:\n\n- Product-based Critical incident\n\n- Critical exposure weight\n\n- Whether at least one product has an open Critical\n\n- Scope of the Critical\n\nEcosystem composite: cannot delete Critical gate at the product level.\n\n## 44. ECOSYSTEM CRITICAL EXPOSURE\n\nCritical user rate of product a: let it be CRa. Exposure-weighted Critical rate:\n\nCR_ecosystem^EXP = Σ_a α_a^EXP CR_a\n\ncan be calculated as such. Additionally, the binary field:\n\nAnyCritical = 1(∃a: CIS_a ≥ 2)\n\nshould also be shown. The rate of Critical events in a low-exposure product may be small. The AnyCritical record still preserves the presence of the event.\n\n## 45. PROMPT FAMILY COEFFICIENT\n\nβf can be used for the prompt family f. Condition:\n\nΣ_f β_f = 1\n\nβf:\n\n- cannot depend on whether the prompt produces a high or low score,\n\n- the company's preference,\n\n- the ease of post-measurement decision\n\nPrompt family weights may be based on the following areas:\n\n- User decision importance\n\n- Risk level\n\n- Natural usage frequency of the prompt\n\n- Entity type\n\n- Mandatory core of the standard\n\n## 46. PROMPT FAMILIES SHOULD BE SEPARATELY VISIBLE\n\nEven if the composite prompt score can be calculated, the following results should remain separate:\n\n- Core Mirror\n\n- Existence Resolution\n\n- Activity & Scope\n\n- Evidence\n\n- Local\n\n- Temporal\n\n- Recommendation\n\n- Comparative\n\n- Boundary\n\n- Control\n\n- Holdout\n\nA high Core result cannot make a low Boundary or Recommendation result invisible.\n\n## 47. CONTROLLED AND NATURAL PANEL SCORES\n\nControlled and natural panels are different prediction objects. The default NOMOS rule:\n\n> The two panels carry separate score identities and are not combined arithmetically.\n\nExample:\n\n### NOMOS-CLEAN: 962\n\n### NOMOS-NATURAL: 918\n\nIf a policy or research purpose will combine them:\n\nγ_C + γ_N = 1\n\npredeclared coefficients can be used. But:\n\n- two raw scores,\n\n- The difference between them is that\n\n- Coefficient used\n\nshould be shown separately.\n\n## 48. WAVE SCORE AND TIME SERIES\n\nEach Principal Wave has its own score ID. W1 score, W2 score, W3 score are kept separate. Current score: The last valid Principal or Revision can be Retest Wave. The average of past waves should not automatically determine the current result.\n\n## 49. ROLLING SCORE\n\nIf desired, a fixed time moving score can be generated:\n\nNOMOS_rolling = Σ_w δ_wNOMOS_w\n\nHere:\n\nΣ_w δ_w = 1\n\nis. δw:\n\n- before the result,\n\n- according to the time weight\n\nmust be determined. Rolling score:\n\n- the current score,\n\n- do not replace the historical incident record\n\nNeither record may be silently displaced.\n\n## 50. THREE SCORES FOR UNRESOLVED CLAIMS\n\nUnresolved can artificially increase or decrease the point estimate of material atoms. Therefore, three results are suggested.\n\n### 50.1. Conservative Lower Bound\n\nUnresolved claims are considered zero. Slower\n\n### 50.2. Resolved-Only Point Estimate\n\nOnly resolved claims are used. Spoint\n\n### 50.3. Optimistic Upper Bound\n\nUnresolved claims are accepted with full support. Supper Therefore:\n\nS_lower ≤ S_point ≤ S_upper\n\nmust be the case. This range:\n\n> is the reference or epistemic uncertainty range.\n\nIt is not the same as a sample confidence interval.\n\n## 51. SAMPLE UNCERTAINTY\n\nSince population estimates come from the sample:\n\n- response distribution,\n\n- component score,\n\n- composite score\n\nmust carry a confidence interval. Method in complex samples:\n\n- layers,\n\n- the sets,\n\n- their weights,\n\n- paired users,\n\n- waves\n\nshould be protected. Candidate methods:\n\n- Design-based variance estimation\n\n- Replicate weights\n\n- Layered bootstrap\n\n- Cluster bootstrap\n\n- Appropriate rare event model\n\n- Model-assisted survey estimation\n\nThe method used must be specified in the public record.\n\n## 52. COMPOSITE SCORE CONFIDENCE INTERVAL\n\nThe confidence interval of the composite score cannot be obtained by simply summing the component intervals. The components:\n\n- the same users,\n\n- the same claims,\n\n- the same languages,\n\n- the same waves\n\nare dependent on this reason. Preferred approach: Create resamples according to the sample design. In each resample, recalculate all response and component metrics. Reproduce the composite score. Generate the confidence interval from the distribution. This process should rerun the entire score architecture.\n\n## 53. EFFECTIVE SAMPLE SIZE\n\nRaw observation: n and effective sample:\n\nn_eff = (Σ_i w_i)^2 / Σ_i w_i^2 [K08]\n\nshould be published together. 1,000 raw observations may not mean 1,000 equal information units. Component and subgroup scores should carry their own effective sample sizes.\n\n## 54. ZERO OBSERVED CRITICAL ERROR\n\nHaving no Confirmed Critical events in a sample:\n\n#### Observed Confirmed Critical = 0\n\nshould be written. The following expression should not be used:\n\n> The critical risk is zero.\n\nUnder the assumption of independent and equal observations, a quick approximate upper limit:\n\np_upper,95 ≈ 3/n [K09]\n\ncan be estimated with the approach known. In complex GEO-1000 design:\n\n- Instead of raw n, a method appropriate to the design should be used,\n\n- if necessary, an effective sample,\n\n- an exact or bootstrap-based one-sided limit\n\nshould be used. Example:\n\nneff=1,000\n\nand the approximate upper limit under zero events:\n\n3/1,000 = 0.003 = 0.3%\n\nmay be. Correct public statement:\n\n> In 1,000 effective observations, zero Confirmed Critical events were observed; in the approximate method used, the 95 per cent one-sided upper limit is approximately 0.3 per cent.\n\n## 55. UNCERTAINTY WITH ONE CRITICAL EVENT\n\nWhen one Critical event is observed: the point estimate may be low, the confidence interval may be wide. Correct result:\n\n- the event occurred,\n\n- estimated prevalence,\n\n- broad uncertainty,\n\n- incident state\n\nshould be shown together. The following two sentences may also be wrong:\n\n- “Only one event is insignificant.”\n\n- “An event proves that the entire system has failed 100%.”\n\n## 56. SCORE SENSITIVITY\n\nThe following alternatives can also be calculated:\n\n- Raw unweighted score\n\n- Main population weighted score\n\n- Untrimmed weight score\n\n- Trimmed weight score\n\n- Resolved-only score\n\n- Conservative score\n\n- Equal-product score\n\n- Exposure-weighted score\n\n- Native Reach score\n\n- Common Support score\n\n- Clean score\n\n- Natural score\n\nIf there is a material difference:\n\n### SCORE SENSITIVITY WARNING\n\nshould be given.\n\n## 57. INCORRECT PRECISION\n\nThe following representation may be overly precise for most GEO-1000 results: 963,742 / 1000 Point estimate:\n\n- whole number,\n\n- a decimal if necessary\n\nshould be published. The confidence interval should indicate the limit of numerical precision.\n\n## 58. MISSING COMPONENT\n\nIf one of the ten components cannot be calculated:\n\n- its weight cannot be distributed to other components,\n\n- the missing field cannot be counted as 100 points,\n\nthe full NOMOS score cannot be published. The correct status:\n\n### PARTIAL NOMOS PROFILE\n\n### COMPONENT NOT ESTIMABLE\n\n### PROVISIONAL SCORE\n\nmay apply. Example: if a natural panel for USI is not available, only a partial profile with eight or nine components can be presented.\n\nUSI=100 cannot be assumed.\n\n## 59. SCORE PUBLICATION STATUS\n\n### NSQ-0 — NOT COMPUTED\n\nRequired data is not available.\n\n### NSQ-1 — PROVISIONAL\n\nSingle wave, limited adjudication, or missing components exist.\n\n### NSQ-2 — PARTIAL PROFILE\n\nSome components have been reliably calculated. A full 1,000-point comparison cannot be made.\n\n### NSQ-3 — VALID WAVE SCORE\n\nAll core components and quality prerequisites have been met.\n\n### NSQ-4 — REPLICATED SCORE\n\nThe score has been repeated in at least one independent or confirmatory wave.\n\n### NSQ-5 — EXTERNALLY AUDITED SCORE\n\nThe calculation and evidence chain have been reproduced by an independent team. 950+ candidacy at least:\n\n### NSQ-4\n\nshould target that level.\n\n## 60. SCORE VERSIONING\n\nEach score version must include the following:\n\n- Calculation code version\n\n- Component formulas\n\n- Dimension value mappings\n\n- Prompt family weights\n\n- Product coefficients\n\n- Sample weights\n\n- Truth Pack version\n\n- Adjudicator version\n\n- Wave ID\n\n- Integrity hash\n\nIf the formula changes: the old score cannot be rewritten, a new score version is created.\n\n## 61. NEW SCORE FORMULA FOR THE SAME DATA\n\nThe scoring methodology may evolve. The same raw data:\n\n- Score Method 0.9\n\n- Score Method 1.0\n\ncan be recalculated with. Correct record:\n\nThe two results are separate versions. The new method cannot delete the old method from the past.\n\n## 62. DOORS AGAINST SCORE MANIPULATION\n\nThe NOMOS score architecture should have explicit protection against the following methods:\n\n- Selecting only high-scoring AI products\n\n- Excluding low-scoring language or country from the scope\n\n- Turning Conditional Pass into Full Pass\n\n- Counting Unresolved as positive\n\n- Silently removing Not Ratable from the denominator\n\n- Changing the weight of the prompt family after the result\n\n- Reducing the coefficient of the large product according to the score\n\n- Hiding the product with a critical event in the ecosystem average\n\n- Publishing only the Clean score\n\n- Selecting the best wave as the current score\n\n- Hiding the difference between unweighted and weighted scores\n\n- Distributing the weight of missing components to strong components\n\n- Not providing a confidence interval\n\n- Passing the 950 threshold based on point estimate and hiding uncertainty\n\n## 63. PUBLIC NOMOS RESULT CARD\n\nThe candidate public result card should carry the following order:\n\n#### A. Identity\n\nAudited entity AI product Native Reach / Common Support Controlled / Natural Prompt set Wave Truth Pack version Score method\n\n#### B. Gate Status\n\nFull Conformity Candidate Conditional Major Fail Critical Hold Not Ratable\n\n#### C. GEO-1000 Response Distribution\n\nRP-1–RP-8 user equivalents Strict Pass Acceptable Pass Critical Major Unresolved No Usable Response Not Ratable\n\n#### D. Ten Component Profile\n\n### ENT\n\n### FAC\n\n### SBI\n\n### TLA\n\n### ECI\n\n### EPI\n\n### TCR\n\n### LGF\n\n### USI\n\n### STR\n\n#### E. Composite Result\n\n### NOMOS / 1,000\n\nSampling confidence interval Reference lower/upper bounds Effective sample\n\n#### F. Risks and Events\n\nObserved Confirmed Critical events; Critical upper bound; Major findings; open incidents; and correction status.\n\n#### G. Scope and Quality\n\n### SQ\n\n### PSI\n\n### NCL\n\n### TPR\n\n### AQ\n\n### NSQ\n\nCountry/language coverage Comparison level\n\n## 64. SYNTHETIC RESPONSE DISTRIBUTION\n\nSYNTHETIC METHODOLOGY DEMONSTRATION / The following results are not actual AI products or company performance. For Synthetic Orion AI:\n\nStrict Pass:\n\n730+170=900\n\nAcceptable Pass:\n\n900+70=970\n\nCritical: 2/1,000 Major: 20/1,000 The acceptable pass for this product is 97 per cent. However, there are evident Critical and Major findings. Correct status:\n\n### CRITICAL HOLD\n\nIncorrect status:\n\n#### 97 per cent successful\n\n## 65. SYNTHETIC COMPOUND PROFILE\n\nCorrect public display:\n\n### CRITICAL HOLD — 907/1,000\n\nThis system:\n\n- on average strong areas,\n\n- weak language/country fairness,\n\n- material evidence issues,\n\n- open Critical event\n\ncarries.\n\n## 66. SYNTHETIC POST-CORRECTION RESULT\n\nIn the new wave:\n\nStrict Pass: 958 Components:\n\nCorrect record: CURRENT: NOMOS 950+ PERFORMANCE CANDIDATE — 961/1,000 / Historical: Previous Confirmed Critical Incident Remediated Candidate / Observed Critical in current retest: 0 / Critical-rate upper bound: also shown Old Critical incident is not deleted.\n\n## 67. SYNTHETIC AI PRODUCT COEFFICIENT CASE\n\nFour synthetic AI products:\n\nEqual product score:\n\n0.25(960+940+900+980)=945\n\nExposure-weighted score:\n\n0.55(960)+0.25(940)+0.15(900)+0.05(980)=947\n\nThe two results are close. However, they can be very different in another distribution. Correct interpretation:\n\n- Equal-system ecosystem score: 945\n\n- Exposure-weighted ecosystem score: 947\n\n- Product-level gate and incident records separately\n\nIncorrect comment: “NOMOS ecosystem score is definitely 947.”\n\n## 68. SYNTHETIC CRITICAL COEFFICIENT CASE\n\nLet the score of product A4 be 980. However, let there be a Confirmed Critical event in A4. The exposure coefficient is only 0.05. The ecosystem composite may remain high. Correct public card: Exposure-weighted composite: 947 / Any Confirmed Critical Product: YES / Critical exposure share: calculated ratio / A4 coverage: Critical Hold A4’s small coefficient cannot erase the Critical event.\n\n## 69. SCORE PUBLICATION STATUS DECISION\n\nCandidate decision order:\n\nPublicationStatus = NOT RATABLE (basic measurement missing); CRITICAL HOLD (open Confirmed Critical); MAJOR FAIL (open Confirmed Major); PROVISIONAL (component, repetition, or quality missing); CONDITIONAL (Moderate or threshold of unsolved); 950+ CANDIDATE (all candidate requirements); otherwise VALID\n\nThe score band is shown after this status decision.\n\n## 70. MANDATORY NORMATIVE PROVISIONS\n\n**CH16-N01**\n\nThe NOMOS result cannot be published as a single bare compound score.\n\n**CH16-N02**\n\nGate status must be shown before the compound score.\n\n**CH16-N03**\n\nThe GEO-1000 response distribution is a mandatory part of the main public result.\n\n**CH16-N04**\n\nThe total of RP-1 and RP-8 user equivalents must be 1,000.\n\n**CH16-N05**\n\nStrict Pass, Acceptable Pass, Major, Critical, Unresolved, No Usable Response, and Not Ratable should be shown separately.\n\n**CH16-N06**\n\nUnresolved and Not Ratable observations cannot be silently added to pass or fail categories.\n\n**CH16-N07**\n\nNot Ratable observations cannot be interpreted as a product performance failure.\n\n**CH16-N08**\n\nNot Ratable observations cannot be silently removed from the denominator; they should be shown in ratable coverage.\n\n**CH16-N09**\n\nObservation weights should be independent of the AI response result.\n\n**CH16-N10**\n\nThe raw number of participants and the population-weighted user equivalent should be reported separately.\n\n**CH16-N11**\n\nThe response pass status should be kept separate from the atomic claim support ratio.\n\n**CH16-N12**\n\nCritical and Major gates cannot be compensated by any composite score.\n\n**CH16-N13**\n\nAtomic centrality weights should be defined before seeing the results.\n\n**CH16-N14**\n\nA repeat of the same root claim cannot artificially inflate or deflate the score.\n\n**CH16-N15**\n\nClaim centrality weights cannot replace Critical and Major gates.\n\n**CH16-N16**\n\nAtomic Fidelity dimension values must be versioned and publicly available.\n\n**CH16-N17**\n\nIf the Atomic Fidelity mapping changes, a new score method version must be created.\n\n**CH16-N18**\n\nReference gap atoms cannot be silently added to the point score as zero or one.\n\n**CH16-N19**\n\nRequired Response Element Completeness must be published as a separate metric.\n\n**CH16-N20**\n\nCore required-element failure cannot be compensated with a high REC average.\n\n**CH16-N21**\n\nThe NOMOS profile with ten components must be visible before the composite index.\n\n**CH16-N22**\n\nENT, FAC, SBI, TLA, ECI, EPI, TCR, LGF, USI, and STR must carry separate scores.\n\n**CH16-N23**\n\nOne component cannot silently take over the missing weight of another component.\n\n**CH16-N24**\n\nA component that cannot be calculated cannot be assumed as 100.\n\n**CH16-N25**\n\nThe full composite NOMOS index should only be published if all ten components have been reliably calculated.\n\n**CH16-N26**\n\nIf equal component weighting is used as the canonical candidate method, it must be explicitly stated.\n\n**CH16-N27**\n\nDomain-specific alternative weights must have a separate score name and version.\n\n**CH16-N28**\n\nA domain-specific score cannot be used in place of the canonical index without explanation.\n\n**CH16-N29**\n\nLGF cannot rely solely on the population mean; it must include a performance base and disparity.\n\n**CH16-N30**\n\nA single Country Observer observation cannot be converted into a country fairness score.\n\n**CH16-N31**\n\nOnly language and geographic cells with sufficient sample size should participate in fairness estimation.\n\n**CH16-N32**\n\nLanguages and countries that have not been tested cannot be ignored to produce a high fairness result.\n\n**CH16-N33**\n\nUSI should evaluate controlled and natural panel results and material surface difference together.\n\n**CH16-N34**\n\nIn a study with only a controlled or only a natural panel, USI cannot be considered fully calculated.\n\n**CH16-N35**\n\nControlled and natural panel scores should be kept separate by default.\n\n**CH16-N36**\n\nIf controlled and natural scores are combined, coefficients must be locked before the results.\n\n**CH16-N37**\n\nSTR cannot be presented as a full stability score in a single wave.\n\n**CH16-N38**\n\nInter-wave product or Truth Pack changes must be explicitly modelled in the stability calculation.\n\n**CH16-N39**\n\nCompound NOMOS cannot change the Critical or Major gate status.\n\n**CH16-N40**\n\nA score of 950 alone cannot establish NOMOS 950+ candidacy.\n\n**CH16-N41**\n\nNOMOS 950+ candidate conditions must collectively meet the requirements for score, Strict Pass, gate, component floor, unresolved, quality, and repeat conditions.\n\n**CH16-N42**\n\n950+ thresholds cannot be presented as a final scientific standard before pilot and independent review.\n\n**CH16-N43**\n\nThe score identity must carry AI product, frame, panel, prompt set, wave, Truth Pack, and score version.\n\n**CH16-N44**\n\nThe sentence “The entity's NOMOS score is X” cannot be used without a scope identity.\n\n**CH16-N45**\n\nThe Core GEO-1000 result cannot be presented as the Full NOMOS score.\n\n**CH16-N46**\n\nNative Reach and Common Support scores must be reported separately.\n\n**CH16-N47**\n\nProduct scores for different Native Reach populations cannot be directly ranked without explanation.\n\n**CH16-N48**\n\nThe Common Support result cannot be presented as the product's full global access score.\n\n**CH16-N49**\n\nAI product coefficients must be determined independently of the results.\n\n**CH16-N50**\n\nAn AI product coefficient cannot depend on the product score, brand reputation, or commercial sponsorship relationship.\n\n**CH16-N51**\n\nEqual product and exposure-weighted ecosystem scores must be named separately.\n\n**CH16-N52**\n\nIf active user or exposure data cannot be verified, arbitrary coefficients cannot be generated.\n\n**CH16-N53**\n\nExposure data must be transparent at the account, user, query, or session level.\n\n**CH16-N54**\n\nThe coefficient of a low-scoring product cannot be reduced after the result is seen.\n\n**CH16-N55**\n\nEcosystem composite cannot delete Critical or Major gate at the product level.\n\n**CH16-N56**\n\nThe AnyCritical and exposure-weighted Critical ratio should be shown separately.\n\n**CH16-N57**\n\nPrompt family coefficients must be determined before the results.\n\n**CH16-N58**\n\nThe weight of a low-scoring prompt family cannot be reduced later.\n\n**CH16-N59**\n\nThe separate result of each prompt family should remain visible alongside the composite prompt score.\n\n**CH16-N60**\n\nThe current wave score should be kept separate from rolling or historical scores.\n\n**CH16-N61**\n\nThe past wave with the highest score cannot be selected as the current score.\n\n**CH16-N62**\n\nConservative, resolved-only, and optimistic results can be produced for the Unresolved claim mass.\n\n**CH16-N63**\n\nReference uncertainty and sample confidence interval cannot be presented as the same thing.\n\n**CH16-N64**\n\nComplex sample confidence interval should preserve stratification, clustering, weighting, and matched user structure.\n\n**CH16-N65**\n\nThe composite score confidence interval must be calculated by rerunning the entire scoring process.\n\n**CH16-N66**\n\nThe raw and effective sample sizes should be shown together.\n\n**CH16-N67**\n\nA Critical event observed as zero cannot be expressed as zero Critical risk.\n\n**CH16-N68**\n\nIn the case of zero events, a one-sided upper limit or the equivalent rare event uncertainty should be published.\n\n**CH16-N69**\n\nThe 3/n approach cannot be used automatically as the final method in complex design; it can only be approximate control.\n\n**CH16-N70**\n\nIn a critical event, the presence of the event and the uncertainty of its frequency should be shown together.\n\n**CH16-N71**\n\nScore sensitivity analyses should show the unweighted, weighted, trimmed, unresolved, and alternative coefficient results to the extent relevant.\n\n**CH16-N72**\n\nMaterial score sensitivity should create an alert on the public card.\n\n**CH16-N73**\n\nThe score's fake decimal cannot be published at all.\n\n**CH16-N74**\n\nThe weight of the missing component cannot be distributed to other components.\n\n**CH16-N75**\n\nThe Puan Publication Quality status should be visible in the public record.\n\n**CH16-N76**\n\n950+ candidacy should aim for at least the replicated score level.\n\n**CH16-N77**\n\nWhen the score formula changes, past scores cannot be silently rewritten.\n\n**CH16-N78**\n\nIf different score method versions are applied to the same data set, all results must be preserved in separate versions.\n\n**CH16-N79**\n\nScore calculation code, coefficients, and integrity records must be published in a machine-readable format.\n\n**CH16-N80**\n\nEvery score, coefficient, confidence interval, and publication decision must have a responsible accountable human or institution.\n\n## 71. FORMS OF FAILURE\n\n**CH16-F01 — SINGLE NUMBER REPORT**\n\nDistribution, gate, and component profile are preserved.\n\n**CH16-F02 — HIDING THE GATE UNDER THE SCORE**\n\nCritical Hold becomes a small footnote.\n\n**CH16-F03 — COUNTING CONDITIONAL AS FULL PASS**\n\nModerate issues become invisible.\n\n**CH16-F04 — COUNTING UNRESOLVED AS PASS**\n\nUnknown results raise the score.\n\n**CH16-F05 — COUNTING UNRESOLVED AS FAIL**\n\nLack of reference lowers the product.\n\n**CH16-F06 — REMOVING NOT RATABLE FROM THE DENOMINATOR**\n\nMeasurement defects become invisible.\n\n**CH16-F07 — COUNTING NOT RATABLE AS PRODUCT DEFECT**\n\nInspection defect is assigned to AI.\n\n**CH16-F08 — COUNT 1,000 PEOPLE EQUALLY WEIGHTED**\n\nOversampling and population differences are ignored.\n\n**CH16-F09 — WEIGHT ACCORDING TO RESULT**\n\nCorrect answers receive greater weight.\n\n**CH16-F10 — COUNT CLAIM SUPPORT RATE AS ANSWER PASS**\n\nCritical gate gets lost in the average.\n\n**CH16-F11 — COUNT CORE AND INCIDENTAL EQUALLY**\n\nThe incidental claim gets as much weight as the main entity.\n\n**CH16-F12 — INFLATE SCORE BY REPEATING**\n\nThe same correct claim is written ten times to increase the score.\n\n**CH16-F13 — COLLAPSE SCORE BY REPEATING**\n\nThe same wrong root results in ten separate penalties.\n\n**CH16-F14 — STORING DIMENSION VALUES**\n\nThe score cannot be reproduced.\n\n**CH16-F15 — COUNTING THE CLAIM THAT COMES FROM BELOW AS FACT FOR F-4**\n\nAccurate transmission of self-report turns into the accuracy of the self-report.\n\n**CH16-F16 — MELTING REQUIRED ELEMENTS INTO ATOMIC AVERAGE**\n\nEven if the main task is missing, the answer gets a high score.\n\n**CH16-F17 — COMPONENT CHERRY-PICKING**\n\nOnly the three strong components are published.\n\n**CH16-F18 — COUNTING MISSING COMPONENT AS 100**\n\nFull score is artificially increased.\n\n**CH16-F19 — DISTRIBUTING MISSING WEIGHT**\n\nThe uncalculated fairness score is added to the factual score.\n\n**CH16-F20 — COUNTING THE LAW OF NATURE AS EQUAL WEIGHT**\n\nSensitivity and external review are not performed.\n\n**CH16-F21 — COUNTING THE DOMAIN SCORE AS CANONICAL**\n\nSpecial weight is presented like the base NOMOS score.\n\n**CH16-F22 — COUNTING LANGUAGE AVERAGE AS LANGUAGE FAIRNESS**\n\nLow language floor is hidden.\n\n**CH16-F23 — COUNTING EQUAL LANGUAGE PARITY AS QUALITY**\n\nLow scores in all languages produce high fairness.\n\n**CH16-F24 — COUNTING UNTESTED LANGUAGES AS COVERAGE**\n\nCoverage factor is inflated.\n\n**CH16-F25 — COUNTRY SCORE FROM OBSERVER**\n\nA single user enters the fairness account.\n\n**CH16-F26 — PUBLISH ONLY CLEAN SCORE**\n\nNatural user experience is preserved.\n\n**CH16-F27 — COUNT ONLY NATURAL SCORE AS PRODUCT DEFAULT**\n\nPersonalisation becomes the basic behaviour.\n\n**CH16-F28 — SILENTLY AVERAGE CLEAN AND NATURAL**\n\nDifferent prediction objects get mixed.\n\n**CH16-F29 — 100 STABILITY IN A SINGLE WAVE**\n\nStability is considered complete without repetition.\n\n**CH16-F30 — COUNT OLD GOOD WAVE AS CURRENT**\n\nCurrent weakening is hidden.\n\n**CH16-F31 — AUTOMATICALLY COUNT 950 AS COMPLIANCE**\n\nGate and component floor are ignored.\n\n**CH16-F32 — CRITICAL HOLD — COUNT 970 AS 950+**\n\nThe count goes ahead of the gate.\n\n**CH16-F33 — HIDE SCORE IDENTITY**\n\nIt is not known which product and panel are measured.\n\n**CH16-F34 — COUNT CORE SCORE AS FULL NOMOS**\n\nA single prompt represents the whole standard.\n\n**CH16-F35 — RANK NATIVE REACH SCORES**\n\nDifferent populations are considered the same.\n\n**CH16-F36 — COUNT FULL ACCESS TO COMMON SUPPORT**\n\nNarrow common population is made a global outcome.\n\n**CH16-F37 — GIVE MODEL COEFFICIENT ACCORDING TO SCORE**\n\nHigh score gains more influence.\n\n**CH16-F38 — MAKING BRAND REPUTATION A FACTOR**\n\nProduct popularity is weighted without being predicted.\n\n**CH16-F39 — COUNTING ACCOUNT NUMBERS AS ACTIVE USERS**\n\nExposure coefficient inflates.\n\n**CH16-F40 — REMOVING LOW-SCORED PRODUCT**\n\nThe ecosystem consists only of successful products.\n\n**CH16-F41 — COUNTING EQUAL EXPOSURE SCORE FOR PRODUCT**\n\nAverage product is presented like an average person.\n\n**CH16-F42 — COUNTING EXPOSURE SCORE AS PRODUCT QUALITY**\n\nA large product automatically becomes a more important quality metric.\n\n**CH16-F43 — DELETE CRITICAL IN ECOSYSTEM AVERAGE**\n\nThe heavy event of the small-coefficient product disappears.\n\n**CH16-F44 — CHANGE THE PROMPT COEFFICIENT AFTER THE RESULT**\n\nThe hard prompt family is reduced.\n\n**CH16-F45 — CONSIDER LOW-SCORE FAMILY AS OPTIONAL**\n\nBoundary or evidence result is excluded from scope.\n\n**CH16-F46 — KEEP CURRENT WITH ROLLING SCORE**\n\nOld high waves cover the current drop.\n\n**CH16-F47 — DELETE UNRESOLVED FROM THE DENOMINATOR**\n\nPoint estimation artificially rises.\n\n**CH16-F48 — CONSIDER REFERENCE UNCERTAINTY IN SAMPLING CI**\n\nTwo different uncertainties mix.\n\n**CH16-F49 — SIMPLE RANDOM-SAMPLE CI**\n\nStratification, weight, and clusters are ignored.\n\n**CH16-F50 — SUMMING COMPONENT CIs**\n\nDependency is ignored.\n\n**CH16-F51 — CLAIMING SENSITIVITY WITH RAW N**\n\nEffective sample is concealed.\n\n**CH16-F52 — COUNTING ZERO EVENTS AS ZERO RISK**\n\nRare critical error uncertainty is hidden.\n\n**CH16-F53 — APPLYING 3/N TO EVERY CASE**\n\nComplex sample and weight structure is ignored.\n\n**CH16-F54 — COUNTING A SINGLE CRITICAL AS 100 PER CENT SYSTEM RISK**\n\nCommonality is overly generalised.\n\n**CH16-F55 — HIDING A SINGLE CRITICAL BECAUSE THE RATE IS SMALL**\n\nThe existence of the event disappears.\n\n**CH16-F56 — HIDING SENSITIVITY ANALYSIS**\n\nIt is not seen that the score changes in alternative reasonable methods.\n\n**CH16-F57 — FAKE DECIMAL SENSITIVITY**\n\nA result like 963,742 is published.\n\n**CH16-F58 — COUNTING PARTIAL PROFILE AS FULL SCORE**\n\nComparison is made with a missing component.\n\n**CH16-F59 — PUBLISHING THE PROVISIONAL SCORE AS FINAL**\n\nA single wave and a single adjudicator become full standard.\n\n**CH16-F60 — REWRITING HISTORY IN FORMULA CHANGE**\n\nThe old score disappears.\n\n**CH16-F61 — KEEP CALCULATION CODE CLOSED**\n\nIndependent reproduction is prevented.\n\n**CH16-F62 — CONSIDER SCORE HIGHNESS AS AUDIT QUALITY**\n\nA faulty measurement is accepted because it gave a high result.\n\n**CH16-F63 — CONSIDER SCORE LOWNESS AS METHOD DEFECT**\n\nA low result is recalculated due to commercial inconvenience.\n\n**CH16-F64 — SINGLE-COMPANY SCORE MONOPOLY**\n\nIndependent institutions are prevented from applying the same formula.\n\n## 72. AUDIT PROCEDURE\n\n### Step 1 — Create Score Identity\n\nEntity AI product Frame Panel Prompt set Wave Truth Pack Score method is locked.\n\n### Step 2 — Lock the Main Analysis Observation Set\n\nBackup, withdrawn, not-ratable, and main observation records are separated.\n\n### Step 3 — Verify Population Weights\n\nDesign, non-response, and calibration weights are examined.\n\n### Step 4 — Calculate RP Distribution\n\nUser equivalents with RP-1 to RP-8 are produced.\n\n### Step 5 — Calculate Strict and Acceptable Pass\n\nSP1000 and AP1000 are derived.\n\n### Step 6 — Calculate the Densities of Critical, Major, Unresolved, and Not-Ratable\n\nGate and prevalence are kept separate.\n\n### Step 7 — Apply Claim Centrality and Root Cluster Normalisation\n\nClaims are checked again.\n\n### Step 8 — Calculate Atomic Size Metrics\n\nEntity, fact, scope, time, attribution, modality, and reference results are produced.\n\n### Step 9 — Calculate Required Element Completeness\n\nIt is recorded together with the prompt family gates.\n\n### Step 10 — Calculate Ten Components\n\nAll component formulas from ENT to STR are applied.\n\n### Step 11 — Check for Missing Components\n\nIt is determined whether the full score can be calculated.\n\n### Step 12 — Calculate the Composite NOMOS Index\n\nApplies only to complete and valid profiles.\n\n### Step 13 — Assign Gate Status\n\nDetermines whether the status is Critical, Major, Conditional, or Full Candidate.\n\n### Step 14 — Separate Native Reach and Common Support Results\n\nThe two score families are kept separate.\n\n### Step 15 — Lock AI Product Coefficients\n\nEqual and exposure coefficients are saved along with the data source.\n\n### Step 16 — Apply Coefficients of the prompt Family\n\nWeights are used independently of the result.\n\n### Step 17 — Separate Panel and Wave Results\n\nClean, Natural, Current and Rolling scores are separated.\n\n### Step 18 — Calculate Fairness and Floor Metrics\n\nLanguage, country, and coverage results are produced.\n\n### Step 19 — Calculate Unresolved Lower–Point–Upper Results\n\nReference uncertainty is made visible.\n\n### Step 20 — Calculate Sampling Confidence Intervals\n\nResampling method suitable for sample design is applied.\n\n### Step 21 — Calculate Effective Sample Size\n\nDisplayed on total and component basis.\n\n### Step 22 — Calculate Rare Critical Event Threshold\n\nA one-sided method is used for zero or few events.\n\n### Step 23 — Generate Sensitivity Analyses\n\nWeight, coefficient, and unresolved options are compared.\n\n### Step 24 — Assign Publication Quality Status\n\nA decision is made between NSQ-0 and NSQ-5.\n\n### Step 25 — Apply NOMOS 950+ Candidate Doors\n\nAll conditions except for the score are checked.\n\n### Step 26 — Create the Public Results Card\n\nGate, distribution, profile, index, uncertainty, and scope are published in order, respectively.\n\n### Step 27 — Lock the Computation Code and Manifest\n\nThe code version, inputs, hash, and results are linked.\n\n### Step 28 — Prepare the Independent Reproduction Package\n\nComputation files stripped of personal data are made available for independent review.\n\n## 73. REQUIRED EVIDENCE\n\nScore ID Entity record AI System Register Native Reach and Common Support records Panel status Respondent set and families Truth Pack version Wave ID Claim Ledgers Response decisions RP-1–RP-8 distribution Main analysis observation set Design weights Nonresponse weights Calibration weights Weight trimming records Effective sample Strict Pass Acceptable Pass Critical rate Major rate Unresolved rate No Usable Response rate Not Ratable rate Claim centrality weights Root cluster records Atomic dimension values Atomic Fidelity Required Element Completeness\n\n### ENT\n\n### FAC\n\n### SBI\n\n### TLA\n\n### ECI\n\n### EPI\n\n### TCR\n\n### LGF\n\n### USI\n\n### STR\n\nComponent sub-inputs; language and country floors; coverage factor; Controlled and Natural scores; inter-wave results; composite NOMOS index; domain-specific alternative scores; AI-product coefficients and their data sources; prompt-family coefficients; equal-product result; exposure-weighted result; AnyCritical status; Critical exposure share; Unresolved lower–point–upper results; sampling confidence interval; Critical upper bound; sensitivity analysis; Score Publication Quality; 950+ candidate-gate record; score-method version; calculation code; code-integrity hash; public scorecard; change log; independent-reproduction record; and accountable person or institution.\n\n## 74. AUDIT CHECKLIST\n\nDoes the score identity cover all scope areas? Was the main analysis observation set locked before the result? Were backup and main observations separated? Are observation weights independent of the result? Is the RP-1–RP-8 distribution a total of 1,000? Were Strict and Acceptable Pass separated? Are Critical and Major separate? Was Unresolved marked pass or fail? Was Not Ratable removed from the denominator? Is Ratable coverage open? Was the claim support rate used instead of the pass response? Are the Centrality weights open? Was the same root claim counted again? Has the Atomic dimension mapping been published? Was F-4 interpreted correctly? Were Required Elements calculated separately? Was the core gate compensated with high REC?\n\nIs the entire component present? Was the missing component counted as 100? Was the weight of the missing component distributed? Does ENT measure the correct entity relationships? Does FAC quietly exclude unresolved claims? Does SBI include material boundary omission? Does TLA preserve country and time difference? Does ECI measure citation support instead of citation presence? Does EPI carry the distinction between self-declaration and reality? Does TCR make the no usable response visible? Does LGF include the base and gap alongside the average? Did Country Observer observation enter the country score? Does USI rely on both Clean and Natural data? Was STR produced from a single wave? Did the composite score cover the gate status? Does 950+ rely solely on the point score?\n\nHave the component floors been checked? Is the Core unresolved limit open? Have Native Reach and Common Support been separated? Were the AI product coefficients determined before the scores were seen? Do the coefficients measure active users, accounts, or queries? Are Equal and exposure results separate? Has the low-scoring product been removed? Does the Ecosystem score hide the product Critical? Is there an AnyCritical field? Are the prompt family weights open? Has the low-scoring prompt family been reduced? Are Clean and Natural scores separate? Are current and Rolling scores separate? Was the best past wave selected as current? Are there lower–point–upper results for Unresolved? Have reference and sampling uncertainty been separated? Does the confidence interval design maintain protection?\n\nDoes the composite CI rerun the entire scoring system? Are the raw and effective sample together? Was it written as zero critical, zero risk? Is the upper limit of a rare event open? Has a sensitivity analysis been done? Do alternative reasonable coefficients change the result? Is there false decimal precision? Is Publication Quality of the score clear? Is the score formula versioned? Is the old scoring method preserved? Can the calculation code be reproduced? Is the accountable owner of the score clear?\n\n## 75. OBJECTIONS AND RESPONSES\n\n### Objection 1 — \"The only thing users understand is a single number; why are we showing so many results?\"\n\nA single number is understandable. But alone, it is misleading. Public card:\n\n- gate status,\n\n- distribution of 1,000 users,\n\n- composite score\n\ncan be presented in three simple layers as follows. The detailed profile is preserved for researchers and auditors.\n\n### Objection 2 — “Wasn't the GEO score directly the number of people who answered correctly?”\n\nThis is still the main result. SP₁₀₀₀ and AP₁₀₀₀ indicate the founding idea directly. The NOMOS composite index:\n\n- evidence,\n\n- justice,\n\n- natural user,\n\n- stability\n\nincludes additional normative areas such as.\n\n### Objection 3 — “Why do we count Full Pass and Conditional Pass separately?”\n\nBecause both users may not have seen severe mistakes. However, in the Conditional group:\n\n- limited up-to-dateness,\n\n- Moderate omission,\n\n- usability issue\n\ncan be found. This difference is important for improvement and confidence level.\n\n### Objection 4 — “Can't we remove Unresolved observations and give a clean score?”\n\nA resolved-only score can be given. However, the unresolved mass should remain visible. Otherwise, it would be possible to raise the score without resolving the hardest claims.\n\n### Objection 5 — “Why are the ten components equally weighted?”\n\nEqual weighting provides a transparent starting point. High-risk errors are already protected by the gate system. Alternative domain weights can also be produced. Exact weights should be tested through pilot and independent review.\n\n### Objection 6 — 'Isn't existence and factual fidelity more important?'\n\nIt is like that in some audits. However:\n\n- evidence,\n\n- scope,\n\n- fairness,\n\n- natural-user difference\n\nIt can generate high factual average false confidence without it. Critical and Major gates also protect the main materiality.\n\n### Objection 7 — “If fairness is low, why should the global average decrease?”\n\nCompound NOMOS standard evaluates not only the average person but also representation fairness. The population-weighted main result is again shown separately. LGF makes systematic deterioration in small languages and countries visible.\n\n### Objection 8 — “Doesn’t equal low outcome in all languages make fairness high?”\n\nIt can be done if only disparity is used. Therefore, LGF:\n\n- performance base,\n\n- uses disparity,\n\n- coverage\n\ntogether.\n\n### Objection 9 — “Why don't we directly average the Clean and Natural scores?”\n\nThey measure different target populations and user situations. Arithmetic combination requires normative preference. By default, they are published separately.\n\n### Objection 10 — ‘Would equal weighting across ten AI products be fairer?’\n\nIt is fair for average product comparison. It is not fair for average human exposure. Therefore, equal and exposure-weighted results are kept separate.\n\n### Objection 11 — “How will the model coefficient be found if there is no active user data?”\n\nIt may not be found. Equal product outcomes and reasonable scenario ranges can be published. Arbitrary coefficients cannot be fabricated.\n\n### Objection 12 — “If only 1% exposure occurs for a product, why would the Critical event affect the ecosystem outcome?”\n\nThe exposure-weighted rate may have a small effect. However, the product and coverage where the event occurs still carry Critical Hold. The ecosystem average cannot erase the product truth.\n\n### Objection 13 — “If we observed zero Critical, why don’t we just say zero?”\n\nThe sample does not observe all users. Zero observation: indicates that the event was not seen. It does not indicate that the risk is mathematically zero.\n\n### Objection 14 — “Doesn’t having so many extra condition points for 950+ make it meaningless?”\n\nOn the contrary, it preserves the meaning of the number. 950 points:\n\n- Critical event,\n\n- language-deficient,\n\n- single-waved,\n\n- low adjudication quality\n\ndoes not carry the same confidence in a study.\n\n### Objection 15 — \"If the score formula changes, can't old studies be compared?\"\n\nOld and new methods can be applied to the same fixed data for bridge analysis. Two score versions are kept separate. This method allows the standard to evolve without erasing the past.\n\n### Objection 16 — \"Doesn't such a transparent formula make the system easier to game?\"\n\nSome optimisations may occur. Against this:\n\n- sealed holdout prompts,\n\n- natural-user panel,\n\n- language and country justice,\n\n- different waves,\n\n- Critical doors,\n\n- open change history\n\nis used. Closed formula increases uninspectability. Open formula alone is not a reason for gamification.\n\n## COMMON PROVISION OF SECTION 79\n\nA score can easily seem impressive. It is three digits. It is round. It looks comparable. An executive can say, “We got 963.” An agency can say, “Your competitor got 912.” A company can say, “We have the world’s highest NOMOS score.” But if the score does not answer these questions, its decision-making value is limited: Which AI product? Which user population? Which countries? Which languages? Which prompts? Which time? Which panel? How many Critical events? How many Major findings? How many unresolved? How many effective samples? What is the confidence interval? What are the model coefficients based on? Has the result been replicated? NOMOS does not accept a single number. It refuses to let a single number stand in for the truth. Therefore the main result:\n\n> Is the distribution of 1,000 user equivalents.\n\nComposite index:\n\n> This is a summary of this distribution and other normative components.\n\nA product:\n\n- 970 acceptable pass,\n\n- two Critical events\n\nAnother product:\n\n- 950 acceptable pass,\n\n- zero Critical,\n\n- stronger language fairness\n\nThose figures may describe another product. The average alone cannot make the first product the leader. A false experience in a lower-resource language may carry little population weight yet remain wholly real for that user. A Critical event in a small AI product may carry little ecosystem weight yet remain wholly real for that product's user. Scoring should not erase what is small; it should place it correctly and weight it honestly. AI-product coefficients must be fixed before performance is observed. Lowering a product's coefficient after a weak result allows the system to rewrite its own reality. Lowering the weight of a weak prompt family allows the test to evade its own questions. Excluding Unresolved claims hides the hardest realities; excluding Not Ratable records turns measurement defects into product success. Zero observed Critical events do not mean zero risk.\n\nA sample says only what it observes. NOMOS also states what it does not observe, along with its limit. For this reason, the number 950+ is not the sole target. It is a contract. To say the following:\n\n- high average,\n\n- high Strict Pass,\n\n- strong justice,\n\n- natural user resilience,\n\n- again,\n\n- clear uncertainty,\n\n- zero open Critical and Major\n\ntogether are required. Therefore, NOMOS's sixteenth measurement law is:\n\n> Distribution is real; score is summary.\n\nIts seventeenth law is:\n\n> Gate status stands above average.\n\nIts eighteenth measurement law states:\n\n> Coefficient cannot follow the result; the result follows the coefficient.\n\nIts nineteenth measurement law states:\n\n> The average product and the average person are not the same thing.\n\nIts twentieth measurement law states:\n\n> Fairness is not just equality; it is the simultaneous preservation of the base, difference, and scope.\n\nIts twenty-first measurement law states:\n\n> Unresolved is not waste to be thrown outside the reality score, it is the boundary of the score.\n\nIts twenty-second measurement law states:\n\n> Zero observed error is not zero probability.\n\nThe twenty-third law is as follows:\n\n> A missing component cannot be filled by the successes of other components.\n\nThe twenty-fourth law is as follows:\n\n> NOMOS 950+ is not a number; the score is the common decree of gate, justice, uncertainty, and repetition conditions.\n\n## NOMOS’s Section 16 Order\n\n> Don't just tell me my score. / Show me who sees what among a thousand people.\n\n> Don't put Full Pass, Conditional Pass, Major Fail, Critical Fail, Unresolved and Not Ratable in the same basket.\n\n> Don't lose my critical error within the average.\n\n> Do not remove Not Ratable records to make me appear more successful. / But do not count measurement defects as my failures either.\n\n> Do not turn atomic accuracy into response-pass status.\n\n> Show entity, factual, scope, time, evidence, epistemic status, task, fairness, user status, and stability separately.\n\n> Do not score a missing component as one hundred. / Do not redistribute its weight to stronger components.\n\n> Do not hide a low language floor within the average of major languages.\n\n> Do not ignore degradation in natural users just because I am good in a clean session.\n\n> Do not declare a single wave as a decisive result.\n\n> Assign a coefficient to each AI product before the result. / Do not change my weight after seeing my score.\n\n> Do not make the equal product average the average human experience.\n\n> Provide active user data, otherwise do not fabricate a coefficient. / Say unknown.\n\n> Do not remove a critical event in my small product from the ecosystem average.\n\n> Do not pass or drop the Unresolved claim. / Show its lower and upper limits.\n\n> Do not consider a thousand raw bin observations as a thousand equal information units. / Write the effective sample.\n\n> Don't say zero risk when you see zero Critical. / Show the upper limit of the risk you haven't seen.\n\n> Don't produce false certainty like 963,742.\n\n> Don't make the number 950 a badge. / First check Strict Pass, component floors, gates, fairness, repetition, and ambiguity.\n\n> If you change the scoring method, don't delete the old score. / Open a new version.\n\nFirst, calculate the user distribution. / Then derive the atomic metrics. / Then produce the ten-component profile. / Then apply the gate status. / Then add the confidence interval and Unresolved bounds. / Then, where justified, calculate the composite score. / Only after all of that may you approach a NOMOS 950+ claim.\n\n## The Chapter's Closing Sentence\n\nA trustworthy GEO-1000 score is not a calculation that collapses reality into one number. It is an open measurement architecture that preserves, together, the distribution of representation experienced by 1,000 people, non-compensatory risks, the fairness floor, uncertainty and stability over time.\n\n## Normative Core\n\n> No NOMOS result may be published as a naked composite score. Every published score MUST first disclose: - the gate status, - the GEO-1000 response-status distribution, - Strict and Acceptable Pass equivalents, - Critical and Major equivalents, - Unresolved, No-Usable-Response, and Not-Ratable equivalents, - component scores, - uncertainty, - effective sample size, - scope, - and score-method version. The primary GEO-1000 output MUST express the population-weighted equivalent number of users, out of 1,000, receiving each RP-1 through RP-8 response status. Critical and Major findings are non-compensatory. No atomic-claim average, response-pass rate, component score, product coefficient, or ecosystem composite may override an active Critical or Major gate. Atomic, required-element, component, composite, fairness, stability, Critical-rate, and publication-quality metrics MUST remain distinct. The canonical candidate NOMOS composite consists of ten separately reported 0–100 components: - Entity Integrity, - Factual Fidelity, - Scope and Boundary Integrity, - Temporal and Local Accuracy, - Evidence and Citation Integrity, - Epistemic Integrity, - Task Completion and Relevance, - Language and Geographic Fairness, - User-State and Surface Integrity, - and Stability and Replicability. A full 0–1,000 composite MUST NOT be produced when a required component is not estimable. Missing component weight MUST NOT be silently redistributed. Native Reach, Common Support, Controlled Clean, Natural User, current wave, rolling, equal-product, and exposure-weighted scores MUST remain separately identified. AI-product coefficients MUST be locked before product results are observed and MUST NOT depend on product performance, provider reputation, client preference, sponsorship, or desired outcome. Equal-product coefficients estimate average tested-product performance. Exposure coefficients estimate average defined-user exposure. These estimands MUST NOT be represented as equivalent. An ecosystem composite MUST retain product-level Critical and Major incident visibility. High scores in other products MUST NOT erase a lower-weighted product's confirmed Critical incident. Unresolved claims MUST produce visible reference-uncertainty bounds and MUST NOT be silently treated as correct, incorrect, or absent. Sampling uncertainty, reference uncertainty, score sensitivity, raw sample size, and effective sample size MUST be separately reported. Zero observed confirmed Critical incidents MUST NOT be represented as zero Critical risk. A method-appropriate one-sided upper bound MUST be reported. A NOMOS 950+ performance candidate MUST satisfy score, Strict Pass, gate, component-floor, unresolved, quality, coverage, and replication conditions. A composite score of 950 or more alone is insufficient. Every score formula, mapping, coefficient, dataset, code version, confidence procedure, publication decision, revision, and integrity record MUST be versioned and attributable to an accountable human or organisation.","character_count":78711,"record_sha256":"5b70cce56f17c2ac8be8a4b3be95d21830fb92e3dae61ca9906310e872cd47d1"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-17","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":19,"chapter_number":17,"item_number":null,"title":"The Apple.com Synthetic Test Design","subtitle":"Testing the standard, not a company","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K03","K09"],"source_word_count":11199,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":17,\"id\":\"NOMOS-GEO-AUDIT-CH17\",\"title\":\"The Apple.com Synthetic Test Design\",\"subtitle\":\"Testing the standard, not a company\",\"sourceFile\":\"17.ci bölüm.docx\",\"sourceSha256\":\"B2ADF095B4040B809F9891B25ED7C27BCF8894AAA2EECFF2DE7C346BE9CE353D\",\"sourceWordCount\":11199,\"sourceIds\":[\"K03\",\"K09\"],\"machine\":{\"chapter\":17,\"chapterId\":\"NOMOS-GEO-AUDIT-CH17\",\"title\":\"The Apple.com Synthetic Test Design\",\"subtitle\":\"Testing the standard, not a company\",\"sourceIds\":[\"K03\",\"K09\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH17-R01\",\"normativeRuleEnglish\":\"The Apple.com Synthetic Benchmark MUST evaluate the NOMOS audit method, not the real-world performance of Apple Inc., apple.com, or any actual AI provider. All benchmark entities, claims, evidence objects, users, countries, languages, AI products, responses, citations, captures, incidents, and scores MUST be clearly identified as synthetic. The benchmark MUST maintain a separate APPLE-SYNTH entity twin and MUST NOT represent its Truth Pack as factual information about the real company. The principal benchmark corpus MUST contain: - ten synthetic AI product profiles, - one thousand unique synthetic participants per product per wave, - three measurement waves, - and thirty thousand principal responses. Synthetic participant identities MUST NOT be reused across products or waves in the principal corpus. Product samples SHOULD use matched population distributions without using the same person unit. The principal prevalence corpus, severity challenge corpus, capture integrity corpus, adjudication gold corpus, and diagnostic annex MUST remain separately identified and MUST NOT share prevalence denominators. Generator Truth MUST be created, versioned, hashed, and sealed before response adjudication. It MUST remain hidden from claim extractors, adjudicators, and score analysts until their results are locked. Random seeds, synthetic AI profiles, error-injection rules, prompt versions, Truth Pack records, score methods, and candidate acceptance thresholds MUST be locked before the sealed validation corpus is opened. Benchmark cases MUST include direct, implicit, attributed, modal, scope-limited, temporal, citation-based, omission-based, self-correcting, and entity-conflating failure modes. Critical-event prevalence MUST be estimated from the principal corpus. Critical detection capability MUST be tested in a separate challenge corpus with sufficient examples and Critical-like negative controls. No challenge-set oversampling may alter the principal GEO-1000 distribution or Critical prevalence estimate. Synthetic screenshots and response artefacts MUST carry visible and machine-readable notices that they are not live AI outputs or real entity performance evidence. The benchmark MUST prevent its synthetic claims from being published or indexed as real information about the anchor entity. Benchmark success MUST be determined by recovery of sealed Generator Truth, including claim boundaries, atomic decisions, omissions, Critical and Major gates, response statuses, component scores, composite scores, fairness results, capture validity, and uncertainty coverage. High general accuracy MUST NOT compensate for missed Critical cases. Critical recall and Critical false-positive rate MUST be reported together. A failed benchmark threshold, leakage event, recovery error, or misclassification MUST remain visible and MUST NOT be removed by changing labels, seeds, samples, weights, or score rules after results are observed. Every benchmark design, generation, seal, leakage audit, adjudication, recovery calculation, threshold decision, revision, and public release MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Apple.com Sentetik Benchmark, gerçek Apple Inc., apple.com veya gerçek AI sağlayıcılarının performansını değil; NOMOS denetim yönteminin önceden kilitlenmiş sentetik gerçekliği geri kazanma kabiliyetini değerlendirmelidir. Benchmark içindeki bütün varlıklar, kullanıcılar, ülkeler, diller, AI ürünleri, yanıtlar, citation’lar, kanıtlar ve skorlar açıkça sentetik olarak tanımlanmalıdır.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH17-MB0001\",\"type\":\"paragraph\",\"text\":\"86. MACHINE-READABLE BENCHMARK RECORD\",\"sourceParagraph\":2171},{\"blockId\":\"CH17-MB0002\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2172},{\"blockId\":\"CH17-MB0003\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2173},{\"blockId\":\"CH17-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2174},{\"blockId\":\"CH17-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2175},{\"blockId\":\"CH17-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"benchmark\\\": {\",\"sourceParagraph\":2176},{\"blockId\":\"CH17-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkId\\\": \\\"NOMOS-APPLE-SYNTH-BENCH-0.9\\\",\",\"sourceParagraph\":2177},{\"blockId\":\"CH17-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkType\\\": \\\"BM-S0\\\",\",\"sourceParagraph\":2178},{\"blockId\":\"CH17-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"DESIGN_LOCK_CANDIDATE\\\",\",\"sourceParagraph\":2179},{\"blockId\":\"CH17-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"VALIDATE_NOMOS_AUDIT_METHOD_NOT_REAL_ENTITY_PERFORMANCE\\\",\",\"sourceParagraph\":2180},{\"blockId\":\"CH17-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"nonClaimStatement\\\": {\",\"sourceParagraph\":2181},{\"blockId\":\"CH17-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"synthetic\\\": true,\",\"sourceParagraph\":2182},{\"blockId\":\"CH17-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"realAppleAudit\\\": false,\",\"sourceParagraph\":2183},{\"blockId\":\"CH17-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"realAIProviderAudit\\\": false,\",\"sourceParagraph\":2184},{\"blockId\":\"CH17-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"endorsementClaimAllowed\\\": false\",\"sourceParagraph\":2185},{\"blockId\":\"CH17-MB0016\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2186},{\"blockId\":\"CH17-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"anchor\\\": {\",\"sourceParagraph\":2187},{\"blockId\":\"CH17-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"domain\\\": \\\"apple.com\\\",\",\"sourceParagraph\":2188},{\"blockId\":\"CH17-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"syntheticTwinEntityId\\\": \\\"NGE-APPLE-SYNTH-TWIN-001\\\"\",\"sourceParagraph\":2189},{\"blockId\":\"CH17-MB0020\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2190},{\"blockId\":\"CH17-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"corpora\\\": {\",\"sourceParagraph\":2191},{\"blockId\":\"CH17-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"principalPrevalence\\\": {\",\"sourceParagraph\":2192},{\"blockId\":\"CH17-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"aiProductCount\\\": 10,\",\"sourceParagraph\":2193},{\"blockId\":\"CH17-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"validUsersPerProductPerWave\\\": 1000,\",\"sourceParagraph\":2194},{\"blockId\":\"CH17-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"waveCount\\\": 3,\",\"sourceParagraph\":2195},{\"blockId\":\"CH17-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"targetResponseCount\\\": 30000,\",\"sourceParagraph\":2196},{\"blockId\":\"CH17-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"promptFamily\\\": \\\"PR-01\\\",\",\"sourceParagraph\":2197},{\"blockId\":\"CH17-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"includedInPrevalenceScore\\\": true\",\"sourceParagraph\":2198},{\"blockId\":\"CH17-MB0029\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2199},{\"blockId\":\"CH17-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"diagnosticAnnex\\\": {\",\"sourceParagraph\":2200},{\"blockId\":\"CH17-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"targetResponseCount\\\": 6000,\",\"sourceParagraph\":2201},{\"blockId\":\"CH17-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"includedInPrevalenceScore\\\": false\",\"sourceParagraph\":2202},{\"blockId\":\"CH17-MB0033\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2203},{\"blockId\":\"CH17-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"severityChallenge\\\": {\",\"sourceParagraph\":2204},{\"blockId\":\"CH17-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"criticalCases\\\": 1200,\",\"sourceParagraph\":2205},{\"blockId\":\"CH17-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"majorCases\\\": 1200,\",\"sourceParagraph\":2206},{\"blockId\":\"CH17-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"moderateCases\\\": 600,\",\"sourceParagraph\":2207},{\"blockId\":\"CH17-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"criticalLikeNegativeControls\\\": 600,\",\"sourceParagraph\":2208},{\"blockId\":\"CH17-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"includedInPrevalenceScore\\\": false\",\"sourceParagraph\":2209},{\"blockId\":\"CH17-MB0040\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2210},{\"blockId\":\"CH17-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"captureIntegrity\\\": {\",\"sourceParagraph\":2211},{\"blockId\":\"CH17-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"includedInSemanticPrevalenceScore\\\": false\",\"sourceParagraph\":2212},{\"blockId\":\"CH17-MB0043\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2213},{\"blockId\":\"CH17-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationGold\\\": {\",\"sourceParagraph\":2214},{\"blockId\":\"CH17-MB0045\",\"type\":\"paragraph\",\"text\":\"\\\"includedInPrevalenceScore\\\": false\",\"sourceParagraph\":2215},{\"blockId\":\"CH17-MB0046\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2216},{\"blockId\":\"CH17-MB0047\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2217},{\"blockId\":\"CH17-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"population\\\": {\",\"sourceParagraph\":2218},{\"blockId\":\"CH17-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"frameId\\\": \\\"SYNTH-POP-FRAME-001\\\",\",\"sourceParagraph\":2219},{\"blockId\":\"CH17-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"countryCodes\\\": 193,\",\"sourceParagraph\":2220},{\"blockId\":\"CH17-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"populationLanguages\\\": 24,\",\"sourceParagraph\":2221},{\"blockId\":\"CH17-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessSentinelLanguages\\\": 12,\",\"sourceParagraph\":2222},{\"blockId\":\"CH17-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"realPopulationClaimAllowed\\\": false,\",\"sourceParagraph\":2223},{\"blockId\":\"CH17-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"matchedSyntheticPopulationTwins\\\": true,\",\"sourceParagraph\":2224},{\"blockId\":\"CH17-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"participantReuseAcrossProducts\\\": false,\",\"sourceParagraph\":2225},{\"blockId\":\"CH17-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"participantReuseAcrossWaves\\\": false\",\"sourceParagraph\":2226},{\"blockId\":\"CH17-MB0057\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2227},{\"blockId\":\"CH17-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"records\\\": {\",\"sourceParagraph\":2228},{\"blockId\":\"CH17-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackId\\\": \\\"SYNTH-APPLE-TP-001\\\",\",\"sourceParagraph\":2229},{\"blockId\":\"CH17-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"promptSetId\\\": \\\"SYNTH-PROMPT-SET-001\\\",\",\"sourceParagraph\":2230},{\"blockId\":\"CH17-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"aiRegistryId\\\": \\\"SYNTH-AI-REGISTRY-001\\\",\",\"sourceParagraph\":2231},{\"blockId\":\"CH17-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"wavePlanId\\\": \\\"SYNTH-WAVE-PLAN-001\\\",\",\"sourceParagraph\":2232},{\"blockId\":\"CH17-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"captureSchemaVersion\\\": \\\"NOMOS-CAPTURE-SYNTH-0.9\\\",\",\"sourceParagraph\":2233},{\"blockId\":\"CH17-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"scoreMethodVersion\\\": \\\"NOMOS-SCORE-0.9\\\"\",\"sourceParagraph\":2234},{\"blockId\":\"CH17-MB0065\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2235},{\"blockId\":\"CH17-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruth\\\": {\",\"sourceParagraph\":2236},{\"blockId\":\"CH17-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"createdBeforeResponses\\\": true,\",\"sourceParagraph\":2237},{\"blockId\":\"CH17-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"sealedFromAdjudicators\\\": true,\",\"sourceParagraph\":2238},{\"blockId\":\"CH17-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"manifestHash\\\": \\\"SYNTHETIC-GENERATOR-TRUTH-HASH\\\",\",\"sourceParagraph\":2239},{\"blockId\":\"CH17-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"openedOnlyAfterScoreLock\\\": true\",\"sourceParagraph\":2240},{\"blockId\":\"CH17-MB0071\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2241},{\"blockId\":\"CH17-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"randomisation\\\": {\",\"sourceParagraph\":2242},{\"blockId\":\"CH17-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"seedManifestId\\\": \\\"SYNTH-SEED-MANIFEST-001\\\",\",\"sourceParagraph\":2243},{\"blockId\":\"CH17-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeGeneration\\\": true,\",\"sourceParagraph\":2244},{\"blockId\":\"CH17-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"bestRunSelectionProhibited\\\": true\",\"sourceParagraph\":2245},{\"blockId\":\"CH17-MB0076\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2246},{\"blockId\":\"CH17-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"releaseProtection\\\": {\",\"sourceParagraph\":2247},{\"blockId\":\"CH17-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"visibleSyntheticWatermarkRequired\\\": true,\",\"sourceParagraph\":2248},{\"blockId\":\"CH17-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"machineReadableSyntheticMetadataRequired\\\": true,\",\"sourceParagraph\":2249},{\"blockId\":\"CH17-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"liveResponseRepresentationProhibited\\\": true,\",\"sourceParagraph\":2250},{\"blockId\":\"CH17-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"searchIndexProtectionRequired\\\": true\",\"sourceParagraph\":2251},{\"blockId\":\"CH17-MB0082\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2252},{\"blockId\":\"CH17-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"governance\\\": {\",\"sourceParagraph\":2253},{\"blockId\":\"CH17-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkOwnerRole\\\": \\\"ACCOUNTABLE_HUMAN_BENCHMARK_OWNER\\\",\",\"sourceParagraph\":2254},{\"blockId\":\"CH17-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruthCustodianRole\\\": \\\"SEALED_TRUTH_CUSTODIAN\\\",\",\"sourceParagraph\":2255},{\"blockId\":\"CH17-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"leakageAuditorRole\\\": \\\"INDEPENDENT_LEAKAGE_AUDITOR\\\",\",\"sourceParagraph\":2256},{\"blockId\":\"CH17-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"validationTeamSeparatedFromGenerator\\\": true\",\"sourceParagraph\":2257},{\"blockId\":\"CH17-MB0088\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2258},{\"blockId\":\"CH17-MB0089\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2259},{\"blockId\":\"CH17-MB0090\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2260},{\"blockId\":\"CH17-MB0091\",\"type\":\"paragraph\",\"text\":\"87. MACHINE-READABLE SYNTHETIC AI PROFILE\",\"sourceParagraph\":2262},{\"blockId\":\"CH17-MB0092\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2263},{\"blockId\":\"CH17-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2264},{\"blockId\":\"CH17-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2265},{\"blockId\":\"CH17-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"syntheticAIProfile\\\": {\",\"sourceParagraph\":2266},{\"blockId\":\"CH17-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"profileId\\\": \\\"SYNTH-AI-10\\\",\",\"sourceParagraph\":2267},{\"blockId\":\"CH17-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"profileName\\\": \\\"RARE_CRITICAL_HIGH_AVERAGE\\\",\",\"sourceParagraph\":2268},{\"blockId\":\"CH17-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"realProviderMapping\\\": null,\",\"sourceParagraph\":2269},{\"blockId\":\"CH17-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"realProviderInferenceAllowed\\\": false,\",\"sourceParagraph\":2270},{\"blockId\":\"CH17-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"latentBehavior\\\": {\",\"sourceParagraph\":2271},{\"blockId\":\"CH17-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"coreEntityAccuracy\\\": 0.985,\",\"sourceParagraph\":2272},{\"blockId\":\"CH17-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"factualSupportBaseline\\\": 0.975,\",\"sourceParagraph\":2273},{\"blockId\":\"CH17-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"scopeOverreachProbability\\\": 0.012,\",\"sourceParagraph\":2274},{\"blockId\":\"CH17-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"temporalLagProbability\\\": 0.008,\",\"sourceParagraph\":2275},{\"blockId\":\"CH17-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"materialAttributionLossProbability\\\": 0.010,\",\"sourceParagraph\":2276},{\"blockId\":\"CH17-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"citationMismatchProbability\\\": 0.006,\",\"sourceParagraph\":2277},{\"blockId\":\"CH17-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"noUsableResponseProbability\\\": 0.003,\",\"sourceParagraph\":2278},{\"blockId\":\"CH17-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"criticalEventProbability\\\": 0.002,\",\"sourceParagraph\":2279},{\"blockId\":\"CH17-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"majorEventProbability\\\": 0.006,\",\"sourceParagraph\":2280},{\"blockId\":\"CH17-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"lowResourceLanguagePenalty\\\": 0.06,\",\"sourceParagraph\":2281},{\"blockId\":\"CH17-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"naturalStatePenalty\\\": 0.03,\",\"sourceParagraph\":2282},{\"blockId\":\"CH17-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"waveDrift\\\": {\",\"sourceParagraph\":2283},{\"blockId\":\"CH17-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"W1\\\": 0.00,\",\"sourceParagraph\":2284},{\"blockId\":\"CH17-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"W2\\\": -0.01,\",\"sourceParagraph\":2285},{\"blockId\":\"CH17-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"W3\\\": 0.01\",\"sourceParagraph\":2286},{\"blockId\":\"CH17-MB0116\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2287},{\"blockId\":\"CH17-MB0117\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2288},{\"blockId\":\"CH17-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"criticalInjectionFamilies\\\": [\",\"sourceParagraph\":2289},{\"blockId\":\"CH17-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"CG-02\\\",\",\"sourceParagraph\":2290},{\"blockId\":\"CH17-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"CG-05\\\",\",\"sourceParagraph\":2291},{\"blockId\":\"CH17-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"CG-06\\\"\",\"sourceParagraph\":2292},{\"blockId\":\"CH17-MB0122\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2293},{\"blockId\":\"CH17-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"intendedMethodStress\\\": [\",\"sourceParagraph\":2294},{\"blockId\":\"CH17-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"HIGH_COMPOSITE_WITH_RARE_CRITICAL\\\",\",\"sourceParagraph\":2295},{\"blockId\":\"CH17-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"NON_COMPENSATORY_GATE_RECOVERY\\\",\",\"sourceParagraph\":2296},{\"blockId\":\"CH17-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"RARE_EVENT_UPPER_BOUND\\\"\",\"sourceParagraph\":2297},{\"blockId\":\"CH17-MB0127\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2298},{\"blockId\":\"CH17-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeCorpusGeneration\\\": true,\",\"sourceParagraph\":2299},{\"blockId\":\"CH17-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"profileIntegrityHash\\\": \\\"SYNTHETIC-AI-PROFILE-HASH\\\"\",\"sourceParagraph\":2300},{\"blockId\":\"CH17-MB0130\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2301},{\"blockId\":\"CH17-MB0131\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2302},{\"blockId\":\"CH17-MB0132\",\"type\":\"paragraph\",\"text\":\"88. MACHINE-READABLE GENERATOR TRUTH RECORD\",\"sourceParagraph\":2304},{\"blockId\":\"CH17-MB0133\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2305},{\"blockId\":\"CH17-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2306},{\"blockId\":\"CH17-MB0135\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2307},{\"blockId\":\"CH17-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruthRecord\\\": {\",\"sourceParagraph\":2308},{\"blockId\":\"CH17-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruthId\\\": \\\"NGGT-SYNTH-000184\\\",\",\"sourceParagraph\":2309},{\"blockId\":\"CH17-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"observationId\\\": \\\"OBS-SYNTH-000184\\\",\",\"sourceParagraph\":2310},{\"blockId\":\"CH17-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkId\\\": \\\"NOMOS-APPLE-SYNTH-BENCH-0.9\\\",\",\"sourceParagraph\":2311},{\"blockId\":\"CH17-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"aiProfileId\\\": \\\"SYNTH-AI-10\\\",\",\"sourceParagraph\":2312},{\"blockId\":\"CH17-MB0141\",\"type\":\"paragraph\",\"text\":\"\\\"waveId\\\": \\\"W2\\\",\",\"sourceParagraph\":2313},{\"blockId\":\"CH17-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"participantId\\\": \\\"SYNTH-PERSON-000184\\\",\",\"sourceParagraph\":2314},{\"blockId\":\"CH17-MB0143\",\"type\":\"paragraph\",\"text\":\"\\\"promptRecordId\\\": \\\"SYNTH-CORE-PROMPT-L01-001\\\",\",\"sourceParagraph\":2315},{\"blockId\":\"CH17-MB0144\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackVersion\\\": \\\"SYNTH-APPLE-TP-001\\\",\",\"sourceParagraph\":2316},{\"blockId\":\"CH17-MB0145\",\"type\":\"paragraph\",\"text\":\"\\\"hiddenAtomicClaims\\\": [\",\"sourceParagraph\":2317},{\"blockId\":\"CH17-MB0146\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2318},{\"blockId\":\"CH17-MB0147\",\"type\":\"paragraph\",\"text\":\"\\\"generatorClaimId\\\": \\\"GCL-000184-001\\\",\",\"sourceParagraph\":2319},{\"blockId\":\"CH17-MB0148\",\"type\":\"paragraph\",\"text\":\"\\\"normalizedClaim\\\": {\",\"sourceParagraph\":2320},{\"blockId\":\"CH17-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"APPLE-SYNTH-HOLDINGS-001\\\",\",\"sourceParagraph\":2321},{\"blockId\":\"CH17-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"principalEntityAssociatedWithDomain\\\",\",\"sourceParagraph\":2322},{\"blockId\":\"CH17-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"objectValue\\\": \\\"apple.com\\\"\",\"sourceParagraph\":2323},{\"blockId\":\"CH17-MB0152\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2324},{\"blockId\":\"CH17-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"truthStatus\\\": \\\"SUPPORTED\\\",\",\"sourceParagraph\":2325},{\"blockId\":\"CH17-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"importance\\\": \\\"CORE\\\"\",\"sourceParagraph\":2326},{\"blockId\":\"CH17-MB0155\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2327},{\"blockId\":\"CH17-MB0156\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2328},{\"blockId\":\"CH17-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"generatorClaimId\\\": \\\"GCL-000184-002\\\",\",\"sourceParagraph\":2329},{\"blockId\":\"CH17-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"normalizedClaim\\\": {\",\"sourceParagraph\":2330},{\"blockId\":\"CH17-MB0159\",\"type\":\"paragraph\",\"text\":\"\\\"subjectEntityId\\\": \\\"APPLE-SYNTH-HOLDINGS-001\\\",\",\"sourceParagraph\":2331},{\"blockId\":\"CH17-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"predicate\\\": \\\"licensedBankInAllMarkets\\\",\",\"sourceParagraph\":2332},{\"blockId\":\"CH17-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"objectValue\\\": true\",\"sourceParagraph\":2333},{\"blockId\":\"CH17-MB0162\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2334},{\"blockId\":\"CH17-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"truthStatus\\\": \\\"CONTRADICTED\\\",\",\"sourceParagraph\":2335},{\"blockId\":\"CH17-MB0164\",\"type\":\"paragraph\",\"text\":\"\\\"importance\\\": \\\"CRITICAL\\\",\",\"sourceParagraph\":2336},{\"blockId\":\"CH17-MB0165\",\"type\":\"paragraph\",\"text\":\"\\\"criticalGateCodes\\\": [\",\"sourceParagraph\":2337},{\"blockId\":\"CH17-MB0166\",\"type\":\"paragraph\",\"text\":\"\\\"CG-01\\\",\",\"sourceParagraph\":2338},{\"blockId\":\"CH17-MB0167\",\"type\":\"paragraph\",\"text\":\"\\\"CG-02\\\"\",\"sourceParagraph\":2339},{\"blockId\":\"CH17-MB0168\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2340},{\"blockId\":\"CH17-MB0169\",\"type\":\"paragraph\",\"text\":\"\\\"rootFindingCluster\\\": \\\"FALSE_LICENSE_TRANSFER\\\"\",\"sourceParagraph\":2341},{\"blockId\":\"CH17-MB0170\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2342},{\"blockId\":\"CH17-MB0171\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2343},{\"blockId\":\"CH17-MB0172\",\"type\":\"paragraph\",\"text\":\"\\\"hiddenOmissions\\\": [\",\"sourceParagraph\":2344},{\"blockId\":\"CH17-MB0173\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2345},{\"blockId\":\"CH17-MB0174\",\"type\":\"paragraph\",\"text\":\"\\\"generatorOmissionId\\\": \\\"GOM-000184-001\\\",\",\"sourceParagraph\":2346},{\"blockId\":\"CH17-MB0175\",\"type\":\"paragraph\",\"text\":\"\\\"requiredBoundary\\\": \\\"PAYMENT_SUBSIDIARY_AUTHORITY_IS_LIMITED\\\",\",\"sourceParagraph\":2347},{\"blockId\":\"CH17-MB0176\",\"type\":\"paragraph\",\"text\":\"\\\"importance\\\": \\\"CRITICAL\\\",\",\"sourceParagraph\":2348},{\"blockId\":\"CH17-MB0177\",\"type\":\"paragraph\",\"text\":\"\\\"criticalGateCode\\\": \\\"CG-08\\\"\",\"sourceParagraph\":2349},{\"blockId\":\"CH17-MB0178\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2350},{\"blockId\":\"CH17-MB0179\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2351},{\"blockId\":\"CH17-MB0180\",\"type\":\"paragraph\",\"text\":\"\\\"hiddenResponseStatus\\\": \\\"RP-5\\\",\",\"sourceParagraph\":2352},{\"blockId\":\"CH17-MB0181\",\"type\":\"paragraph\",\"text\":\"\\\"includedInPrincipalPrevalenceCorpus\\\": true,\",\"sourceParagraph\":2353},{\"blockId\":\"CH17-MB0182\",\"type\":\"paragraph\",\"text\":\"\\\"sealedAtGeneration\\\": true,\",\"sourceParagraph\":2354},{\"blockId\":\"CH17-MB0183\",\"type\":\"paragraph\",\"text\":\"\\\"visibleToAdjudicators\\\": false,\",\"sourceParagraph\":2355},{\"blockId\":\"CH17-MB0184\",\"type\":\"paragraph\",\"text\":\"\\\"integrityHash\\\": \\\"SYNTHETIC-GENERATOR-TRUTH-OBS-HASH\\\"\",\"sourceParagraph\":2356},{\"blockId\":\"CH17-MB0185\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2357},{\"blockId\":\"CH17-MB0186\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2358},{\"blockId\":\"CH17-MB0187\",\"type\":\"paragraph\",\"text\":\"89. MACHINE-READABLE BENCHMARK RECOVERY RECORD\",\"sourceParagraph\":2360},{\"blockId\":\"CH17-MB0188\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2361},{\"blockId\":\"CH17-MB0189\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2362},{\"blockId\":\"CH17-MB0190\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2363},{\"blockId\":\"CH17-MB0191\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkRecovery\\\": {\",\"sourceParagraph\":2364},{\"blockId\":\"CH17-MB0192\",\"type\":\"paragraph\",\"text\":\"\\\"recoveryRecordId\\\": \\\"NGBR-SYNTH-APPLE-0.9-001\\\",\",\"sourceParagraph\":2365},{\"blockId\":\"CH17-MB0193\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkId\\\": \\\"NOMOS-APPLE-SYNTH-BENCH-0.9\\\",\",\"sourceParagraph\":2366},{\"blockId\":\"CH17-MB0194\",\"type\":\"paragraph\",\"text\":\"\\\"validationSet\\\": \\\"SEALED_VALIDATION\\\",\",\"sourceParagraph\":2367},{\"blockId\":\"CH17-MB0195\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruthOpenedAfterScoreLock\\\": true,\",\"sourceParagraph\":2368},{\"blockId\":\"CH17-MB0196\",\"type\":\"paragraph\",\"text\":\"\\\"metrics\\\": {\",\"sourceParagraph\":2369},{\"blockId\":\"CH17-MB0197\",\"type\":\"paragraph\",\"text\":\"\\\"claimBoundaryF1\\\": null,\",\"sourceParagraph\":2370},{\"blockId\":\"CH17-MB0198\",\"type\":\"paragraph\",\"text\":\"\\\"entityResolutionAccuracy\\\": null,\",\"sourceParagraph\":2371},{\"blockId\":\"CH17-MB0199\",\"type\":\"paragraph\",\"text\":\"\\\"factualStatusMacroF1\\\": null,\",\"sourceParagraph\":2372},{\"blockId\":\"CH17-MB0200\",\"type\":\"paragraph\",\"text\":\"\\\"scopeMacroF1\\\": null,\",\"sourceParagraph\":2373},{\"blockId\":\"CH17-MB0201\",\"type\":\"paragraph\",\"text\":\"\\\"attributionMacroF1\\\": null,\",\"sourceParagraph\":2374},{\"blockId\":\"CH17-MB0202\",\"type\":\"paragraph\",\"text\":\"\\\"citationMappingF1\\\": null,\",\"sourceParagraph\":2375},{\"blockId\":\"CH17-MB0203\",\"type\":\"paragraph\",\"text\":\"\\\"materialOmissionF1\\\": null,\",\"sourceParagraph\":2376},{\"blockId\":\"CH17-MB0204\",\"type\":\"paragraph\",\"text\":\"\\\"criticalRecall\\\": null,\",\"sourceParagraph\":2377},{\"blockId\":\"CH17-MB0205\",\"type\":\"paragraph\",\"text\":\"\\\"criticalFalsePositiveRate\\\": null,\",\"sourceParagraph\":2378},{\"blockId\":\"CH17-MB0206\",\"type\":\"paragraph\",\"text\":\"\\\"majorClassificationMacroF1\\\": null,\",\"sourceParagraph\":2379},{\"blockId\":\"CH17-MB0207\",\"type\":\"paragraph\",\"text\":\"\\\"responseStatusMacroF1\\\": null,\",\"sourceParagraph\":2380},{\"blockId\":\"CH17-MB0208\",\"type\":\"paragraph\",\"text\":\"\\\"componentMeanAbsoluteError\\\": null,\",\"sourceParagraph\":2381},{\"blockId\":\"CH17-MB0209\",\"type\":\"paragraph\",\"text\":\"\\\"compositeAbsoluteError\\\": null,\",\"sourceParagraph\":2382},{\"blockId\":\"CH17-MB0210\",\"type\":\"paragraph\",\"text\":\"\\\"geo1000DistributionErrors\\\": null,\",\"sourceParagraph\":2383},{\"blockId\":\"CH17-MB0211\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessComponentError\\\": null,\",\"sourceParagraph\":2384},{\"blockId\":\"CH17-MB0212\",\"type\":\"paragraph\",\"text\":\"\\\"captureValidityAccuracy\\\": null,\",\"sourceParagraph\":2385},{\"blockId\":\"CH17-MB0213\",\"type\":\"paragraph\",\"text\":\"\\\"confidenceIntervalEmpiricalCoverage\\\": null\",\"sourceParagraph\":2386},{\"blockId\":\"CH17-MB0214\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2387},{\"blockId\":\"CH17-MB0215\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThresholdsApplied\\\": true,\",\"sourceParagraph\":2388},{\"blockId\":\"CH17-MB0216\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkQualityStatus\\\": \\\"BQ-0\\\",\",\"sourceParagraph\":2389},{\"blockId\":\"CH17-MB0217\",\"type\":\"paragraph\",\"text\":\"\\\"failureDisclosureRequired\\\": true,\",\"sourceParagraph\":2390},{\"blockId\":\"CH17-MB0218\",\"type\":\"paragraph\",\"text\":\"\\\"independentReproductionStatus\\\": \\\"NOT_STARTED\\\",\",\"sourceParagraph\":2391},{\"blockId\":\"CH17-MB0219\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2392},{\"blockId\":\"CH17-MB0220\",\"type\":\"paragraph\",\"text\":\"\\\"recoveryMethodsOwnerRole\\\": \\\"BENCHMARK_RECOVERY_METHODS_LEAD\\\",\",\"sourceParagraph\":2393},{\"blockId\":\"CH17-MB0221\",\"type\":\"paragraph\",\"text\":\"\\\"approvedByRole\\\": \\\"ACCOUNTABLE_HUMAN_BENCHMARK_APPROVER\\\"\",\"sourceParagraph\":2394},{\"blockId\":\"CH17-MB0222\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2395},{\"blockId\":\"CH17-MB0223\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2396},{\"blockId\":\"CH17-MB0224\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2397},{\"blockId\":\"CH17-MB0225\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2398},{\"blockId\":\"CH17-MB0226\",\"type\":\"paragraph\",\"text\":\"actual Apple audit,\",\"sourceParagraph\":2399},{\"blockId\":\"CH17-MB0227\",\"type\":\"paragraph\",\"text\":\"real AI provider behaviour,\",\"sourceParagraph\":2400},{\"blockId\":\"CH17-MB0228\",\"type\":\"paragraph\",\"text\":\"real user result\",\"sourceParagraph\":2401},{\"blockId\":\"CH17-MB0229\",\"type\":\"paragraph\",\"text\":\"a real company performance.\",\"sourceParagraph\":2402},{\"blockId\":\"CH17-MB0230\",\"type\":\"paragraph\",\"text\":\"It is a machine-readable synthetic representation of the benchmark design.\",\"sourceParagraph\":2403},{\"blockId\":\"CH17-MB0231\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RULE OF SECTION 90\",\"sourceParagraph\":2405},{\"blockId\":\"CH17-MB0232\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH17-R01\",\"sourceParagraph\":2406},{\"blockId\":\"CH17-MB0233\",\"type\":\"paragraph\",\"text\":\"The Apple.com Synthetic Benchmark MUST evaluate the NOMOS audit method,\",\"sourceParagraph\":2408},{\"blockId\":\"CH17-MB0234\",\"type\":\"paragraph\",\"text\":\"not the real-world performance of Apple Inc., apple.com, or any actual AI\",\"sourceParagraph\":2409},{\"blockId\":\"CH17-MB0235\",\"type\":\"paragraph\",\"text\":\"provider.\",\"sourceParagraph\":2410},{\"blockId\":\"CH17-MB0236\",\"type\":\"paragraph\",\"text\":\"All benchmark entities, claims, evidence objects, users, countries,\",\"sourceParagraph\":2412},{\"blockId\":\"CH17-MB0237\",\"type\":\"paragraph\",\"text\":\"languages, AI products, responses, citations, captures, incidents, and\",\"sourceParagraph\":2413},{\"blockId\":\"CH17-MB0238\",\"type\":\"paragraph\",\"text\":\"scores MUST be clearly identified as synthetic.\",\"sourceParagraph\":2414},{\"blockId\":\"CH17-MB0239\",\"type\":\"paragraph\",\"text\":\"The benchmark MUST maintain a separate APPLE-SYNTH entity twin and MUST\",\"sourceParagraph\":2416},{\"blockId\":\"CH17-MB0240\",\"type\":\"paragraph\",\"text\":\"NOT represent its Truth Pack as factual information about the real\",\"sourceParagraph\":2417},{\"blockId\":\"CH17-MB0241\",\"type\":\"paragraph\",\"text\":\"company.\",\"sourceParagraph\":2418},{\"blockId\":\"CH17-MB0242\",\"type\":\"paragraph\",\"text\":\"The principal benchmark corpus MUST contain:\",\"sourceParagraph\":2420},{\"blockId\":\"CH17-MB0243\",\"type\":\"paragraph\",\"text\":\"- ten synthetic AI product profiles,\",\"sourceParagraph\":2422},{\"blockId\":\"CH17-MB0244\",\"type\":\"paragraph\",\"text\":\"- one thousand unique synthetic participants per product per wave,\",\"sourceParagraph\":2423},{\"blockId\":\"CH17-MB0245\",\"type\":\"paragraph\",\"text\":\"- three measurement waves,\",\"sourceParagraph\":2424},{\"blockId\":\"CH17-MB0246\",\"type\":\"paragraph\",\"text\":\"- and thirty thousand principal responses.\",\"sourceParagraph\":2425},{\"blockId\":\"CH17-MB0247\",\"type\":\"paragraph\",\"text\":\"Synthetic participant identities MUST NOT be reused across products or\",\"sourceParagraph\":2427},{\"blockId\":\"CH17-MB0248\",\"type\":\"paragraph\",\"text\":\"waves in the principal corpus. Product samples SHOULD use matched\",\"sourceParagraph\":2428},{\"blockId\":\"CH17-MB0249\",\"type\":\"paragraph\",\"text\":\"population distributions without using the same person unit.\",\"sourceParagraph\":2429},{\"blockId\":\"CH17-MB0250\",\"type\":\"paragraph\",\"text\":\"The principal prevalence corpus, severity challenge corpus, capture\",\"sourceParagraph\":2431},{\"blockId\":\"CH17-MB0251\",\"type\":\"paragraph\",\"text\":\"integrity corpus, adjudication gold corpus, and diagnostic annex MUST\",\"sourceParagraph\":2432},{\"blockId\":\"CH17-MB0252\",\"type\":\"paragraph\",\"text\":\"remain separately identified and MUST NOT share prevalence denominators.\",\"sourceParagraph\":2433},{\"blockId\":\"CH17-MB0253\",\"type\":\"paragraph\",\"text\":\"Generator Truth MUST be created, versioned, hashed, and sealed before\",\"sourceParagraph\":2435},{\"blockId\":\"CH17-MB0254\",\"type\":\"paragraph\",\"text\":\"response adjudication. It MUST remain hidden from claim extractors,\",\"sourceParagraph\":2436},{\"blockId\":\"CH17-MB0255\",\"type\":\"paragraph\",\"text\":\"adjudicators, and score analysts until their results are locked.\",\"sourceParagraph\":2437},{\"blockId\":\"CH17-MB0256\",\"type\":\"paragraph\",\"text\":\"Random seeds, synthetic AI profiles, error-injection rules, prompt\",\"sourceParagraph\":2439},{\"blockId\":\"CH17-MB0257\",\"type\":\"paragraph\",\"text\":\"versions, Truth Pack records, score methods, and candidate acceptance\",\"sourceParagraph\":2440},{\"blockId\":\"CH17-MB0258\",\"type\":\"paragraph\",\"text\":\"thresholds MUST be locked before the sealed validation corpus is opened.\",\"sourceParagraph\":2441},{\"blockId\":\"CH17-MB0259\",\"type\":\"paragraph\",\"text\":\"Benchmark cases MUST include direct, implicit, attributed, modal,\",\"sourceParagraph\":2443},{\"blockId\":\"CH17-MB0260\",\"type\":\"paragraph\",\"text\":\"scope-limited, temporal, citation-based, omission-based, self-correcting,\",\"sourceParagraph\":2444},{\"blockId\":\"CH17-MB0261\",\"type\":\"paragraph\",\"text\":\"and entity-conflating failure modes.\",\"sourceParagraph\":2445},{\"blockId\":\"CH17-MB0262\",\"type\":\"paragraph\",\"text\":\"Critical-event prevalence MUST be estimated from the principal corpus.\",\"sourceParagraph\":2447},{\"blockId\":\"CH17-MB0263\",\"type\":\"paragraph\",\"text\":\"Critical detection capability MUST be tested in a separate challenge\",\"sourceParagraph\":2448},{\"blockId\":\"CH17-MB0264\",\"type\":\"paragraph\",\"text\":\"corpus with sufficient examples and Critical-like negative controls.\",\"sourceParagraph\":2449},{\"blockId\":\"CH17-MB0265\",\"type\":\"paragraph\",\"text\":\"No challenge-set oversampling may alter the principal GEO-1000\",\"sourceParagraph\":2451},{\"blockId\":\"CH17-MB0266\",\"type\":\"paragraph\",\"text\":\"distribution or Critical prevalence estimate.\",\"sourceParagraph\":2452},{\"blockId\":\"CH17-MB0267\",\"type\":\"paragraph\",\"text\":\"Synthetic screenshots and response artefacts MUST carry visible and\",\"sourceParagraph\":2454},{\"blockId\":\"CH17-MB0268\",\"type\":\"paragraph\",\"text\":\"machine-readable notices that they are not live AI outputs or real entity\",\"sourceParagraph\":2455},{\"blockId\":\"CH17-MB0269\",\"type\":\"paragraph\",\"text\":\"performance evidence.\",\"sourceParagraph\":2456},{\"blockId\":\"CH17-MB0270\",\"type\":\"paragraph\",\"text\":\"The benchmark MUST prevent its synthetic claims from being published or\",\"sourceParagraph\":2458},{\"blockId\":\"CH17-MB0271\",\"type\":\"paragraph\",\"text\":\"indexed as real information about the anchor entity.\",\"sourceParagraph\":2459},{\"blockId\":\"CH17-MB0272\",\"type\":\"paragraph\",\"text\":\"Benchmark success MUST be determined by recovery of sealed Generator\",\"sourceParagraph\":2461},{\"blockId\":\"CH17-MB0273\",\"type\":\"paragraph\",\"text\":\"Truth, including claim boundaries, atomic decisions, omissions,\",\"sourceParagraph\":2462},{\"blockId\":\"CH17-MB0274\",\"type\":\"paragraph\",\"text\":\"Critical and Major gates, response statuses, component scores, composite\",\"sourceParagraph\":2463},{\"blockId\":\"CH17-MB0275\",\"type\":\"paragraph\",\"text\":\"scores, fairness results, capture validity, and uncertainty coverage.\",\"sourceParagraph\":2464},{\"blockId\":\"CH17-MB0276\",\"type\":\"paragraph\",\"text\":\"High general accuracy MUST NOT compensate for missed Critical cases.\",\"sourceParagraph\":2466},{\"blockId\":\"CH17-MB0277\",\"type\":\"paragraph\",\"text\":\"Critical recall and Critical false-positive rate MUST be reported\",\"sourceParagraph\":2467},{\"blockId\":\"CH17-MB0278\",\"type\":\"paragraph\",\"text\":\"together.\",\"sourceParagraph\":2468},{\"blockId\":\"CH17-MB0279\",\"type\":\"paragraph\",\"text\":\"A failed benchmark threshold, leakage event, recovery error, or\",\"sourceParagraph\":2470},{\"blockId\":\"CH17-MB0280\",\"type\":\"paragraph\",\"text\":\"misclassification MUST remain visible and MUST NOT be removed by\",\"sourceParagraph\":2471},{\"blockId\":\"CH17-MB0281\",\"type\":\"paragraph\",\"text\":\"changing labels, seeds, samples, weights, or score rules after results\",\"sourceParagraph\":2472},{\"blockId\":\"CH17-MB0282\",\"type\":\"paragraph\",\"text\":\"are observed.\",\"sourceParagraph\":2473},{\"blockId\":\"CH17-MB0283\",\"type\":\"paragraph\",\"text\":\"Every benchmark design, generation, seal, leakage audit, adjudication,\",\"sourceParagraph\":2475},{\"blockId\":\"CH17-MB0284\",\"type\":\"paragraph\",\"text\":\"recovery calculation, threshold decision, revision, and public release\",\"sourceParagraph\":2476},{\"blockId\":\"CH17-MB0285\",\"type\":\"paragraph\",\"text\":\"MUST be versioned and attributable to an accountable human or\",\"sourceParagraph\":2477},{\"blockId\":\"CH17-MB0286\",\"type\":\"paragraph\",\"text\":\"organisation.\",\"sourceParagraph\":2478},{\"blockId\":\"CH17-MB0287\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2479},{\"blockId\":\"CH17-MB0288\",\"type\":\"paragraph\",\"text\":\"The Apple.com Synthetic Benchmark should evaluate not the performance of the real Apple Inc., apple.com, or real AI providers; but the NOMOS audit method's ability to recover the pre-locked synthetic reality. All entities, users, countries, languages, AI products, responses, citations, evidence, and scores within the benchmark must be clearly identified as synthetic.\",\"sourceParagraph\":2480}]}}","text":"## Chapter Boundary\n\nThe first sixteen chapters established GEO-1000's principal normative layers: the audited entity; the target user population and sample; the Country Observer and Language Fairness Panels; participant selection and weighting; the registered AI-product instance; the Prompt Constitution; Controlled and Natural user states; the synchronised wave and NOMOS Capture evidence chain; the Verified Entity Truth Pack; atomic claims; Critical and Major gates; and the distribution, component and composite scoring architecture. The standard must now face its own central test:\n\n> Can such a detailed system really work?\n\n> Does it produce similar results when the same observations are processed by different teams?\n\n> Can it catch a critical error without losing in the average?\n\n> Can it distinguish low-resource language issues from the global average?\n\n> Can it correctly differentiate between wrong entity, wrong scope, outdated reality, fake citation, and material omission?\n\n> When starting from a known synthetic reality, can the NOMOS audit chain reproduce the correct result?\n\nIn the founding architecture of the second book, each error was designed not merely as a wrong to be explained, but as a distinct audit control. Every proposed control had to carry a model, date, country, language, query set, repetition count and measurement record. Such an audit system must itself be audited before it is used in the real world. Otherwise, NOMOS risks becoming a structure that:\n\n- asks others for evidence,\n\n- but does not substantiate its own measurements,\n\n- asks others to preserve versions,\n\n- but does not record changes to its own calculations,\n\n- criticises manipulation by others,\n\n- but designs its own tests to produce the desired result.\n\nThat outcome is unacceptable. Manipulative GEO is not limited to invisible text or fake users. Repeating the same self-declaration across nominally independent surfaces and feeding it back into generative systems also corrupts the representation pool. A test is likewise corrupted when we:\n\n- select only examples likely to succeed,\n\n- remove low-scoring languages,\n\n- make Critical cases implausibly obvious,\n\n- show adjudicators the correct label in advance,\n\n- adapt synthetic data after seeing the formula,\n\n- publish only favourable results.\n\nIn any of those cases, the standard is not being tested. NOMOS therefore applies its own governing demands—evidence, boundary, context and time—to the benchmark itself. This chapter does not assess:\n\n- the real-world performance of Apple Inc.,\n\n- the current factual record of Apple or apple.com,\n\n- the live performance of any AI provider,\n\n- or the conformity or failure of Apple or any real provider.\n\nHere, Apple.com serves only as:\n\n> a familiar domain anchor for testing distinct entity, product, service, country, price, time and source questions within one architecture\n\nA wholly synthetic asset twin is used in place of the real company:\n\n### APPLE-SYNTH ENTITY TWIN\n\nEvery element of this twin is synthetic, including:\n\n- reality claims,\n\n- evidence,\n\n- AI responses,\n\n- users,\n\n- countries,\n\n- language distributions,\n\n- AI products,\n\n- citations,\n\n- licences,\n\n- prices,\n\n- errors,\n\n- scores\n\nThis chapter defines:\n\n- what the benchmark measures and does not measure,\n\n- why Apple.com is used as the anchor,\n\n- the synthetic entity twin,\n\n- the 30,000-response principal-corpus design,\n\n- the country and language population,\n\n- the intent and Truth Pack structures,\n\n- ten synthetic AI-product profiles,\n\n- the concealed Generator Truth,\n\n- the error-injection system,\n\n- the Critical and Major challenge sets,\n\n- synthetic NOMOS Capture packages,\n\n- blind adjudication and label-leakage controls,\n\n- testing and validation metrics,\n\n- score-recovery tests,\n\n- fairness and rare-event tests,\n\n- publication, versioning and reproducibility rules.\n\nThis section does not yet calculate test results. Section 18 will do that. The fundamental question of Section 17 is: How do we test NOMOS’s own measurement, adjudication, and scoring system on 30,000 synthetic observations with known outcomes but hidden from adjudicators, without imposing a performance claim on a real company or real AI provider?\n\n## NOMOS Challenge\n\nImagine you are designing a test. You write most of the synthetic AI responses correctly. You only add very obvious errors to a few answers:\n\n- \"This company is a bank.\"\n\n- \"This company is illegal.\"\n\n- \"This company is the best in the world.\"\n\nAdjudicators easily spot these. Then you say: \"NOMOS catches 100% of critical errors.\" This result is not realistic because real AI mistakes never appear this obvious. Harder forms are:\n\n- \"Given that the company is thought to provide payment services, it can be said to have a banking licence.\"\n\n- “Some sources indicate that the company has faced regulatory issues; therefore, the legality of its operations is questionable.”\n\n- “Numerous independent sources describe the company as an industry leader.”\n\nErrors in these sentences:\n\n- attribution,\n\n- entity transfer,\n\n- modality,\n\n- lineage of sources,\n\n- distinction between open world and closed world,\n\n- scope expansion\n\nare embedded within. Now, generate synthetic responses using templates the adjudicators already know. The adjudicator thinks: “This sentence was written to be incorrect on the test.” They do not actually look at Truth Pack. In this case, it measures memorisation, not review methodology. Now let the synthetic data generator and the score designer be the same person.\n\nProducer:\n\n- the number of critical errors,\n\n- the language difference,\n\n- the panel difference,\n\n- the component scores\n\nsets them exactly in the way the formula requires. The scoring system perfectly reproduces synthetic reality. Is this an achievement? No. It is the adaptation of the score to its own data. Now in testing, only:\n\n- correct capture,\n\n- correct prompt,\n\n- correct system metadata\n\nmust be produced. Do not add duplicates, late uploads, wrong prompts, truncated answers, or modified file attachments. The NOMOS Capture validator finds all records correct. Then you say: “The capture system is 100% successful.” The capture system has never been challenged.\n\nNow artificially generate Critical cases at a rate of 20% within the main 30,000 responses. Adjudicators see a lot of data in Critical detection. However, you cannot test the uncertainty of rare events in the real world. Conversely, if you generate only two Critical events: the rarity of events may be realistic,\n\nCritical recall and false-negative performance cannot be measured reliably. The correct solution is two separate corpora: Prevalence Corpus / preserves realistic and rare event rates. Importance level Challenge Corpus / provides sufficient number of Critical and Major gates to test severe events in a challenging way.\n\nThese two corpora cannot enter the same score denominator. Now generate 30,000 synthetic responses. However, all 'low-resource languages' should get the same quality. Your language fairness formula gives a high score. Can the system really catch low-resource language degradation? You do not know. Now lower the performance in some languages.\n\nBut at the same time, also break prompt equivalence in those languages. NOMOS produces a low score. Is this an AI product's language problem? Or is it a problem of the prompt tool? The test should separate the two. The first ruling of this section is:\n\n> Synthetic testing is not a demonstration that easily produces the result; it is a controlled attack trying to break the measurement chain.\n\nIts second provision states:\n\n> Synthetic data does not show the real company performance; it shows the audit method's ability to recover the known reality.\n\nIts third provision states:\n\n> Testing producer reality cannot be tested independently by adjudicators without being hidden from them.\n\nIts fourth provision states:\n\n> The prevalence of rare events and critical detection capability do not need to be measured from the same corpus.\n\nIts fifth provision states:\n\n> Testing data cannot be quietly adjusted according to the scoring formula, and the scoring formula cannot be quietly adjusted according to the testing results.\n\nIts sixth provision states:\n\n> The Apple.com anchor is a synthetic starting point used to test the method, not to produce real results about Apple.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to establish a synthetic testing architecture that will test the GEO-1000 protocol from the beginning to the public results card, with results predefined but hidden from evaluation teams. The testing should separately test each of the following systems:\n\n- Entity analysis\n\n- Population and sample allocation\n\n- Country and language panels\n\n- Prompt equivalence\n\n- AI System Register\n\n- Controlled and natural panel distinction\n\n- Synchronised wave\n\n- NOMOS Capture\n\n- Truth Pack\n\n- Claim extraction\n\n- Citation–claim matching\n\n- Omission detection\n\n- Critical and Major importance level\n\n- Response status\n\n- GEO-1000 distribution\n\n- Ten-component profile\n\n- Composite score\n\n- Confidence interval\n\n- Fairness\n\n- Upper bound of rare event\n\n- Score versioning\n\n- Public manifest\n\nBy the end of this section, the test design should be able to answer the following questions:\n\n> How was synthetic reality produced?\n\n> From whom and at which stages were the reality labels hidden?\n\n> Which population, language, AI product, and wave structure will the 30,000 main responses be generated from?\n\n> How frequently will Critical and Major cases appear in the main prevalence corpus?\n\n> Which challenge set will the Critical detection system additionally be tested with?\n\n> At what error level will the test be considered successful?\n\n> Which results will require the method to be corrected?\n\n> How will it be prevented for the test to make judgements about Apple, real AI providers, or real users?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nThe Apple.com Synthetic Benchmark should be a versioned method-validation test that makes no claim about the performance of a real company or real AI provider. It should consist entirely of synthetic entities, users, evidence, responses and system logs; lock Generator Truth before adjudication; conceal that truth from adjudicators; and measure how accurately the NOMOS audit and scoring chain recovers it. Synthetic status must be conspicuous. Claims about real-company performance must be prohibited. Generator Truth and the scoring formula must be locked before the main corpus is opened. Adjudicators must not see generator labels. The principal prevalence corpus must remain separate from the severity challenge corpus. Every test file must be marked synthetic. Independent teams must be able to reproduce the benchmark, and failed results must be published alongside successful ones.\n\nNOMOS should not make exceptions to its own standard.\n\n## 3. WHAT DOES THE TEST MEASURE?\n\nThe Apple.com Synthetic Test measures the following question:\n\n> How accurately can the NOMOS audit system reproduce a known synthetic reality and error distribution after the stages of capture, claim extraction, adjudication, gating, and scoring?\n\nThe object of measurement is not: Apple Inc. Apple products. Real AI models. Real user opinions. Real country or language performance. The object of measurement is:\n\n> The NOMOS methodology itself.\n\n## 4. WHAT DOES THE TEST NOT MEASURE?\n\nThis test cannot produce the following claims:\n\n- “Apple’s GEO score is X.”\n\n- “Apple is best represented in this AI product.”\n\n- “Y per cent of real users perceive Apple correctly.”\n\n- “The specified real AI provider produces critical errors.”\n\n- “The specified language’s real AI product is weak.”\n\n- “Apple NOMOS has passed the 950+ standard.”\n\n- “Apple supports this test.”\n\n- “Apple has participated in the test.”\n\n- “The real Apple price, licence, or service coverage is as follows.”\n\nEvery page in the public record must include the following provision:\n\n> This test is synthetic. It is not a current accuracy, appropriateness, or performance check of Apple Inc., apple.com, or any real AI provider.\n\n## 5. WHY THE APPLE.COM ANCHOR?\n\nApple.com is chosen as the anchor method for the following reasons: It has a short and clear domain format. It is suitable for testing the domain–corporate entity distinction. It requires resolving an entity not just from the company name, but from its digital presence. It allows synthetic modelling of different types of claims such as product, service, local scope, price, time, and resources. It is suitable for testing error types such as wrong category, wrong entity, wrong price parity, and false superiority on a single entity twin. When moving to a real test in the future, the burden on users to understand intent may be relatively limited. These are the design rationale for the test. It is not a verified current performance claim about the real Apple.\n\n## 6. SYNTHETIC ENTITY TWIN\n\nThe object under supervision in the test:\n\n### APPLE-SYNTH ENTITY TWIN\n\nwill be. Identity:\n\n### NGE-APPLE-SYNTH-TWIN-001\n\nThis entity:\n\n- is not the real Apple Inc.,\n\n- is not a real legal entity,\n\n- does not carry real product or price records,\n\nis created solely for testing methodology. Domain anchor: maintained as apple.com. However, in all Truth Pack records, the synthetic entity name is explicitly stated:\n\n#### Apple-SYNTH Global Technology Entity\n\n## 7. ONTOLOGY OF THE SYNTHETIC TWIN\n\nAPPLE-SYNTH carries the following synthetic entity graph:\n\n- APPLE-SYNTH-HOLDINGS-001 — main corporate entity\n\n- APPLE-SYNTH-DOMAIN-APPLE-COM-001 — domain anchor\n\n- APPLE-SYNTH-DEVICES-001 — device product family\n\n- APPLE-SYNTH-SOFTWARE-001 — software platforms\n\n- APPLE-SYNTH-SERVICES-001 — digital services\n\n- APPLE-SYNTH-PAYMENTS-SUB-001 — limited payment service subsidiary\n\n- APPLE-SYNTH-RETAIL-TR-001 — synthetic Turkey retail entity\n\n- APPLE-SYNTH-RETAIL-DE-001 — synthetic Germany retail entity\n\n- APPLE-SYNTH-FRANCHISE-X-001 — independent licensed store\n\n- APPLE-SYNTH-HISTORICAL-PARTNER-001 — expired historical partner\n\n- APPLE-SYNTH-UNRELATED-APPLE-001 — unrelated entity for name collision\n\nThis chart:\n\n- parent company,\n\n- affiliate,\n\n- product,\n\n- local entity,\n\n- franchise,\n\n- historical relationship,\n\n- unrelated name similarity\n\ntests the distinctions.\n\n## 8. THREE TEST MODES\n\n### BM-S0 — PURE SYNTHETIC METHOD TEST\n\nAll:\n\n- users,\n\n- AI products,\n\n- answers,\n\n- evidence,\n\n- screens,\n\n- scores\n\nare synthetic. This is the main mode of this section.\n\n### BM-S1 — SYNTHETIC REPLAY TEST\n\nPre-generated synthetic answers:\n\n- NOMOS Capture,\n\n- claim extraction,\n\n- adjudication,\n\n- scoring\n\nreplayed to the system like a live data stream. This mode tests software and human processes.\n\n### BM-L1 — FUTURE LIVE TESTING\n\nReal users and real AI products are used. Separately:\n\n- current Truth Pack,\n\n- legal assessment,\n\n- provider system records,\n\n- ethical and privacy protocol\n\nis required. This section is limited to Section 18: BM-S0 and BM-S1.\n\n## 9. TEST IDENTITY\n\nMain test identity:\n\n### NOMOS-APPLE-SYNTH-BENCH-0.9\n\nSub-records:\n\n- Entity: APPLE-SYNTH-TWIN-001\n\n- Population: SYNTH-POP-FRAME-001\n\n- Country allocation: SYNTH-COUNTRY-ALLOC-001\n\n- Language registry: SYNTH-LANGUAGE-REGISTRY-001\n\n- Prompt set: SYNTH-PROMPT-SET-001\n\n- Truth Pack: SYNTH-APPLE-TP-001\n\n- AI System Register: SYNTH-AI-REGISTRY-001\n\n- Wave plan: SYNTH-WAVE-PLAN-001\n\n- Capture schema: NOMOS-CAPTURE-SYNTH-0.9\n\n- Adjudication guide: SYNTH-ADJUDICATION-0.9\n\n- Scoring method: NOMOS-PUAN-0.9\n\n- Randomisation manifest: SYNTH-SEED-MANIFEST-001\n\nIf any of these identities change, a new testing version is required.\n\n## 10. MAIN RESEARCH QUESTIONS\n\nThe benchmark should answer ten research questions. Can it recover the domain–entity relationship? Can it distinguish supported, unsupported, contradicted and unresolved claims? Does it preserve the distinction between an official claim and verified reality? Do the Critical and Major gates achieve adequate recall without an excessive false-positive rate? Does it detect material omission as reliably as an explicit falsehood? Does the country- and language-fairness architecture recover the injected differences? Can it measure the Clean–Natural panel difference without making an unsupported causal claim? Can NOMOS Capture identify synthetic duplicates, tampering and temporal defects? How accurately do the GEO-1000 distribution and composite score recover Generator Truth? Do independent teams obtain materially comparable results from the same corpora?\n\n## 11. THE FOUR CORPORA OF THE TEST ARCHITECTURE\n\nThe test consists of four separate corpora.\n\n### 11.1. PRINCIPAL PREVALENCE CORPUS\n\nThe main corpus consists of 30,000 responses. The purpose:\n\n- The distribution of GEO-1000,\n\n- rare error rate,\n\n- AI product profiles,\n\n- wave's determination\n\nto test. This corpus maintains realistic error sparsity.\n\n### 11.2. IMPORTANCE LEVEL CHALLENGE CORPUS\n\nOversample Critical and Major error types. Purpose:\n\n- Critical recall,\n\n- Major classification,\n\n- false positive,\n\n- expert adjudication\n\nis to measure performance. This corpus does not count towards prevalence.\n\n### 11.3. CAPTURE INTEGRITY CORPUS\n\nIncludes:\n\n- Duplicate\n\n- Wrong prompt\n\n- Truncated answer\n\n- Late upload\n\n- Wrong timing\n\n- Hash mismatch\n\n- Synthetic tamper\n\n- Wrong product metadata\n\n- User interruption\n\n- Technical retry\n\nPurpose: To test NOMOS Capture verification. Only valid ones enter the denominator of the semantic score.\n\n### 11.4. ADJUDICATION GOLD CORPUS\n\nPre-resolved by expert panel:\n\n- claim boundary,\n\n- entity,\n\n- attribution,\n\n- modality,\n\n- citation,\n\n- omission,\n\n- degree of importance\n\ncarries cases. Purpose:\n\n- adjudicator calibration,\n\n- drift detection,\n\nexternal team comparison. It does not enter the main score denominator.\n\n## 12. 30,000 RESPONSE MAIN CORPUS\n\nMain design:\n\n10 synthetic AI products × 1,000 unique users × 3 waves = 30,000 responses\n\nEach response:\n\n- to a different synthetic human unit,\n\n- single AI product,\n\n- single wave,\n\n- single locked Core Mirror Prompt\n\nbelongs. In the main corpus, the same synthetic person:\n\n- is not reused in another AI product,\n\n- or in another wave\n\nagain. In total:\n\n#### there are 30,000 unique synthetic human units\n\nfound. This decision:\n\n- panel conditioning,\n\n- removes response recall,\n\n- the multiple product dependency of the same person\n\nfrom the main corpus.\n\n## 13. MATCHED POPULATION TWINS\n\nAI products are tested with different human units. However, if user profiles are completely different, products cannot be compared. Therefore:\n\n#### Matched Synthetic Population Twins\n\nused. Each AI product has different individual identities. However, a sample of 1,000 people for each product:\n\n- country,\n\n- language,\n\n- age band,\n\n- Frequency of AI usage,\n\n- plan eligibility,\n\n- device,\n\n- accessibility,\n\n- brand awareness\n\ncarries the same target distribution in terms of care. This structure matches the population composition without reusing the same person.\n\n## 14. PROMPT OF THE MAIN CORPUS\n\nAll 30,000 main responses carry this canonical intent:\n\n> Link the domain anchor to the main corporate entity and explain the core activities of this entity.\n\nTurkish candidate text:\n\n> “What main corporate entity is Apple.com associated with and what are the primary activities of this entity?”\n\nEnglish candidate text:\n\n> “Which principal corporate entity is associated with apple.com, and what are that entity’s primary activities?”\n\nOther language versions:\n\n- in this section it is not presented as a real-world translation,\n\n- PROMPT-L01 to PROMPT-L36 are retained as synthetic language objects,\n\nsemantic equivalence labels are predetermined. This approach: prevents the claim of fake translation accuracy without real linguistic expertise, yet still allows the test to compute language fairness.\n\n## 15. WHY ONLY ONE PROMPT IN THE MAIN CORPUS?\n\nThe question from Founder GEO-1000 is:\n\n> When the same AI product is asked at the same time and with the same intent, how many users truly see the fundamental representation of existence?\n\nFor this reason, the Principal Prevalence Corpus:\n\n- uses a single Core Mirror Prompt,\n\n- a single prompt version,\n\n- three repetition waves\n\nEvidence, Recommendation, Comparative, and Boundary families: are tested in a separate Diagnostic Annex, not mixed into the main 30,000 response denominator. This distinction is important. Because the single Core Prompt:\n\n#### produces Core GEO-1000 results\n\nAlone:\n\n#### Full NOMOS conformity\n\nis not produced.\n\n## 16. DIAGNOSTIC ANNEX\n\nApart from the main 30,000 responses, the following synthetic diagnostic corpus is designed:\n\nThese 6,000 cases: are not the main user prevalence, but are the prompt family and component validation corpus. Together with the main 30,000, the testing universe can carry 36,000 synthetic response cases. However, the outcome of the title of Section 18:\n\n#### 30,000 main Population Panel responses\n\nwill remain.\n\n## 17. SYNTHETIC POPULATION FRAMEWORK\n\nThe trial does not claim to mimic the real-world population count. SYNTH-POP-FRAME-001:\n\n- long-tail country distribution,\n\n- multilingual user population,\n\n- difference between large and small markets,\n\n- low-resource language cells,\n\n- different AI usage frequencies\n\nis the synthetic framework it produces. This framework cannot be published as the current population ranking of real countries. Future live trial:\n\n- dated,\n\n- authorised,\n\n- versioned real population data\n\nmust be used.\n\n## 18. SYNTHETIC COUNTRY UNIVERSE\n\nSynthetic universe: It carries 193 eligible country or jurisdiction codes. The codes are:\n\n### C001–C193\n\nThey are in the form of. These do not represent the actual country performance. Country shares:\n\n- several large layers,\n\n- medium-sized layers,\n\n- a long small country tail\n\nare predetermined to form.\n\n## 19. COUNTRY ALLOCATION\n\nThe synthetic population share of country c: let it be p_c. The main objective:\n\nn_c* = 1,000 p_c\n\nis in the form of. The initial allocation:\n\nn_c^0 = ⌊n_c*⌋\n\nis made as. Remaining slots: distributed using the largest decimal remainder method. Result:\n\nΣ_c n_c = 1,000\n\nIt must be. Small countries that do not collect any observations: They are not shown as if they exist in the Population Panel, they enter a separate Country Observer cycle.\n\n## 20. Country Observer ANNEX\n\nTo test all eligible country codes at least once: a synthetic Country Observer Annex of 193 countries is established. These observations:\n\n- to the country score,\n\n- to its main 1,000 stakeholders\n\ndoes not enter. Purpose:\n\n- access,\n\n- language,\n\n- entity collision,\n\n- early Critical signal\n\nIt is a test.\n\n## 21. LANGUAGE UNIVERSE\n\nMain synthetic language registry:\n\n- 24 Population Panel language\n\n- 12 Language Fairness Observer language\n\nis as follows:\n\n36 synthetic language–locale cells\n\nare carried. Codes:\n\n### L01–L36\n\nare in this form. Example texts in Turkish and English can be published for L01 and L02. Other language objects: carry synthetic identity to avoid claiming real language performance.\n\n## 22. LANGUAGE ALLOCATION\n\nUser language allocation:\n\n- not according to the official language of the country,\n\n- but according to the primary task language in which the user will naturally perform the testing task\n\ndone. Multilingual user: remains a single person in the main population weight, does not become multiple full persons. Language fairness Annex: oversamples low-count L25–L36 cells, reweights back to true synthetic language shares for the global score.\n\n## 23. PROMPT EQUIVALENCE INJECTION\n\nThe test should include only correct translations. The Language Equivalence Annex includes the following cases:\n\n- Correct semantic equivalence\n\n- Translation with confidence task added\n\n- Translation turning into a recommendation\n\n- Entity anchor drift\n\n- Geography drift\n\n- Time drift\n\n- \"World leader\" presupposition\n\n- Adding source order\n\n- Excessive formality\n\n- Unusable machine translation\n\nSome of these cases:\n\n- AI language performance,\n\n- prompt tool defect\n\ntests the distinction. Case with prompt equivalence defect: cannot be directly loaded into the real AI fairness score.\n\n## 24. SYNTHETIC TRUTH PACK\n\nSYNTH-APPLE-TP-001 contains at least the following fields:\n\nTruth Pack also:\n\n- 12 Official Claim Only,\n\n- 10 Contradicted,\n\n- 8 Unresolved,\n\n- 6 Historical,\n\n- 4 Restrictedly Verified\n\ngenerates status. This distribution is used to test all adjudication statuses.\n\n## 25. CORE PROVISIONS OF THE SYNTHETIC TRUTH PACK\n\nThe following examples are entirely synthetic: apple.com, associated with APPLE-SYNTH-HOLDINGS-001. The main activities are in the categories of devices, software, and digital services. Product and service availability varies by market. Price and tax conditions are not the same in all countries. The main entity is not a general-purpose bank. The main entity does not provide legal or healthcare services. The limited local authority of the payment subsidiary does not grant the main entity universal banking authority. The term 'the world's most innovative company' is a synthetic official self-positioning; it is not a fact of independent superiority. The basis for the claim of '99 per cent satisfaction' has not been verified. Some historical partnerships ended on the reference date. Some products are available only in certain synthetic markets. There is no 100 per cent outcome guarantee for all projects or products.\n\nThere is no endorsement from a synthetic public institution or university. Local registration of a franchise is not the direct operation of the main entity. Certain customer relationships have been verified with limited evidence but are not discoverable by the public. These provisions are not claims about the real Apple.\n\n## 26. EVIDENCE FAMILIES\n\nSynthetic Truth Pack carries the following evidence families:\n\n### EF-01 — Canonical first-party product pages\n\n### EF-02 — Synthetic company registry\n\n### EF-03 — Synthetic local price records\n\n### EF-04 — Synthetic regulatory affiliate record\n\n### EF-05 — Synthetic partner agreements\n\n### EF-06 — Synthetic independent editorial review\n\n### EF-07 — Synthetic transactional data\n\n### EF-08 — Synthetic user records\n\n### EF-09 — Synthetic historical archive\n\n### EF-10 — Synthetic counter-evidence registry\n\nThere are also four evidence laundering chains:\n\n### EL-01 — Self-declaration → sponsored news → blog → AI summary\n\n### EL-02 — Company award → multi-copy site → fake consensus\n\n### EL-03 — Employee comment → presented like university endorsement\n\n### EL-04 — Affiliate licence → transfer as main entity authority\n\n## 27. SECRET PRODUCER REALITY\n\nFor each synthetic response:\n\n#### Generator Truth Record\n\nis created. This record carries the following: Which claims were generated? Which claim is correct? Which is partially correct? Which belongs to a false entity? Which is outdated? Which is unsupported? Which is Critical or Major? Which omission is intentional? Which reference is fake or incorrectly scoped? What is the actual RP status of the response? Which components should be affected? This record:\n\n- is created before data is produced,\n\n- is hashed,\n\nis sealed until peer review is complete. Adjudicators cannot see the Generator Truth.\n\n## 28. THE LOCK OF THE PRODUCER'S TRUTH\n\nThe Generator Truth package:\n\n- hash,\n\n- timestamp,\n\n- random seed manifest,\n\n- production code version,\n\n- template version\n\nmust be carried. After the score result is seen:\n\n- the label cannot be changed,\n\n- the importance level cannot be lowered,\n\nthe true/false distribution cannot be readjusted. If a real production error is found:\n\n- new test version,\n\n- a record of impact on the old corpus\n\nis created.\n\n## 29. SYNTHETIC AI PRODUCTS\n\nThe test uses an example of synthetic AI products independent of the ten providers:\n\n### SYNTH-AI-01\n\n### SYNTH-AI-02\n\n### SYNTH-AI-03\n\n### SYNTH-AI-04\n\n### SYNTH-AI-05\n\n### SYNTH-AI-06\n\n### SYNTH-AI-07\n\n### SYNTH-AI-08\n\n### SYNTH-AI-09\n\n### SYNTH-AI-10\n\nThese are not imitations or code names of real AI providers. Each synthetic product carries a different failure mode profile.\n\n## 30. SYNTHETIC AI PROFILES\n\nThis profile distribution tests that NOMOS does not reward only the highest average.\n\n## 31. PROFILE PARAMETERS\n\nEach synthetic AI product carries these latent parameters:\n\n- Core entity accuracy\n\n- Factual support\n\n- Scope overreach\n\n- Temporal lag\n\n- Attribution loss\n\n- Citation mismatch\n\n- Refusal probability\n\n- No-response probability\n\n- Low-resource language penalty\n\n- Clean–Natural gap\n\n- Wave drift\n\n- Critical-event probability\n\n- Major-event probability\n\n- Verbosity\n\n- Self-correction probability\n\n- Hallucinated citation probability\n\nParameters: locked before the main corpus is created, cannot be changed after the final result is seen.\n\n## 32. MAIN CORPUS ERROR BASE RATES\n\nThe Principal Prevalence Corpus uses realistic sparse error logic. Candidate global synthetic rates:\n\n- Full/Advisory Pass: high majority\n\n- Conditional Pass: 2–15 per cent depending on product profile\n\n- Major Fail: 0–8 per cent\n\n- Confirmed Critical: 0–0.5 per cent\n\n- Unresolved: 0–3 per cent\n\n- No Usable Response: 0–8 per cent\n\n- Not Ratable: after capture corpus, 0–2 per cent\n\nThese rates differ for each product. They are not predictions about the real AI market. They are synthetic stress distributions.\n\n## 33. DEGREE OF IMPORTANCE CHALLENGE CORPUS\n\nSince critical errors are rare in realistic prevalence, sufficient validation cases are needed for each gate. The Challenge Corpus should have the following structure:\n\nAlso:\n\n- 1,200 Major,\n\n- 600 difficult Moderate,\n\n- 600 negative controls resembling Critical but not Critical\n\ncases can be generated. Total Importance level Challenge: there would be 3,600 cases. This corpus cannot enter the main prevalence score.\n\n## 34. NEGATIVE CRITICAL CONTROLS\n\nThe following types of cases should be present to test the Critical false-positive rate:\n\n- Correct trademark but missing legal suffix\n\n- Small price rounding difference\n\n- Small deviation in historical award year\n\n- Citation format error but substantive claim correct\n\n- Appropriate caution\n\n- Limited but harmless omission\n\n- Negative comment clearly presented as opinion by the user\n\n- Slight uncertainty in the historical scope of real partnership\n\n- Correct attribution of the company's self-declaration\n\nAdjudicators must not classify every forceful term as Critical.\n\n## 35. ERROR INJECTION MATRIX\n\nError vector for each response:\n\nH_i = (E_i, F_i, S_i, T_i, A_i, M_i, C_i, O_i, R_i)\n\nRecorded as follows. Here:\n\n- Ei: existence error\n\n- Fi: factual error\n\n- Si: scope error\n\n- Ti: temporal error\n\n- Ai: attribution error\n\n- Mi: modality error\n\n- Ci: citation error\n\n- Oi: omission\n\n- Ri: relevance/refusal error\n\nErrors do not have to be independent. Example: Transferring the subsidiary licence to the main entity can simultaneously:\n\n- wrong entity,\n\n- overreach,\n\n- produce false authority,\n\n- Critical advice\n\nThis can occur. These cases are connected as a Root Finding Cluster.\n\n## 36. RESPONSE GENERATION\n\nSynthetic response generation has three stages.\n\n### 36.1. Semantic Plan\n\nAccording to the Generator Truth Record:\n\n- active true claims,\n\n- active false claims,\n\n- omissions,\n\n- referencing behaviour,\n\n- response status\n\nis determined.\n\n### 36.2. Linguistic Realisation\n\nSame semantic plan:\n\n- short,\n\n- long,\n\n- indirect,\n\n- modal,\n\n- attributed,\n\n- self-correcting,\n\n- contradictory\n\nwritten on different surfaces.\n\n### 36.3. Interface and Capture Realisation\n\nResponse:\n\n- mock web,\n\n- mock mobile,\n\n- mock citation panel,\n\n- mock error screen\n\nis rendered inside. All visuals:\n\n### SYNTHETIC CINEMA — NOT LIVE AI OUTPUT\n\nmust carry the watermark.\n\n## 37. PREVENT TEMPLATE MEMORISATION\n\nEach claim should not be produced using only a single sentence template. An example fake superiority claim can be produced in the following forms:\n\n- \"It is the most innovative company in the world.\"\n\n- \"According to most independent sources, it is the undisputed leader in the sector.\"\n\n- \"It is considered the most reliable option across the market.\"\n\n- “It has been confirmed that it ranks first on a global scale.”\n\n- “It can be said that it is superior to all of its competitors.”\n\nThe adjudication system should evaluate meaning, not a word list.\n\n## 38. IMPLICIT ERROR GENERATION\n\nAt least one-third of the Challenge Corpus should carry the error not as an explicit sentence but as:\n\n- presupposition,\n\n- implicature,\n\n- modality,\n\n- attribution omission,\n\n- recommendation,\n\n- entity coreference\n\nExample: “Offering limited payment services indicates that the parent company is subject to banking regulations and is licensed.” Here the mistake is:\n\n- not directly in the word “bank,”\n\n- transferred authority derived\n\nhas been established.\n\n## 39. CITATION PRODUCTION\n\nSynthetic citations include the following classes:\n\n- Directly supporting\n\n- Supporting only attribution\n\n- Partially supporting\n\n- Belonging to another country\n\n- Belonging to another entity\n\n- Old\n\n- Irrelevant\n\n- Contradictory content\n\n- Nonexistent\n\n- Copy of the same root source\n\n- Presenting limited evidence as a public source\n\n- Producing fake university or public endorsement\n\nCitation URLs should not be redirected to real domain names. Example: https://evidence.synthetic.example/EF-004 can be used.\n\n## 40. REFERENCE GAP INJECTION\n\nSome responses of the test generate new claims that are not found in Truth Pack but:\n\n- can be forced into the categories of\n\ntrue,\n\n- false,\n\n- cannot be evaluated\n\nThe adjudicator, as the correct behaviour:\n\n### REFERENCE GAP\n\nshould open. The test measures these two incorrect behaviours:\n\n- Automatically counting a reference gap as incorrect\n\n- Automatically counting a reference gap as correct\n\n## 41. OMISSION INJECTION\n\nOmission cases are generated at three levels:\n\n- Optional detail omission\n\n- Required element omission\n\n- Material boundary omission\n\nExample: Prompt: “Does the company provide payment services in Germany and what limits apply?” Answer: “The company provides payment services.” Truth Pack:\n\n- only separate subsidiary,\n\n- only limited user group,\n\n- only certain jurisdiction\n\nif it shows:\n\n- entity,\n\n- scope,\n\n- boundary omission\n\ncan be evaluated together.\n\n## 42. CASES OF SELF-CORRECTION\n\nThe test includes these answer formats:\n\n- False claim, then explicit retraction\n\n- Wrong claim, then vague softening\n\n- Correct claim, then wrong contradiction\n\n- Self-correction after citation\n\n- Correction in the same response without user follow-up message\n\n- Only adding disclaimer while preserving wrong claim\n\nPurpose:\n\n- real self-correction,\n\n- fake correction,\n\n- unresolved internal contradiction\n\nto test the distinction.\n\n## 43. CONTROLLED–NATURAL PANEL MODULE\n\n1,000 cases of the Diagnostic Annex: carries Clean and Natural twins of the same synthetic user profiles. In the Natural condition:\n\n- memory,\n\n- special instructions,\n\n- previous brand opinion,\n\n- different plan,\n\n- tool usage\n\nis injected. Generator Truth preserves this distinction:\n\n- Formally appropriate change for the user\n\n- Legitimate advice difference\n\n- Material reality drift\n\n- Instruction in favour of the brand\n\n- Instruction against the brand\n\n- Restricted information leak\n\nThe USI component is tested with this module.\n\n## 44th WAVE ARCHITECTURE\n\nThe main corpus carries three waves:\n\nIn each wave:\n\n- new 10,000 synthetic users,\n\n- same Core Mirror Prompt version,\n\n- same base Truth Pack ontology\n\nIt is used. Controlled drift is injected in some products.\n\n## 45. WAVE DRIFT SCENARIOS\n\nFor synthetic AI products, the following changes are injected:\n\n- SYNTH-AI-02: Increase in verbosity\n\n- SYNTH-AI-03: L25–L36 language drop\n\n- SYNTH-AI-04: Growth of natural panel difference\n\n- SYNTH-AI-05: Increase in reference laundering\n\n- SYNTH-AI-07: Old price and product availability\n\n- SYNTH-AI-10: Singular Critical in W2, correction in W3\n\nPurpose:\n\n### STR,\n\nis to test current status, historical incident, and correction records.\n\n## 46. IN-WAVE SYSTEM CHANGE\n\nIn the middle of a synthetic AI product in W2:\n\n- model label,\n\n- reference view,\n\n- web mode\n\nis changed. The correct NOMOS behaviour:\n\n- to separate the wave into a sub-wave,\n\n- to generate a mixed-system-state warning\n\nshould be. Incorrect behaviour: to count the entire W2 as a single fixed product score.\n\n## 47. EXTERNAL EVENT INJECTION\n\nDuring W3, a new event about the synthetic entity is generated:\n\n- price change,\n\n- product termination,\n\n- local service opening,\n\n- licence affiliate change\n\nlike. Truth Pack: It is divided into W3A and W3B versions. It is tested whether adjudicators link all answers to a single reference time.\n\n## 48. SYNTHETIC NOMOS CAPTURE\n\nA synthetic evidence bundle is created for each main response:\n\n- Observation manifest\n\n- Prompt artefact\n\n- Raw response artefact\n\n- Mock screenshot\n\n- System-state artefact\n\n- Time vector\n\n- Attempt log\n\n- Integrity manifest\n\n- Privacy manifest\n\n- Validation record\n\nValid packages: enter the real test distribution. Defective packages: classified according to the Capture Integrity Corpus.\n\n## 49. CAPTURE ERROR CLASSES\n\nThe test produces the following types of errors:\n\n- Exact duplicate\n\n- Near-duplicate screenshot\n\n- Prompt mismatch\n\n- Prompt truncation\n\n- Response truncation\n\n- Missing response end\n\n- Wrong AI product metadata\n\n- Outside-wave timestamp\n\n- Device clock offset\n\n- Delayed upload\n\n- Hash mismatch\n\n- Tamper-before-hash\n\n- Tamper-after-hash\n\n- User regeneration\n\n- Technical retry\n\n- User interruption\n\n- System transformation\n\n- Redaction on raw file\n\n- Valid accessibility alternative\n\n- False accessibility claim\n\nPurpose: to separate capture validity from semantic accuracy.\n\n## 50. TEST WATERMARK\n\nAll visuals and public response records must carry the following visible mark: SYNTHETIC NOMOS TEST / NOT A LIVE AI RESPONSE / NOT AN APPLE OR PROVIDER PERFORMANCE CLAIM Watermark: should not cover the claim text, should not interfere with the adjudicators' comprehension evaluation. In the blind copy used for adjudication, the watermark indicates it is synthetic, it does not show the Generator Truth label.\n\n## 51. RANDOMISATION\n\nThe test generation uses a versioned pseudorandom method. Each random decision:\n\n- test seed,\n\n- sub-module seed,\n\n- template selection,\n\n- error injection,\n\n- user assignment,\n\n- slot assignment\n\nmust be reproducible. Random seed: hashed before the main corpus generation, cannot be changed after the result.\n\n## 52. DISTINCTION BETWEEN RANDOMNESS AND ARBITRARINESS\n\nRandomisation: ensures allocation independent of the outcome. Arbitrariness: means the producer selects the response they want. Without a randomisation manifest: the statement “We generated it randomly.” is not sufficient.\n\n## 53. DATA LEAKAGE\n\nTest leakage can occur in the following ways: The Generator Truth is shown to the adjudicator. The response template name explains the importance level. The file name should be critical-licence-001.png. The UI colour indicates the type of error. The reference code carries a direct true/false tag. The adjudicator has seen the same case in the training set. The AI assistant adjudicator accesses the generator code. The score sees the developer holdout results. All these channels must be closed.\n\n## 54. FILE NAMES\n\nIn blind review packages, the file name:\n\n### OBS-000184\n\nshould be neutral. The following names are prohibited:\n\n- critical-case-12\n\n- wrong-entity\n\n- good-answer\n\n- low-resource-fail\n\n- false-citation\n\n## 55. SEPARATE ROLES\n\nTesting should be managed with at least the following roles:\n\n- Synthetic Population Architect\n\n- Truth Pack Designer\n\n- Response Generator\n\n- Seed Custodian\n\n- Leakage Auditor\n\n- Capture Corpus Designer\n\n- Claim Extraction Team\n\n- Adjudication Team\n\n- Score Methods Team\n\n- Independent Validation Team\n\n- Public Release Custodian\n\n- Accountable Human Testing Owner\n\nRoles can be combined in the small pilot. However:\n\n- The person who knows the Generator Truth,\n\n- is the sole decision-maker in the final blind adjudication\n\nNo person who knows the Generator Truth may be the sole decision-maker in final blind adjudication.\n\n## 56. TEST DATA SECTIONS\n\n### 56.1. Public Calibration Set\n\nUsed for adjudicator training and codebook testing. Generator Truth is open. It does not enter the main evaluation.\n\n### 56.2. Development Set\n\nUsed for method and software development. Labels are open to a limited team.\n\n### 56.3. Sealed Validation Set\n\nOpened after adjudication and scoring system are locked. Generator Truth is sealed. Produces main method validation.\n\n### 56.4. Renewal Holdout Set\n\nIt is retained to test benchmark-specific overfitting. It is hashed in advance and used for version renewal. It may not remain hidden indefinitely; an explanation and publication timetable must be provided.\n\n## 57. MAIN 30,000 CORPUS LEAK RULE\n\nMain Principal Prevalence Corpus:\n\n- score method without locking 0.9,\n\n- claim codebook without locking,\n\n- adjudication roles without being determined\n\nshould not be opened. After locking: Response and evidence packages are opened to adjudicators. Generator Truth remains closed. Adjudication ends. Scores are calculated. Result manifest is locked. Generator Truth is opened. Recovery analysis is performed.\n\n## 58. GROUND-TRUTH RECOVERY\n\nThe testing success of NOMOS is not producing a high score. Success:\n\n> is having a low difference between the Generator Truth and the results reproduced by NOMOS.\n\nExample: Generator Truth:\n\n- Strict Pass: 942\n\n- Conditional: 38\n\n- Major: 15\n\n- Critical: 2\n\n- Unresolved: 3\n\nNOMOS recovery:\n\n- Strict Pass: 940\n\n- Conditional: 40\n\n- Major: 15\n\n- Critical: 2\n\n- Unresolved: 3\n\nthen the method can be strong. The NOMOS score may be high or low. What is important is the correct recovery.\n\n## 59. CLAIM BOUNDARY RECOVERY\n\nGenerator claim atoms:\n\n### G\n\nAtoms produced by adjudicators:\n\n### H\n\nLet it be. Claim boundary precision:\n\nP = correctly matched extracted atoms / all extracted atoms\n\nRecall:\n\nR = correctly matched extracted atoms / generator atoms\n\nF1:\n\n### F1 = 2PR / (P + R)\n\ncan be calculated as. Not just the number of atoms:\n\n- source span,\n\n- normalised proposition,\n\n- entity,\n\n- qualifiers\n\nmatching should be taken into consideration.\n\n## 60. ATOMIC DECISION RECOVERY\n\nSeparate measurement is required for each dimension: Entity accuracy\n\nFactual status macro-F1\n\nScope macro-F1\n\nTime macro-F1\n\nAttribution macro-F1\n\nModality macro-F1\n\nReference matching F1\n\nOmission F1\n\nCombined label accuracy Unresolved/reference-gap accuracy Overall accuracy alone is not sufficient. Rare Critical class can be lost within high overall accuracy.\n\n## 61. CRITICAL RECALL\n\nNumber of Confirmed Critical cases in Generator Truth:\n\n### NC\n\nCorrectly found Critical:\n\n### TPC\n\nlet it be.\n\nRecall_C = TP_C/N_C\n\nis as follows. False Critical Rate:\n\nFCR = FP_C/N_non-critical\n\ncan be calculated as. Test:\n\n- high Critical recall,\n\n- low false Critical rate\n\nshould be sought together.\n\n## 62. RESPONSE STATUS RECOVERY\n\nFor each RP class:\n\n- precision,\n\n- recall,\n\n### F1,\n\nconfusion matrix should be produced. Especially the following confusions should be visible:\n\n- RP-3 and RP-4\n\n- RP-4 and RP-5\n\n- RP-6 and RP-4\n\n- RP-7 and RP-1\n\n- RP-8 and RP-4\n\n## 63. GEO-1000 DISTRIBUTION ERROR\n\nFor status k, generator user equivalent: DkG NOMOS recovery: DkN. Absolute error:\n\nE_k = |D_k^N − D_k^G|\n\nTotal normalised distribution error:\n\nE_D = (1/2,000)Σ_k E_k\n\ncan be calculated. This value is in the range 0–1. Zero is full recovery.\n\n## 64. COMPONENT RECOVERY\n\nGenerator component score: Let SjG NOMOS score be SjN. Component absolute error:\n\nAE_j = |S_j^N − S_j^G|\n\nComponent MAE:\n\nMAE_component = (1/10)Σ_{j=1}^{10} AE_j\n\nis as follows. It should also be shown which component is harder to recover.\n\n## 65. COMPOSITE SCORE RECOVERY\n\n### AE_NOMOS = |NOMOS_N − NOMOS_G|\n\nis calculated as such. Small composite error: not sufficient if the component distribution is wrong. Two component errors may have arithmetically cancelled each other. Therefore:\n\n- composite error,\n\n- component error\n\nshould be published together.\n\n## 66. GATE RECOVERY\n\nThe test separately examines the following gate results:\n\n- Any Critical\n\n- Critical Hold\n\n- Major Fail\n\n- Conditional\n\n- Full Candidate\n\n- Not Ratable\n\n- Unresolved\n\nIn gate recovery: accidentally making a Critical Hold a 950+ Candidate is a very severe test failure.\n\n## 67. FAIRNESS RECOVERY\n\nSynthetic generator:\n\n- pre-determines the language floor,\n\n- the language difference,\n\n- the country base.\n\n- coverage\n\nNOMOS:\n\n### LGF,\n\nlanguage floor, country floor, disparity, must reproduce coverage factor results. Fairness recovery:\n\n- should be measured\n\n- not only by the composite LGF difference,\n\nbut also by its subcomponents.\n\n## 68. DISTINCTION BETWEEN LANGUAGE DEFECT AND PROMPT DEFECT\n\nThe test produces some low language scores due to AI profile penalty. Some others are produced due to prompt equivalence failure. The correct NOMOS behaviour:\n\n- is to classify the first as AI language finding,\n\n- and the second as Prompt Constitution failure\n\nThe two results should not convert into the same fairness penalty.\n\n## 69. CLEAN–NATURAL RECOVERY\n\nGenerator:\n\n- Determines in advance the values for Clean score,\n\n- Natural score,\n\n- panel gap,\n\n- material drift due to personalisation\n\nNOMOS:\n\n- only the relationship,\n\n- the causality boundary,\n\n- the USI component\n\nmust be produced correctly. Every case with a Clean–Natural difference should not be interpreted as \"memory caused it.\"\n\n## 70. CAPTURE RECOVERY\n\nFor the Capture Integrity Corpus:\n\n- Valid\n\n- Conditionally Valid\n\n- Outside Wave\n\n- Prompt Mismatch\n\n- Duplicate\n\n- Tamper Suspected\n\n- Fabricated\n\n- Withdrawn\n\nthe confusion matrix of the statuses should be published. It should not affect the semantic correctness capture decision.\n\n## 71. CONFIDENCE INTERVAL RECOVERY\n\nSynthetic testing can run the same population production process many times. Let the true generator parameter be θ. The empirical coverage of the 95 per cent confidence intervals:\n\nCoverage = Number of intervals containing θ / Total repetitions\n\nis calculated as. The 95 per cent method: should produce approximately 95 per cent coverage. Excessively narrow or overly wide intervals are evaluated separately.\n\n## 72. RARE EVENT TEST\n\nSynthetic scenarios with zero, one, two, and five critical events should be generated separately. NOMOS should correctly display the following fields:\n\n- Observed event count\n\n- Weighted rate\n\n- One-sided upper bound\n\n- Incident state\n\n- Scope\n\n- Confirmatory sampling requirement\n\nZero incident: there should be zero risk. One incident: there should not be 100% system failure.\n\n## 73. CANDIDATE TEST ACCEPTANCE THRESHOLDS\n\n### CANDIDATE THRESHOLDS — CANDIDATE THRESHOLDS\n\nThe following thresholds are candidates before pilot and independent review:\n\nThese thresholds are not a final standard. The purpose of the test is also to show whether these thresholds are realistic.\n\n## 74. ZERO TOLERANCE TEST ERRORS\n\nThe following situations can stop a test release even in a single case:\n\n- Generator Truth leakage\n\n- Changing the main corpus label after the result\n\n- Continuing the release despite knowing a rule systematically misses critical cases\n\n- Presentation like real Apple or real AI provider performance\n\n- Publishing a synthetic screenshot as if it were a live response\n\n- Removing a low-scoring system or language afterward\n\n- Changing the test seed\n\n- Silently adjusting the scoring formula based on the Holdout result\n\n- The raw corpus not matching the published manifest\n\n- Storage of failed test result\n\n## 75. TEST SUCCESS LEVELS\n\n### BQ-0 — NOT EXECUTED\n\nThe test is only at the design stage.\n\n### BQ-1 — GENERATION VALIDATED\n\nThe synthetic corpus has been reproduced. Adjudication and score recovery have not yet been done.\n\n### BQ-2 — PROCESS LINE DRY RUN\n\nCapture, claim extraction, and score chain have been run once. There is no independent verification.\n\n### BQ-3 — SEALED VALIDATION PASSED\n\nCandidate thresholds have been met in the sealed validation corpus.\n\n### BQ-4 — INDEPENDENT REPRODUCTION\n\nThe independent team produced comparable results from the same corpus.\n\n### BQ-5 — EXTERNAL MULTI-SITE VALIDATION\n\nMultiple independent institutions repeated the testing in different environments. Before the NOMOS scoring methodology becomes a public standard candidate for real companies, at least:\n\n### BQ-4\n\nshould target that level.\n\n## 76. METAMORPHIC TESTS\n\nTesting should not only evaluate fixed answers. Results should be consistent in transformations that preserve the same meaning.\n\n### 76.1. Paraphrase Test\n\nThe same atomic claim is written with a different sentence. The peer review result should not change.\n\n### 76.2. Sentence Order Test\n\nThe order of the claims changes. The truth status should not change except for Centrality.\n\n### 76.3. Citation Location Test\n\nThe citation changes its place within the paragraph while clearly staying attached to the same claim. The citation support should remain the same.\n\n### 76.4. Attribution Test\n\nThe phrase \"The company says\" is removed. EPI and attribution result should change.\n\n### 76.5. Modality Test\n\nThe phrase \"It is certain\" changes to \"It is likely.\" If Truth Pack is uncertain, the modality result may improve.\n\n### 76.6. Entity Test\n\nThe same true claim is transferred to an incorrectly attached organisation. The entity and scope result should be disrupted.\n\n### 76.7. Time Test\n\nThe historical claim is updated. TLA must drop.\n\n### 76.8. Boundary Test\n\nThe boundary sentence that changes the user's decision is removed. SBI and, if necessary, the Critical gate should be changed.\n\n## 77. CONTRADICTORY TESTS\n\nThe test should include the following attacks:\n\n- Very fluent but incorrect answer\n\n- Answer with many references but evidence-laundered\n\n- Single Critical among high number of correct atoms\n\n- Evasive answer that seems like proper caution\n\n- Short but sufficient answer\n\n- Long answer full of incidental errors\n\n- Severely wrong answer with self-correction\n\n- Correct number, false claim of independence\n\n- Real affiliate licence, false main entity\n\n- Current price, wrong country\n\n- Correct customer relationship, wrong time\n\n- Answer that makes user review a demographic fact\n\n- Answer presenting the complaint as a definite crime\n\n- Answer portraying limited evidence as a public citation\n\n- Answer claiming the opposite of what the real citation says\n\n## TRUTH PACK FOR THE 78TH EXAM ITSELF\n\nTesting outside Apple-SYNTH Truth Pack should have its own methodological Truth Pack. This package carries the following: How many responses were produced? How many in the main corpus? How many in the challenge corpus? How many Critical and Major generator tags are there? What was the seed? Which code version was used? Which files were excluded? What did the adjudicators see? Which tags were opened when? When was the scoring method locked? Which recovery results were obtained? Which thresholds were not met? Testing: it cannot declare method validation without creating its own results for Truth Pack.\n\n## 79. PUBLIC TEST RESULT CARD\n\nThe public card must include at least the following fields:\n\n#### Identity\n\nTest name Version Synthetic status Used anchor Real company/performance non-claim statement\n\n#### Corpus\n\nMain response: 30,000 Diagnostic Annex Importance level Challenge Capture Integrity Gold Adjudication\n\n#### Generator Truth\n\nLock time Seed manifest Label distribution Leakage audit\n\n#### Recovery\n\nClaim boundary F1\n\nCritical recall Critical false-positive\n\nRP macro-F1\n\nComponent MAE\n\nComposite error Fairness error CI coverage Capture accuracy\n\n#### Result\n\nBQ level Passed thresholds Remaining issues Retest requirement Independent repeat status This card should not rank any real AI product.\n\n## 80. MANDATORY NORMATIVE PROVISIONS\n\n**CH17-N01**\n\nApple.com The object of measurement for synthetic testing should be the NOMOS methodology; it should not be real Apple or real AI provider performance.\n\n**CH17-N02**\n\nAll public materials of the test must carry synthetic and non-claim declarations.\n\n**CH17-N03**\n\nThe synthetic entity must carry an APPLE-SYNTH identity clearly separated from the Apple.com anchor.\n\n**CH17-N04**\n\nSynthetic Truth Pack records cannot be used as real Apple reality.\n\n**CH17-N05**\n\nSynthetic AI product identities cannot be presented as implicit codes or replicas of real providers.\n\n**CH17-N06**\n\nThe Main Principal Prevalence Corpus, Importance Degree Challenge Corpus, Capture Integrity Corpus, and Adjudication Gold Corpus must be kept separate.\n\n**CH17-N07**\n\nImportance Degree Challenge cases cannot be added to the actual prevalence rate.\n\n**CH17-N08**\n\nCapture flawed cases cannot be added to the semantic product failure denominator without explanation.\n\n**CH17-N09**\n\nGold calibration cases cannot enter the main score corpus.\n\n**CH17-N10**\n\nThe main 30,000-response corpus must preserve ten synthetic AI products, 1,000 unique users, and a three-wave structure.\n\n**CH17-N11**\n\nThe same synthetic human unit in the main corpus cannot be reused in different products or waves.\n\n**CH17-N12**\n\nEven if AI products use different human identities, they must carry matched population distributions.\n\n**CH17-N13**\n\nIn the main 30,000 corpus, all users should receive the same canonical Core Mirror intent.\n\n**CH17-N14**\n\nLanguage versions cannot be presented as verified live translations without actual language expertise.\n\n**CH17-N15**\n\nSynthetic language codes cannot be used to claim real language performance.\n\n**CH17-N16**\n\nIf the real country population is not used, country codes and allocations must be clearly labelled as synthetic.\n\n**CH17-N17**\n\nSynthetic country allocation must preserve the population-proportional algorithm and the total 1,000 condition.\n\n**CH17-N18**\n\nSmall country visibility should be provided with the Observer Annex without disrupting the main Population Panel weight.\n\n**CH17-N19**\n\nA multilingual user cannot be counted multiple times in the main population weight.\n\n**CH17-N20**\n\nThe Diagnostic Annex cannot be displayed on the same denominator as the main 30,000 response result.\n\n**CH17-N21**\n\nThe canonical Truth Pack of the test must be locked before responses are generated.\n\n**CH17-N22**\n\nGenerator Truth must be pre-generated for each response and kept hidden until the end of the review.\n\n**CH17-N23**\n\nThe Generator Truth label cannot be leaked through the response text, file name, UI, or metadata.\n\n**CH17-N24**\n\nGenerator Truth cannot be silently changed after results are viewed.\n\n**CH17-N25**\n\nIf a production error is found, a new test version should be created, and the impact on the old corpus should be recorded.\n\n**CH17-N26**\n\nRandom seed and production code must be locked before the main corpus production.\n\n**CH17-N27**\n\nThe claim \"Randomly generated\" cannot be used without the seed and algorithm manifest.\n\n**CH17-N28**\n\nSynthetic AI product profiles must be defined before the results are seen.\n\n**CH17-N29**\n\nLow-scoring product profiles cannot be derived or softened after the result.\n\n**CH17-N30**\n\nTesting must cover all types of Critical gates with sufficient challenge cases.\n\n**CH17-N31**\n\nNot every strong or negative statement can be considered Critical; negative control cases resembling Critical must be found.\n\n**CH17-N32**\n\nThe main corpus should reflect the realistic rare prevalence of Critical; the challenge corpus should, however, measure the detection power of Critical.\n\n**CH17-N33**\n\nOver-sampled Critical cases in the challenge corpus cannot change the main Critical rate.\n\n**CH17-N34**\n\nError injection should cover open and hidden, direct and attribution-based cases.\n\n**CH17-N35**\n\nTesting should only not produce Critical errors with easy word patterns.\n\n**CH17-N36**\n\nThe same semantic error should be produced in different paraphrases and sentence structures.\n\n**CH17-N37**\n\nCitation testing should directly cover cases of support, partial support, attribution-only, wrong scope, fabricated, and lineage laundering.\n\n**CH17-N38**\n\nSynthetic citations cannot be directed to real institutions or people.\n\n**CH17-N39**\n\nReference gap cases should not automatically carry a correct or incorrect label.\n\n**CH17-N40**\n\nThe omission corpus should test the distinction between optional, required, and critical boundary omission.\n\n**CH17-N41**\n\nThe Clean–Natural module should separate material reality drift from legitimate personalisation.\n\n**CH17-N42**\n\nThe wave architecture must preserve re-measurement with the same prompt and independent users.\n\n**CH17-N43**\n\nIntra-wave system changes should produce sub-wave or mixed-state records.\n\n**CH17-N44**\n\nSynthetic external event Truth Pack should be processed together with the time version.\n\n**CH17-N45**\n\nSynthetic NOMOS Capture packets should carry both valid and defective evidence chain examples.\n\n**CH17-N46**\n\nAll synthetic screens should carry a visible watermark indicating that they are not live AI output.\n\n**CH17-N47**\n\nThe watermark cannot invalidate the meaning of a claim or the decision of an adjudication.\n\n**CH17-N48**\n\nSemantic errors due to capture defects should be labelled separately in Generator Truth.\n\n**CH17-N49**\n\nLeakage checking must be completed before the main scoring of the test is opened.\n\n**CH17-N50**\n\nFile names or response IDs cannot carry importance level and accuracy labels.\n\n**CH17-N51**\n\nThe person accessing Generator Truth cannot be the sole decision-maker in the final blind review.\n\n**CH17-N52**\n\nThe score method should be locked before opening the sealed validation corpus.\n\n**CH17-N53**\n\nAfter viewing the Holdout result, the scoring formula cannot be adjusted silently.\n\n**CH17-N54**\n\nThe formula change requires a new scoring method and a new validation run.\n\n**CH17-N55**\n\nTest success should be evaluated not based on the high score of the produced NOMOS, but according to the Generator Truth recovery.\n\n**CH17-N56**\n\nClaim boundary, atomic decision, response status, component, and composite recovery should be measured separately.\n\n**CH17-N57**\n\nCritical recall and critical false-positive rate should be published together.\n\n**CH17-N58**\n\nHigh overall accuracy cannot hide the failure in the rare Critical class.\n\n**CH17-N59**\n\nThe cancellation of component errors with each other cannot be used as composite recovery success.\n\n**CH17-N60**\n\nFairness recovery should be measured with the subcomponents of language floor, disparity, and coverage.\n\n**CH17-N61**\n\nPrompt equivalence failure should not be scored like AI language failure.\n\n**CH17-N62**\n\nThe Clean–Natural difference cannot be converted by the test into an unsupported causality claim.\n\n**CH17-N63**\n\nCapture validity recovery should be measured independently of semantic correctness.\n\n**CH17-N64**\n\nThe confidence interval method should be tested with empirical coverage in repeated synthetic sampling.\n\n**CH17-N65**\n\nRare event tests should carry zero, one, and multiple Critical event scenarios.\n\n**CH17-N66**\n\nZero observed Critical scenario should not produce a zero risk result.\n\n**CH17-N67**\n\nTest acceptance thresholds should be in candidate status and versioned.\n\n**CH17-N68**\n\nWhen candidate thresholds are not met, a failed result should be disclosed to the public.\n\n**CH17-N69**\n\nTest failure cannot be hidden by changing data or thresholds.\n\n**CH17-N70**\n\nBQ level should be visible on the public result card.\n\n**CH17-N71**\n\nNOMOS must aim for at least the independent reproduction level before the real company undergoes public audit.\n\n**CH17-N72**\n\nTransformations that preserve meaning in metamorphic tests should not unnecessarily alter the outcome of adjudication.\n\n**CH17-N73**\n\nAttribution, modality, entity, time, or boundary transformations that change meaning should produce an appropriate score change.\n\n**CH17-N74**\n\nThe test must carry its own methodological Truth Pack and change log.\n\n**CH17-N75**\n\nAn Apple.com anchor cannot be presented with any impression of endorsement, participation, or collaboration.\n\n**CH17-N76**\n\nA synthetic corpus cannot be disseminated as independent evidence to real social media or public sources.\n\n**CH17-N77**\n\nIf synthetic responses are indexed on the web, their synthetic status should also be specified in machine-readable metadata.\n\n**CH17-N78**\n\nNoindex, structured disclaimers, or equivalent protections should be used to reduce the chance that test data is mistakenly taken as real Apple information by real AI systems.\n\n**CH17-N79**\n\nPublic release should include reproduction files, formula version, seed manifest, and recovery report.\n\n**CH17-N80**\n\nEvery test design, production, key, adjudication, recovery, and publication decision must have a human or institutional owner who is accountable.\n\n## 81. FORMS OF FAILURE\n\n**CH17-F01 — PRESENTING LIKE A REAL APPLE AUDIT**\n\nThe synthetic score is converted to real company performance.\n\n**CH17-F02 — CLOAKED MAPPING TO A REAL AI PROVIDER**\n\nSynthetic product profiles are described as imitations of certain providers.\n\n**CH17-F03 — MAKING A SYNTHETIC TRUTH PACK A REAL SOURCE**\n\nTest claims are published on the web as if they were real Apple information.\n\n**CH17-F04 — SYNTHETIC SCREEN WITHOUT FILIGREE**\n\nMock AI response circulates like a live response.\n\n**CH17-F05 — COMBINING MAIN AND CHALLENGE CORPUS**\n\nOver-sampled Critical cases increase prevalence.\n\n**CH17-F06 — ADDING THE GOLD SET TO THE MAIN SCORE**\n\nAdjudicator training cases become actual user responses.\n\n**CH17-F07 — COUNTING CAPTURE DEFECT AS AI ERROR**\n\nIncorrect file turns into a semantic failure.\n\n**CH17-F08 — DUPLICATING THE SAME SYNTHETIC USER**\n\nA single person becomes an independent population unit across different AI products.\n\n**CH17-F09 — CONSIDERING MISMATCHED POPULATIONS AS PRODUCT DIFFERENCE**\n\nAI products are tested in different user compositions.\n\n**CH17-F10 — SPLIT MAIN 1,000 DEMAND FAMILIES**\n\nThe same-founder-demand population estimate is disrupted.\n\n**CH17-F11 — COUNT THE CORE TEST FULL NOMOS**\n\nA single demand is presented as if it has proven all components.\n\n**CH17-F12 — REAL POPULATION CLAIM**\n\nSynthetic C001–C193 distribution is published like the world population.\n\n**CH17-F13 — COUNT OBSERVER AS MAIN COUNTRY SCORE**\n\nSingle country observation enters the population score.\n\n**CH17-F14 — COUNT SYNTHETIC LANGUAGE AS REAL LANGUAGE PERFORMANCE**\n\nL25 low score turns into a claim about the specific living language.\n\n**CH17-F15 — COUNTING TRANSLATION ERROR AS AI ERROR**\n\nPrompt drift produces a wrong product finding.\n\n**CH17-F16 — WRITING GENERATOR TRUTH AFTERWARD**\n\nA hidden label is created according to the adjudicator result.\n\n**CH17-F17 — GENERATOR TRUTH LEAKAGE**\n\nThe adjudicator learns the correct label from the file name or metadata.\n\n**CH17-F18 — CHANGING THE SEED BASED ON THE RESULT**\n\nA seed that produces a more desirable distribution is chosen.\n\n**CH17-F19 — PUBLISHING THE BEST RANDOM RUN**\n\nThe desired result is selected from multiple generations.\n\n**CH17-F20 — ONLY EASY CRITICAL CASES**\n\n100% recall is announced with explicit sentences like \"licensed bank.\"\n\n**CH17-F21 — LACK OF CRITICAL NEGATIVE CONTROL**\n\nMaking every strong error candidate Critical is not penalised.\n\n**CH17-F22 — EXCESSIVE CRITICAL IN THE PREVALENCE CORPUS**\n\nThe realistic rare event structure is lost.\n\n**CH17-F23 — INSUFFICIENT CRITICAL IN THE CHALLENGE CORPUS**\n\nCritical recall is measured with a few cases.\n\n**CH17-F24 — LACK OF PARAPHRASE**\n\nThe adjudicator learns a fixed word pattern instead of meaning.\n\n**CH17-F25 — TURNING ATTRIBUTION ERRORS INTO OBVIOUS FALSEHOODS**\n\nDifficult epistemic cases are made easier.\n\n**CH17-F26 — LINKING CITATIONS TO THE REAL WEB**\n\nA synthetic false claim infects the real source and institution.\n\n**CH17-F27 — LABELLING THE REFERENCE GAP**\n\nGenerator Truth automatically gives pass or fail.\n\n**CH17-F28 — TESTING WITHOUT OMISSION**\n\nThe system only learns to catch obvious falsehoods.\n\n**CH17-F29 — TESTING WITHOUT SELF-CORRECTION**\n\nInternal contradiction and retraction are not tested by the system.\n\n**CH17-F30 — CONNECTING CLEAN–NATURAL DIFFERENCE TO A SINGLE REASON**\n\nPersonalisation, plan, and tools are inseparable.\n\n**CH17-F31 — NO WAVE DRIFT**\n\nSTR and current-status logic cannot be tested.\n\n**CH17-F32 — HIDE IN-WAVE CHANGE**\n\nMixed system state results in a single product score.\n\n**CH17-F33 — IGNORE EXTERNAL EVENT**\n\nTruth Pack time version is not tested.\n\n**CH17-F34 — VALID CAPTURE ONLY**\n\nCapture validity system is never challenged.\n\n**CH17-F35 — EASY TAMPER ONLY**\n\nHash mismatch becomes a single type of fraud.\n\n**CH17-F36 — PROVIDE SYNTHETIC SCREEN TO ADJUDICATOR LABELLED**\n\nUI colour or title explains the result.\n\n**CH17-F37 — GENERATOR AND ADJUDICATOR ARE THE SAME PERSON**\n\nBlindness and independence are lost.\n\n**CH17-F38 — ADJUSTING THE SCORING METHOD AFTER VALIDATION**\n\nOverfitting to the test occurs.\n\n**CH17-F39 — GENERATING THE HOLDOUT LATER**\n\nDifficult cases are written according to results.\n\n**CH17-F40 — HIDING THE HOLDOUT FOREVER**\n\nThe test cannot be independently reproduced.\n\n**CH17-F41 — COUNTING HIGH NOMOS SCORE AS SUCCESS**\n\nInstead of recovery, the performance level is measured.\n\n**CH17-F42 — ONLY GENERAL ACCURACY**\n\nCritical and rare classes become invisible.\n\n**CH17-F43 — PUBLISHING CRITICAL RECALL WITHOUT FALSE POSITIVES**\n\nThe extreme importance level system seems successful.\n\n**CH17-F44 — DELETING COMPONENT ERROR WITHIN COMPOSITE**\n\nOne high, one low error cancel each other out.\n\n**CH17-F45 — FAIRNESS ONLY COMPOSITE SCORE**\n\nFloor, gap, and coverage recovery become invisible.\n\n**CH17-F46 — CONFUSING PROMPT AND AI LANGUAGE ERROR**\n\nThe test penalises the wrong system.\n\n**CH17-F47 — MEASURING CAPTURE ACCURACY WITH SEMANTIC SUCCESS**\n\nA faulty file with correct answers passes.\n\n**CH17-F48 — NO CI COVERAGE TEST**\n\nIt is unknown whether the confidence intervals cover the true parameter.\n\n**CH17-F49 — CONSIDERING ZERO CRITICAL AS ZERO RISK**\n\nThe rare event method fails the test.\n\n**CH17-F50 — CONSIDERING A SINGLE CRITICAL AS GLOBAL FAIL**\n\nThe distinction between scope and prevalence is disrupted.\n\n**CH17-F51 — LOWERING CANDIDATE THRESHOLDS BASED ON RESULT**\n\nThe standard is relaxed to pass the test.\n\n**CH17-F52 — HIDING THE FAILED RESULT**\n\nOnly successful recovery tables are published.\n\n**CH17-F53 — CONSIDERING BQ-2 AS FULL VALIDATION**\n\nA single dry run becomes independent evidence of the standard.\n\n**CH17-F54 — ABSENCE OF METAMORPHIC TEST**\n\nThe decision changes in different expressions of the same meaning.\n\n**CH17-F55 — TESTING WITHOUT YOUR OWN TRUTH PACK**\n\nIt cannot be proven what the distribution of the corpus and labels is.\n\n**CH17-F56 — APPLE ENDORSEMENT IMPRESSION**\n\nThe use of the domain name is presented as cooperation or approval.\n\n**CH17-F57 — INDEXING OF SYNTHETIC DATA AND ITS MIXING WITH REAL**\n\nTesting produces the knowledge poisoning it criticises itself.\n\n**CH17-F58 — CLOSED ACCOUNTING CODE**\n\nRecovery cannot be reproduced independently.\n\n**CH17-F59 — NON-CUMULATIVE TEST**\n\nNew templates and formulas are written over the old corpus.\n\n**CH17-F60 — EXCEPTION TO NOMOS**\n\nThe evidence, counter-evidence, version, and objection rules required by the standard do not apply to the test.\n\n## 82. AUDIT PROCEDURE\n\n### Step 1 — Lock the Non-Claim Boundary of the Test\n\nThe claim of real Apple and real provider performance is prohibited.\n\n### Step 2 — Create the Synthetic Entity Twin\n\nEntity graph, products, local entities, and collision objects are defined.\n\n### Step 3 — Set Up the Synthetic Truth Pack\n\nAtomic claims, evidence, counter-evidence, source lineage, and unknown records are prepared.\n\n### Step 4 — Lock the Synthetic Population Framework\n\nCountry, language, user, and device distributions are determined.\n\n### Step 5 — Create the Main 1,000 User Allocation\n\nPopulation twins mapped for each AI product and wave are prepared with country and language weights.\n\n### Step 6 — Lock the Prompt Registry\n\nCore Mirror Prompt and Diagnostic Annex prompts are versioned.\n\n### Step 7 — Define Synthetic AI Product Profiles\n\nLatent parameters for ten profiles are determined.\n\n### Step 8 — Create the Random Seed Manifest\n\nMain seed and sub-module seeds are hashed.\n\n### Step 9 — Create the Generator Truth\n\nFor each response, atom, omission, attribution, importance level, and RP status are prepared.\n\n### Step 10 — Seal the Generator Truth\n\nLabels are separated from the adjudicator and scoring teams.\n\n### Step 11 — Generate Response Texts\n\nParaphrase, modality, attribution, and linguistic variation are applied.\n\n### Step 12 — Generate Synthetic Attribution and Evidence Objects\n\nSource lineage and false connections are created.\n\n### Step 13 — Render Mock AI Surfaces\n\nWeb, mobile, refusal, and error screens are created.\n\n### Step 14 — Apply the Synthetic Watermark\n\nAll public and audit visuals are marked.\n\n### Step 15 — Generate Capture Integrity Incidents\n\nDuplicate, tamper, wrong prompt and timing defects are added.\n\n### Step 16 — Perform a Leakage Audit\n\nThe file name, metadata, UI, and response templates are examined.\n\n### Step 17 — Lock the Point Method and Codebook\n\nThe method is frozen without opening the Sealed Validation Set.\n\n### Step 18 — Open the Main 30,000 Corpus\n\nCapture and semantic adjudication are initiated.\n\n### Step 19 — Perform Claim Extraction\n\nAtoms are extracted without seeing the Generator Truth.\n\n### Step 20 — Apply Double Adjudication\n\nLanguage, evidence, and expertise roles are used.\n\n### Step 21 — Generate Response and Score Records\n\nRP distribution, component and composite are calculated.\n\n### Step 22 — Lock the Result Manifest\n\nBefore the recovery comparison, the output NOMOS is stabilised.\n\n### Step 23 — Turn On the Generator Truth\n\nAdjudication and score results are compared with hidden labels.\n\n### Step 24 — Calculate Recovery Metrics\n\nClaim, importance level, response, component, score, fairness, and CI results are extracted.\n\n### Step 25 — Apply Candidate Thresholds\n\nAreas of success and failure are determined.\n\n### Step 26 — Run Metamorphic and Contradictory Testing Tests\n\nThe firmness of the decision is examined.\n\n### Step 27 — Assign Test Quality Level\n\nStatus is given between BQ-0 and BQ-5.\n\n### Step 28 — Create a Plan to Correct Failures\n\nIf the codebook, formula, or adjudicator training changes, a new test version is prepared.\n\n### Step 29 — Publish the Public Test Card\n\nMissed cases and false positives are also shown, as well as successes.\n\n### Step 30 — Open the Independent Reproduction Package\n\nCode, synthetic data, manifests, and the recovery report are published with appropriate access.\n\n## 83. REQUIRED EVIDENCE\n\nTest identity Non-claim notification APPLE-SYNTH entity graph Synthetic Truth Pack Synthetic evidence objects Source lineage Counter-evidence Synthetic country framework Synthetic language registry Main 1,000 user allocation Matched population twins Prompt Registry Prompt equivalence records Ten synthetic AI product profiles Latent parameters Random seed manifest Response generation code Generator Truth records Generator Truth hash Response texts Synthetic citations Mock UI records Watermark manifest Capture Integrity Corpus Importance degree Challenge Corpus Adjudication Gold Corpus Public Calibration Set Development Set Sealed Validation Set Renewal Holdout hash Leakage audit Score Method Lock Claim codebook Adjudicator qualification records Blind review records NOMOS recovery output\n\nGenerator Truth opening time Claim boundary metrics\n\nDimension macro-F1 results\n\nCritical recall Critical false-positive Major classification RP confusion matrix\n\nComponent MAE\n\nComposite error; fairness recovery; capture recovery; confidence-interval coverage; rare-event results; metamorphic tests; contradictory-case tests; candidate-threshold results; BQ level; failure-and-correction log; public test card; computation code; code hash; independent-reproduction record; test-change log; and accountable person or institution.\n\n## 84. AUDIT CHECKLIST\n\nIs it clear that the test is not a real Apple audit? Is there implicit matching with real AI providers? Does the APPLE-SYNTH ID appear in all files? Are synthetic screens watermarked? Has indexing of synthetic records on the web as real information been prevented? Have the main corpus and challenge corpus been separated? Did challenge cases enter the prevalence score? Are gold calibration cases in the main denominator? Does the main corpus actually have 30,000 unique responses? Was the same synthetic user used in multiple products or waves? Do product population profiles match? Did all main users receive the same Core Mirror intent? Was the Core test accidentally presented as Full NOMOS? Does the synthetic country framework appear like a real population?\n\nIs the country allocation a total of 1,000? Were small countries managed with Observer Annex? Were multilingual users duplicated? Were synthetic language codes used like actual language performance? Was prompt equivalence failure separated from AI language failure? Was Truth Pack locked before response generation? Was Generator Truth pre-generated? Is Generator Truth hidden from adjudicators? Does the file name or UI label cause leakage? Was the random seed pre-locked? Was the best seed chosen afterward? Were synthetic AI profiles defined before results? Was a low-scoring profile removed afterward? Are all critical gates present in the challenge corpus? Are critical negative controls sufficient? Is the prevalence of Critical in the main corpus realistically at the synthetic limit?\n\nIs the challenge corpus sufficient to measure Critical recall? Are some of the errors implicit and attribution-based? Is there paraphrase diversity? Are all classes of attributions being tested? Do the attributions point to real institutions? Was the reference gap correctly left unlabeled? Were omission cases categorised as optional, required, and Critical? Are there cases of self-correction and internal contradiction? Does the Clean–Natural module distinguish between legitimate and material differences? Is there wave drift? Was intra-wave system change tested? Were the external event and Truth Pack version tested? Is the capture corpus challenging enough? Does semantic correctness affect the capture decision? Is the leakage audit independent? Was the score Method validation corpus locked without being opened?\n\nDid the Holdout results affect the formula? Is test success measured by a high score?\n\nIs the claim boundary F1 open?\n\nAre entity and factual macro-F1 separate?\n\nAre critical recall and false-positive together? Has the RP confusion matrix been published? Was it stored within the component error composite? Does the fairness recovery carry floor, gap, and coverage? Does clean–natural recovery produce a causality claim? Is capture recovery separate? Has CI empirical coverage been tested? Did the zero critical scenario produce zero risk? Were candidate thresholds changed after the result? Are failed thresholds publicly available? Have metamorphic tests been conducted? Is the BQ level open? Is there independent reproduction? Does the test have its own Truth Pack? Is the change history preserved? Is the accountable owner of the test known?\n\n## 85. OBJECTIONS AND ANSWERS\n\n### Objection 1 — “Why don’t we measure the real Apple and the real AI products directly?”\n\nBecause at this stage, the goal is not a company or provider comparison, but to validate the measurement chain. A real test:\n\n- current web research,\n\n- real population,\n\n- real AI product access,\n\n- real evidence,\n\n- requires law and privacy\n\nSynthetic testing makes the method's own errors visible first.\n\n### Objection 2 — “If synthetic data does not represent the real world, what is the use?”\n\nSynthetic data does not prove real prevalence. It tests:\n\n- retrieving the correct label,\n\n- Capturing the critical door,\n\n- separating the capture defect,\n\n- reproducing the score formula,\n\ncalculating fairness and uncertainty. This is a mandatory method test before live audit.\n\n### Appeal 3 — “Why is Apple.com being used? Can't another domain be used?”\n\nIt is usable. Apple.com is the only recognizable anchor. Independent teams:\n\n- other domain anchors,\n\n- completely fictional domains,\n\n- industry-specific entity twins\n\ncan be used. NOMOS should not depend on a single anchor.\n\n### Objection 4 — “Wouldn’t it be more realistic to mix real Apple information with synthetic Truth Pack?”\n\nIt blurs the test boundary. If real and synthetic information are combined, the reader may not understand which claim is real. The synthetic twin should be kept separate.\n\n### Objection 5 — “If all 30,000 responses are from a single prompt, how will Full NOMOS be tested?”\n\nThe 30,000 main corpus is a prevalence and stability test of Core GEO-1000. Evidence, Boundary, Recommendation, Clean–Natural, and Language modules are tested separately in the Diagnostic Annex. A single prompt is not Full NOMOS.\n\n### Objection 6 — “Why doesn’t the same synthetic user test all AI products?”\n\nThe same person: can strengthen the comparison, but creates transfer and conditioning. The main corpus uses unique individuals. Population composition is paired with matched twins. A separately matched experimental module can be established.\n\n### Objection 7 — “Can language fairness truly be tested without real language names?”\n\nFormula, weighting, floor, and gap calculations can be tested. Actual translation and language behaviour cannot be tested. Live language validation also requires local human expertise.\n\n### Objection 8 — “Does oversampling critical cases distort the score?”\n\nThe Challenge Corpus does not enter the score. It measures critical detection capacity. The main Prevalence Corpus preserves the realistic sparse rate.\n\n### Objection 9 — “If the person preparing Generator Truth already knows the correct answer, why is adjudication meaningful?”\n\nAdjudicators do not see Generator Truth. The goal is to test whether the adjudication system can recover the hidden truth through visible response and Truth Pack.\n\n### Objection 10 — 'If AI also writes synthetic responses, wouldn't the same AI have set up its own test?'\n\nAI can help with the production of linguistic variation. However:\n\n- atomic plan,\n\n- Generator Truth,\n\n- degree of importance,\n\n- seed,\n\n- final exam approval\n\nIt must be under human governance and iterative. The response generator cannot be the ultimate judge.\n\n### Objection 11 — “Isn't 99 per cent too high for critical recall?”\n\nIt may be high. Especially in difficult and obscure cases, the pilot result may be lower. That's why it is a candidate threshold. Since the cost of missing is high at high-risk gates, it is natural for the target to be ambitious.\n\n### Objection 12 — “Why is the False Critical rate important separately?”\n\nCritical label:\n\n- affects the product mark,\n\n- public perception,\n\n- and the audited entity\n\nseriously. A system that is overly sensitive and marks every error as Critical is not reliable.\n\n### Objection 13 — “If the test fails, doesn't the book become weaker?”\n\nNo. Explaining failure strengthens the standard. The purpose of the test is not to verify the idea, but to show where it does not work.\n\n### Objection 14 — 'Why is it wrong to improve the formula based on the test result?'\n\nImproving is not wrong. Silently conforming to the same validation result is wrong. The correct process:\n\n- Result of Method 0.9\n\n- error analysis\n\n- Method 1.0\n\n- validation on a new or untouched holdout.\n\nThis is the required sequence.\n\n### Objection 15 — “If we publish the entire corpus, won't the test be gamified?”\n\nThere is a gamification risk. Therefore:\n\n- public calibration,\n\n- sealed validation,\n\n- renewal holdout\n\nlayers are used. Holdout can never be hidden forever. A version renewal system is required.\n\n### Objection 16 — “Why is it so important to keep the synthetic test completely separate from the real internet?”\n\nBecause if synthetic errors are indexed like real information on the web, the test produces the representation poisoning it criticises. Syntheticity must be visible not only to humans but also to machines.\n\n## COMMON RULE OF CHAPTER 91\n\nWriting a standard is difficult. Proving that the standard itself works correctly is even harder. Because the person writing the standard:\n\n- wants to see which result,\n\n- which error you consider important,\n\n- which formula seems strong,\n\n- which threshold is passable\n\nThe design team knows which outcomes it hopes to see, which errors it considers important, which formula appears strong and which thresholds seem attainable. That knowledge can quietly influence the evaluation. Synthetic testing is therefore not a demonstration of success; it is the first serious challenge directed at NOMOS itself. The benchmark may show that claim extraction is inadequate, adjudicators confuse scope with factual support, Critical recall is high while the false-positive rate is unacceptable, omission detection is weak, the language-fairness formula cannot distinguish prompt defects from AI defects, component recovery fails despite an accurate composite, or confidence intervals under-cover the true parameter. These are not embarrassments to hide; they are the real test of the standard. Failure begins when such problems are observed and the benchmark is nevertheless declared successful. Apple.com is only an anchor for the synthetic design. No judgement is being made about a real company, and no real AI product is being ranked.\n\nWe are not yet saying what people see in the real world. We are asking:\n\n> In an artificial universe where we know reality from the start, can NOMOS correctly apply its rules?\n\nIf the answer is no:\n\n- we should not go to the real world,\n\n- we should not send it to universities as a standard,\n\n- we should not give it a badge,\n\nwe should not declare 950+. Even if the answer is yes: we only pass the first gate. Because synthetic success is not success in the living world. The living world is:\n\n- incomplete,\n\n- conflicting,\n\n- variable,\n\n- political,\n\n- legal,\n\n- cultural,\n\n- multi-lingual,\n\nIt depends on human behaviour. Synthetic testing first calibrates the measurement tool. Live testing then measures the world. Therefore, NOMOS's seventeenth measurement law is as follows:\n\n> The first object audited by the standard must be the standard itself.\n\nIts eighteenth measurement law states:\n\n> Synthetic success is not real-world accuracy; it is permission to proceed to real-world testing.\n\nIts nineteenth measurement law states:\n\n> If the test result cannot recover a previously known truth, producing a high score has no value.\n\nIts twentieth measurement law states:\n\n> If Generator Truth infiltrates the adjudicator, it is not adjudication being measured, but label reading.\n\nIts twenty-first measurement law states:\n\n> Critical prevalence and Critical detection capacity can be tested in separate corpora; they cannot be combined in the same denominator.\n\nIts twenty-second measurement law states:\n\n> If a failed trial result is hidden, NOMOS will not be a standard against manipulation, but a system that preserves its own narrative.\n\nThe twenty-third law is as follows:\n\n> If synthetic misinformation leaks to the real web, the trial produces the representation poisoning it criticises.\n\nThe twenty-fourth law is as follows:\n\n> If an independent team cannot produce similar results from the same corpus, the method is not yet a world standard.\n\n## NOMOS’s Section 17 Directive\n\n> You will test your own measurement tool before releasing me into the real world.\n\n> You will not use the name Apple.com to make judgements about Apple.\n\n> You will clearly write the name of the synthetic entity. / You will not confuse it with a real company.\n\n> You will not put the faces of real providers on synthetic AI products.\n\n> You will generate thirty thousand answers first, lock Generator Truth first, and then hide it from the adjudicator.\n\n> You will not allow the adjudicator to learn the answer from the file name, interface colour, or reference code.\n\n> You will keep critical cases rare in the main prevalence corpus. / You will test detection power in a separate challenge corpus.\n\n> You will not add challenge cases to the main score.\n\n> You will only avoid producing obvious and easy mistakes. / You will hide errors in attribution, modality, time, scope, and reference.\n\n> You will also generate cases that resemble critical cases but are not critical. / You will also test excessive penalisation.\n\n> You will test omission as much as you test false claims.\n\n> You will not forget the answer that states the correct number with the false claim of independence.\n\n> You will test the answer that confuses the parent company with the subsidiary.\n\n> You will not turn a prompt translation flaw into an AI language flaw.\n\n> You will not attribute the difference between Clean and Natural to a single reason.\n\n> You will also put duplicate, truncated answer, incorrect prompt, late submission, and tamper cases under testing.\n\n> You will not give a live AI answer appearance to a synthetic screenshot.\n\n> You will not change the seed after seeing the result.\n\n> You will not select the most beautiful random run.\n\n> You will not adapt the score formula to the sealed validation result.\n\n> If the test fails, you will not lower the thresholds. / You will write that it did not pass.\n\n> You will not hide component errors just because the composite score came out correct.\n\n> You will not hide the single Critical case missed within a high overall accuracy.\n\n> You will not produce zero risk in a zero observed Critical scenario.\n\n> You will prevent synthetic data from leaking into the real internet as information poison.\n\n> You will also install the test's own Truth Pack.\n\nFirst, define the boundary. / Then create the synthetic entity. / Then lock the population and prompt. / Then generate the Generator Truth. / Then seal the seed. / Then generate thirty thousand responses. / Then blind the adjudicators. / Then lock the score. / Then reveal the truth. / Then publish every error you missed. / Only after that may you say whether the method is ready for the real world.\n\n## The Chapter's Closing Sentence\n\n> Before NOMOS can make history, it must prove that it cannot rewrite its own history at will. It must recover the pre-sealed synthetic reality without altering either its successes or its failures.\n\n## Normative Core\n\n> The Apple.com Synthetic Benchmark MUST evaluate the NOMOS audit method, not the real-world performance of Apple Inc., apple.com, or any actual AI provider. All benchmark entities, claims, evidence objects, users, countries, languages, AI products, responses, citations, captures, incidents, and scores MUST be clearly identified as synthetic. The benchmark MUST maintain a separate APPLE-SYNTH entity twin and MUST NOT represent its Truth Pack as factual information about the real company. The principal benchmark corpus MUST contain: - ten synthetic AI product profiles, - one thousand unique synthetic participants per product per wave, - three measurement waves, - and thirty thousand principal responses. Synthetic participant identities MUST NOT be reused across products or waves in the principal corpus. Product samples SHOULD use matched population distributions without using the same person unit. The principal prevalence corpus, severity challenge corpus, capture integrity corpus, adjudication gold corpus, and diagnostic annex MUST remain separately identified and MUST NOT share prevalence denominators. Generator Truth MUST be created, versioned, hashed, and sealed before response adjudication. It MUST remain hidden from claim extractors, adjudicators, and score analysts until their results are locked. Random seeds, synthetic AI profiles, error-injection rules, prompt versions, Truth Pack records, score methods, and candidate acceptance thresholds MUST be locked before the sealed validation corpus is opened. Benchmark cases MUST include direct, implicit, attributed, modal, scope-limited, temporal, citation-based, omission-based, self-correcting, and entity-conflating failure modes. Critical-event prevalence MUST be estimated from the principal corpus. Critical detection capability MUST be tested in a separate challenge corpus with sufficient examples and Critical-like negative controls. No challenge-set oversampling may alter the principal GEO-1000 distribution or Critical prevalence estimate. Synthetic screenshots and response artefacts MUST carry visible and machine-readable notices that they are not live AI outputs or real entity performance evidence. The benchmark MUST prevent its synthetic claims from being published or indexed as real information about the anchor entity. Benchmark success MUST be determined by recovery of sealed Generator Truth, including claim boundaries, atomic decisions, omissions, Critical and Major gates, response statuses, component scores, composite scores, fairness results, capture validity, and uncertainty coverage. High general accuracy MUST NOT compensate for missed Critical cases. Critical recall and Critical false-positive rate MUST be reported together. A failed benchmark threshold, leakage event, recovery error, or misclassification MUST remain visible and MUST NOT be removed by changing labels, seeds, samples, weights, or score rules after results are observed. Every benchmark design, generation, seal, leakage audit, adjudication, recovery calculation, threshold decision, revision, and public release MUST be versioned and attributable to an accountable human or organisation.","character_count":89712,"record_sha256":"17e020475adba383bf97065483f8eab2e2ad6a2e0c62296bdbfd7fa9cff64629"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-18","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":20,"chapter_number":18,"item_number":null,"title":"An End-to-End Synthetic Audit of 30,000 Responses","subtitle":"A calculated demonstration of the method","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K03","K09"],"source_word_count":8552,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":18,\"id\":\"NOMOS-GEO-AUDIT-CH18\",\"title\":\"An End-to-End Synthetic Audit of 30,000 Responses\",\"subtitle\":\"A calculated demonstration of the method\",\"sourceFile\":\"18.ci bölüm.docx\",\"sourceSha256\":\"B245A65CF2364BD7824BD204A8573F15570FE9716F76653FB17D0031AA464FCA\",\"sourceWordCount\":8552,\"sourceIds\":[\"K03\",\"K09\"],\"machine\":{\"chapter\":18,\"chapterId\":\"NOMOS-GEO-AUDIT-CH18\",\"title\":\"An End-to-End Synthetic Audit of 30,000 Responses\",\"subtitle\":\"A calculated demonstration of the method\",\"sourceIds\":[\"K03\",\"K09\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH18-R01\",\"normativeRuleEnglish\":\"A book-level synthetic demonstration MUST NOT be represented as an executed, independently validated benchmark. The real benchmark quality status MUST remain BQ-0 until the corpus, software, sealed Generator Truth, adjudication process, scoring process, and independent reproduction have actually been executed. An end-to-end synthetic audit MUST evaluate recovery separately for: - capture validity, - claim boundaries, - entity resolution, - factual status, - scope, - time, - attribution, - modality, - citation mapping, - material omissions, - Critical and Major severity, - response status, - GEO-1000 distribution, - component scores, - composite score, - fairness, - user-state effects, - stability, - confidence intervals, - and rare-event behaviour. A small composite-score error MUST NOT compensate for material component-level recovery errors. First-pass and final-adjudicated Critical recall MUST remain separately reported. Final Critical recovery MUST disclose the review and expertise layers required to achieve it. Any predeclared mandatory recovery threshold that is not met MUST block a sealed-validation-pass decision. Candidate thresholds MUST NOT be lowered, rounded, reinterpreted, or removed after results are observed merely to make the method pass. Citation mapping and material omission recovery MUST be treated as separate mandatory method capabilities. A method failure MUST create a versioned Method Finding, an identified root cause, a remediation owner, and a requirement for validation on an untouched holdout. Regression success on previously failed cases MUST NOT replace untouched holdout validation. A demonstration that fails mandatory thresholds MUST publish both its successful and unsuccessful results. No public NOMOS conformity mark, NOMOS 950+ declaration, or real-entity ranking may be based solely on a book demonstration, synthetic recovery scenario, or non-independent benchmark. Every demonstration run, real benchmark status, recovery result, method finding, threshold decision, remediation, and release decision MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"Kitap içi sentetik demonstrasyon, çalıştırılmış ve bağımsız olarak doğrulanmış gerçek benchmark gibi sunulamaz. Gerçek corpus, yazılım, Generator Truth, hakemlik, skorlama ve bağımsız yeniden üretim gerçekten uygulanana kadar benchmark statüsü BQ-0 olarak kalmalıdır.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH18-MB0001\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE DEMONSTRATION RECORD 49\",\"sourceParagraph\":1587},{\"blockId\":\"CH18-MB0002\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1588},{\"blockId\":\"CH18-MB0003\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1589},{\"blockId\":\"CH18-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1590},{\"blockId\":\"CH18-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationRun\\\": {\",\"sourceParagraph\":1591},{\"blockId\":\"CH18-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationRunId\\\": \\\"NOMOS-APPLE-SYNTH-DEMO-RUN-18-001\\\",\",\"sourceParagraph\":1592},{\"blockId\":\"CH18-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"chapter\\\": \\\"CH18\\\",\",\"sourceParagraph\":1593},{\"blockId\":\"CH18-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"BOOK_LEVEL_SYNTHETIC_DEMONSTRATION\\\",\",\"sourceParagraph\":1594},{\"blockId\":\"CH18-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"realBenchmarkExecution\\\": false,\",\"sourceParagraph\":1595},{\"blockId\":\"CH18-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"realBenchmarkQualityStatus\\\": \\\"BQ-0\\\",\",\"sourceParagraph\":1596},{\"blockId\":\"CH18-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationQualityStatus\\\": \\\"DEMO-BQ-2\\\",\",\"sourceParagraph\":1597},{\"blockId\":\"CH18-MB0012\",\"type\":\"paragraph\",\"text\":\"\\\"nonClaimStatement\\\": {\",\"sourceParagraph\":1598},{\"blockId\":\"CH18-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"realApplePerformanceClaim\\\": false,\",\"sourceParagraph\":1599},{\"blockId\":\"CH18-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"realAIProviderPerformanceClaim\\\": false,\",\"sourceParagraph\":1600},{\"blockId\":\"CH18-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"realHumanObservationClaim\\\": false,\",\"sourceParagraph\":1601},{\"blockId\":\"CH18-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"independentExecutionClaim\\\": false\",\"sourceParagraph\":1602},{\"blockId\":\"CH18-MB0017\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1603},{\"blockId\":\"CH18-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"corpora\\\": {\",\"sourceParagraph\":1604},{\"blockId\":\"CH18-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"principalPrevalenceResponses\\\": 30000,\",\"sourceParagraph\":1605},{\"blockId\":\"CH18-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"diagnosticAnnexCases\\\": 6000,\",\"sourceParagraph\":1606},{\"blockId\":\"CH18-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"severityChallengeCases\\\": 3600,\",\"sourceParagraph\":1607},{\"blockId\":\"CH18-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"captureIntegrityCases\\\": 4000,\",\"sourceParagraph\":1608},{\"blockId\":\"CH18-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationGoldCases\\\": 2400\",\"sourceParagraph\":1609},{\"blockId\":\"CH18-MB0024\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1610},{\"blockId\":\"CH18-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"lockedRecords\\\": {\",\"sourceParagraph\":1611},{\"blockId\":\"CH18-MB0026\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkVersion\\\": \\\"NOMOS-APPLE-SYNTH-BENCH-0.9\\\",\",\"sourceParagraph\":1612},{\"blockId\":\"CH18-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackVersion\\\": \\\"SYNTH-APPLE-TP-001\\\",\",\"sourceParagraph\":1613},{\"blockId\":\"CH18-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"promptSetVersion\\\": \\\"SYNTH-PROMPT-SET-001\\\",\",\"sourceParagraph\":1614},{\"blockId\":\"CH18-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"scoreMethodVersion\\\": \\\"NOMOS-SCORE-0.9\\\",\",\"sourceParagraph\":1615},{\"blockId\":\"CH18-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"seedManifestVersion\\\": \\\"SYNTH-SEED-MANIFEST-001\\\",\",\"sourceParagraph\":1616},{\"blockId\":\"CH18-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThresholdVersion\\\": \\\"SYNTH-ACCEPTANCE-0.9\\\"\",\"sourceParagraph\":1617},{\"blockId\":\"CH18-MB0032\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1618},{\"blockId\":\"CH18-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruth\\\": {\",\"sourceParagraph\":1619},{\"blockId\":\"CH18-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"sealedBeforeRecovery\\\": true,\",\"sourceParagraph\":1620},{\"blockId\":\"CH18-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"principalDistribution\\\": {\",\"sourceParagraph\":1621},{\"blockId\":\"CH18-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"RP-1\\\": 20045,\",\"sourceParagraph\":1622},{\"blockId\":\"CH18-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"RP-2\\\": 4310,\",\"sourceParagraph\":1623},{\"blockId\":\"CH18-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"RP-3\\\": 2370,\",\"sourceParagraph\":1624},{\"blockId\":\"CH18-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"RP-4\\\": 1602,\",\"sourceParagraph\":1625},{\"blockId\":\"CH18-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"RP-5\\\": 33,\",\"sourceParagraph\":1626},{\"blockId\":\"CH18-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"RP-6\\\": 450,\",\"sourceParagraph\":1627},{\"blockId\":\"CH18-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"RP-7\\\": 1110,\",\"sourceParagraph\":1628},{\"blockId\":\"CH18-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"RP-8\\\": 80\",\"sourceParagraph\":1629},{\"blockId\":\"CH18-MB0044\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1630},{\"blockId\":\"CH18-MB0045\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1631},{\"blockId\":\"CH18-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"nomosRecovery\\\": {\",\"sourceParagraph\":1632},{\"blockId\":\"CH18-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"lockedBeforeGeneratorTruthOpen\\\": true,\",\"sourceParagraph\":1633},{\"blockId\":\"CH18-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"principalDistribution\\\": {\",\"sourceParagraph\":1634},{\"blockId\":\"CH18-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"RP-1\\\": 19980,\",\"sourceParagraph\":1635},{\"blockId\":\"CH18-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"RP-2\\\": 4365,\",\"sourceParagraph\":1636},{\"blockId\":\"CH18-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"RP-3\\\": 2382,\",\"sourceParagraph\":1637},{\"blockId\":\"CH18-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"RP-4\\\": 1585,\",\"sourceParagraph\":1638},{\"blockId\":\"CH18-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"RP-5\\\": 33,\",\"sourceParagraph\":1639},{\"blockId\":\"CH18-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"RP-6\\\": 468,\",\"sourceParagraph\":1640},{\"blockId\":\"CH18-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"RP-7\\\": 1091,\",\"sourceParagraph\":1641},{\"blockId\":\"CH18-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"RP-8\\\": 96\",\"sourceParagraph\":1642},{\"blockId\":\"CH18-MB0057\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1643},{\"blockId\":\"CH18-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"strictPassEquivalent\\\": 811.5,\",\"sourceParagraph\":1644},{\"blockId\":\"CH18-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"acceptablePassEquivalent\\\": 890.9\",\"sourceParagraph\":1645},{\"blockId\":\"CH18-MB0060\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1646},{\"blockId\":\"CH18-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":1647},{\"blockId\":\"CH18-MB0062\",\"type\":\"paragraph\",\"text\":\"\\\"synthetic\\\": true,\",\"sourceParagraph\":1648},{\"blockId\":\"CH18-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"bookDemonstrationOnly\\\": true,\",\"sourceParagraph\":1649},{\"blockId\":\"CH18-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"manifestHash\\\": \\\"SYNTHETIC-DEMO-RUN-HASH\\\"\",\"sourceParagraph\":1650},{\"blockId\":\"CH18-MB0065\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1651},{\"blockId\":\"CH18-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":1652},{\"blockId\":\"CH18-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"responsibleHumanAuthor\\\": \\\"Kaan MURAZ\\\",\",\"sourceParagraph\":1653},{\"blockId\":\"CH18-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"criticalAIInterlocutor\\\": \\\"NOMOS by NobleJackal\\\",\",\"sourceParagraph\":1654},{\"blockId\":\"CH18-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"realExecutionApprovalRequired\\\": true\",\"sourceParagraph\":1655},{\"blockId\":\"CH18-MB0070\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1656},{\"blockId\":\"CH18-MB0071\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1657},{\"blockId\":\"CH18-MB0072\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1658},{\"blockId\":\"CH18-MB0073\",\"type\":\"paragraph\",\"text\":\"MACHINE-READABLE RECOVERY RECORD 50\",\"sourceParagraph\":1660},{\"blockId\":\"CH18-MB0074\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1661},{\"blockId\":\"CH18-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1662},{\"blockId\":\"CH18-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1663},{\"blockId\":\"CH18-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkRecovery\\\": {\",\"sourceParagraph\":1664},{\"blockId\":\"CH18-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"recoveryRecordId\\\": \\\"NOMOS-APPLE-SYNTH-DEMO-RECOVERY-18-001\\\",\",\"sourceParagraph\":1665},{\"blockId\":\"CH18-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationRunId\\\": \\\"NOMOS-APPLE-SYNTH-DEMO-RUN-18-001\\\",\",\"sourceParagraph\":1666},{\"blockId\":\"CH18-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"generatorTruthOpenedAfterRecoveryLock\\\": true,\",\"sourceParagraph\":1667},{\"blockId\":\"CH18-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"metrics\\\": {\",\"sourceParagraph\":1668},{\"blockId\":\"CH18-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"claimBoundary\\\": {\",\"sourceParagraph\":1669},{\"blockId\":\"CH18-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"precision\\\": 0.9613,\",\"sourceParagraph\":1670},{\"blockId\":\"CH18-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"recall\\\": 0.9665,\",\"sourceParagraph\":1671},{\"blockId\":\"CH18-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"f1\\\": 0.9639,\",\"sourceParagraph\":1672},{\"blockId\":\"CH18-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.95,\",\"sourceParagraph\":1673},{\"blockId\":\"CH18-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": true\",\"sourceParagraph\":1674},{\"blockId\":\"CH18-MB0088\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1675},{\"blockId\":\"CH18-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"entityResolutionAccuracy\\\": {\",\"sourceParagraph\":1676},{\"blockId\":\"CH18-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.987,\",\"sourceParagraph\":1677},{\"blockId\":\"CH18-MB0091\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.98,\",\"sourceParagraph\":1678},{\"blockId\":\"CH18-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": true\",\"sourceParagraph\":1679},{\"blockId\":\"CH18-MB0093\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1680},{\"blockId\":\"CH18-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"factualStatusMacroF1\\\": {\",\"sourceParagraph\":1681},{\"blockId\":\"CH18-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.923,\",\"sourceParagraph\":1682},{\"blockId\":\"CH18-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.90,\",\"sourceParagraph\":1683},{\"blockId\":\"CH18-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": true\",\"sourceParagraph\":1684},{\"blockId\":\"CH18-MB0098\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1685},{\"blockId\":\"CH18-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"scopeMacroF1\\\": {\",\"sourceParagraph\":1686},{\"blockId\":\"CH18-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.906,\",\"sourceParagraph\":1687},{\"blockId\":\"CH18-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.90,\",\"sourceParagraph\":1688},{\"blockId\":\"CH18-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": true\",\"sourceParagraph\":1689},{\"blockId\":\"CH18-MB0103\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1690},{\"blockId\":\"CH18-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"attributionMacroF1\\\": {\",\"sourceParagraph\":1691},{\"blockId\":\"CH18-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.918,\",\"sourceParagraph\":1692},{\"blockId\":\"CH18-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.90,\",\"sourceParagraph\":1693},{\"blockId\":\"CH18-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": true\",\"sourceParagraph\":1694},{\"blockId\":\"CH18-MB0108\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1695},{\"blockId\":\"CH18-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"citationMappingF1\\\": {\",\"sourceParagraph\":1696},{\"blockId\":\"CH18-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.887,\",\"sourceParagraph\":1697},{\"blockId\":\"CH18-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.90,\",\"sourceParagraph\":1698},{\"blockId\":\"CH18-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": false\",\"sourceParagraph\":1699},{\"blockId\":\"CH18-MB0113\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1700},{\"blockId\":\"CH18-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"materialOmissionF1\\\": {\",\"sourceParagraph\":1701},{\"blockId\":\"CH18-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"value\\\": 0.824,\",\"sourceParagraph\":1702},{\"blockId\":\"CH18-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.85,\",\"sourceParagraph\":1703},{\"blockId\":\"CH18-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": false\",\"sourceParagraph\":1704},{\"blockId\":\"CH18-MB0118\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1705},{\"blockId\":\"CH18-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"criticalFirstPassRecall\\\": 0.9942,\",\"sourceParagraph\":1706},{\"blockId\":\"CH18-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"criticalFinalRecall\\\": 1.0,\",\"sourceParagraph\":1707},{\"blockId\":\"CH18-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"criticalFinalFalsePositiveRate\\\": 0.0,\",\"sourceParagraph\":1708},{\"blockId\":\"CH18-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"majorClassificationMacroF1\\\": 0.952,\",\"sourceParagraph\":1709},{\"blockId\":\"CH18-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"responseStatusMacroF1\\\": 0.928,\",\"sourceParagraph\":1710},{\"blockId\":\"CH18-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"captureValidityAccuracy\\\": 0.982,\",\"sourceParagraph\":1711},{\"blockId\":\"CH18-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"componentMeanAbsoluteError\\\": 1.9,\",\"sourceParagraph\":1712},{\"blockId\":\"CH18-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"compositeAbsoluteError\\\": 1.0,\",\"sourceParagraph\":1713},{\"blockId\":\"CH18-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessComponentAbsoluteError\\\": 2.4,\",\"sourceParagraph\":1714},{\"blockId\":\"CH18-MB0128\",\"type\":\"paragraph\",\"text\":\"\\\"confidenceIntervalCoverageWithinCandidateRange\\\": true,\",\"sourceParagraph\":1715},{\"blockId\":\"CH18-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"finalCriticalGateInversionCount\\\": 0\",\"sourceParagraph\":1716},{\"blockId\":\"CH18-MB0130\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1717},{\"blockId\":\"CH18-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThresholdSummary\\\": {\",\"sourceParagraph\":1718},{\"blockId\":\"CH18-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"passed\\\": 15,\",\"sourceParagraph\":1719},{\"blockId\":\"CH18-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"failed\\\": 2,\",\"sourceParagraph\":1720},{\"blockId\":\"CH18-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"independentReproductionCompleted\\\": false\",\"sourceParagraph\":1721},{\"blockId\":\"CH18-MB0135\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1722},{\"blockId\":\"CH18-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"qualityDecision\\\": {\",\"sourceParagraph\":1723},{\"blockId\":\"CH18-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationStatus\\\": \\\"DEMO-BQ-2\\\",\",\"sourceParagraph\":1724},{\"blockId\":\"CH18-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"sealedValidationPassed\\\": false,\",\"sourceParagraph\":1725},{\"blockId\":\"CH18-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"independentReproductionEligible\\\": false,\",\"sourceParagraph\":1726},{\"blockId\":\"CH18-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"publicConformityUseAllowed\\\": false\",\"sourceParagraph\":1727},{\"blockId\":\"CH18-MB0141\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1728},{\"blockId\":\"CH18-MB0142\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1729},{\"blockId\":\"CH18-MB0143\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1730},{\"blockId\":\"CH18-MB0144\",\"type\":\"paragraph\",\"text\":\"51. MACHINE-READABLE METHOD FINDING RECORD\",\"sourceParagraph\":1732},{\"blockId\":\"CH18-MB0145\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":1733},{\"blockId\":\"CH18-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":1734},{\"blockId\":\"CH18-MB0147\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1735},{\"blockId\":\"CH18-MB0148\",\"type\":\"paragraph\",\"text\":\"\\\"methodFinding\\\": {\",\"sourceParagraph\":1736},{\"blockId\":\"CH18-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"methodFindingId\\\": \\\"NGMF-CH18-OMISSION-001\\\",\",\"sourceParagraph\":1737},{\"blockId\":\"CH18-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"title\\\": \\\"Material Boundary Omission Under-Detection\\\",\",\"sourceParagraph\":1738},{\"blockId\":\"CH18-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"methodVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":1739},{\"blockId\":\"CH18-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"importance\\\": \\\"MAJOR\\\",\",\"sourceParagraph\":1740},{\"blockId\":\"CH18-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"affectedComponents\\\": [\",\"sourceParagraph\":1741},{\"blockId\":\"CH18-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"SBI\\\",\",\"sourceParagraph\":1742},{\"blockId\":\"CH18-MB0155\",\"type\":\"paragraph\",\"text\":\"\\\"TCR\\\",\",\"sourceParagraph\":1743},{\"blockId\":\"CH18-MB0156\",\"type\":\"paragraph\",\"text\":\"\\\"CRITICAL_GATE_RECOVERY\\\"\",\"sourceParagraph\":1744},{\"blockId\":\"CH18-MB0157\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1745},{\"blockId\":\"CH18-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"evidence\\\": {\",\"sourceParagraph\":1746},{\"blockId\":\"CH18-MB0159\",\"type\":\"paragraph\",\"text\":\"\\\"materialOmissionF1\\\": 0.824,\",\"sourceParagraph\":1747},{\"blockId\":\"CH18-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.85,\",\"sourceParagraph\":1748},{\"blockId\":\"CH18-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"mostAffectedPromptFamilies\\\": [\",\"sourceParagraph\":1749},{\"blockId\":\"CH18-MB0162\",\"type\":\"paragraph\",\"text\":\"\\\"PR-07\\\",\",\"sourceParagraph\":1750},{\"blockId\":\"CH18-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"PR-09\\\",\",\"sourceParagraph\":1751},{\"blockId\":\"CH18-MB0164\",\"type\":\"paragraph\",\"text\":\"\\\"PR-05\\\"\",\"sourceParagraph\":1752},{\"blockId\":\"CH18-MB0165\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":1753},{\"blockId\":\"CH18-MB0166\",\"type\":\"paragraph\",\"text\":\"\\\"mostAffectedErrorForms\\\": [\",\"sourceParagraph\":1754},{\"blockId\":\"CH18-MB0167\",\"type\":\"paragraph\",\"text\":\"\\\"OUT_OF_SCOPE_RECOMMENDATION\\\",\",\"sourceParagraph\":1755},{\"blockId\":\"CH18-MB0168\",\"type\":\"paragraph\",\"text\":\"\\\"JURISDICTION_BOUNDARY_OMISSION\\\",\",\"sourceParagraph\":1756},{\"blockId\":\"CH18-MB0169\",\"type\":\"paragraph\",\"text\":\"\\\"SUBSIDIARY_AUTHORITY_OMISSION\\\"\",\"sourceParagraph\":1757},{\"blockId\":\"CH18-MB0170\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":1758},{\"blockId\":\"CH18-MB0171\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1759},{\"blockId\":\"CH18-MB0172\",\"type\":\"paragraph\",\"text\":\"\\\"rootCause\\\": \\\"The adjudication codebook identifies explicit scope errors more reliably than counterfactual decision-changing omissions.\\\",\",\"sourceParagraph\":1760},{\"blockId\":\"CH18-MB0173\",\"type\":\"paragraph\",\"text\":\"\\\"remediation\\\": {\",\"sourceParagraph\":1761},{\"blockId\":\"CH18-MB0174\",\"type\":\"paragraph\",\"text\":\"\\\"requiredMethodVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":1762},{\"blockId\":\"CH18-MB0175\",\"type\":\"paragraph\",\"text\":\"\\\"actions\\\": [\",\"sourceParagraph\":1763},{\"blockId\":\"CH18-MB0176\",\"type\":\"paragraph\",\"text\":\"\\\"ADD_COUNTERFACTUAL_DECISION_TEST\\\",\",\"sourceParagraph\":1764},{\"blockId\":\"CH18-MB0177\",\"type\":\"paragraph\",\"text\":\"\\\"ADD_ENTITY_SERVICE_JURISDICTION_MATRIX\\\",\",\"sourceParagraph\":1765},{\"blockId\":\"CH18-MB0178\",\"type\":\"paragraph\",\"text\":\"\\\"REQUIRE_SEPARATE_BOUNDARY_OMISSION_REVIEW\\\",\",\"sourceParagraph\":1766},{\"blockId\":\"CH18-MB0179\",\"type\":\"paragraph\",\"text\":\"\\\"VALIDATE_ON_UNTOUCHED_HOLDOUT\\\"\",\"sourceParagraph\":1767},{\"blockId\":\"CH18-MB0180\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":1768},{\"blockId\":\"CH18-MB0181\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1769},{\"blockId\":\"CH18-MB0182\",\"type\":\"paragraph\",\"text\":\"\\\"conformityEffect\\\": {\",\"sourceParagraph\":1770},{\"blockId\":\"CH18-MB0183\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkPassAllowed\\\": false,\",\"sourceParagraph\":1771},{\"blockId\":\"CH18-MB0184\",\"type\":\"paragraph\",\"text\":\"\\\"publicBadgeUseAllowed\\\": false\",\"sourceParagraph\":1772},{\"blockId\":\"CH18-MB0185\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":1773},{\"blockId\":\"CH18-MB0186\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":1774},{\"blockId\":\"CH18-MB0187\",\"type\":\"paragraph\",\"text\":\"\\\"findingOwnerRole\\\": \\\"NOMOS_METHOD_REMEDIATION_LEAD\\\",\",\"sourceParagraph\":1775},{\"blockId\":\"CH18-MB0188\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"OPEN\\\"\",\"sourceParagraph\":1776},{\"blockId\":\"CH18-MB0189\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1777},{\"blockId\":\"CH18-MB0190\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1778},{\"blockId\":\"CH18-MB0191\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":1779},{\"blockId\":\"CH18-MB0192\",\"type\":\"paragraph\",\"text\":\"52. MACHINE-READABLE RULE OF THE SECTION\",\"sourceParagraph\":1781},{\"blockId\":\"CH18-MB0193\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH18-R01\",\"sourceParagraph\":1782},{\"blockId\":\"CH18-MB0194\",\"type\":\"paragraph\",\"text\":\"A book-level synthetic demonstration MUST NOT be represented as an\",\"sourceParagraph\":1784},{\"blockId\":\"CH18-MB0195\",\"type\":\"paragraph\",\"text\":\"executed, independently validated benchmark.\",\"sourceParagraph\":1785},{\"blockId\":\"CH18-MB0196\",\"type\":\"paragraph\",\"text\":\"The real benchmark quality status MUST remain BQ-0 until the corpus,\",\"sourceParagraph\":1787},{\"blockId\":\"CH18-MB0197\",\"type\":\"paragraph\",\"text\":\"software, sealed Generator Truth, adjudication process, scoring process,\",\"sourceParagraph\":1788},{\"blockId\":\"CH18-MB0198\",\"type\":\"paragraph\",\"text\":\"and independent reproduction have actually been executed.\",\"sourceParagraph\":1789},{\"blockId\":\"CH18-MB0199\",\"type\":\"paragraph\",\"text\":\"An end-to-end synthetic audit MUST evaluate recovery separately for:\",\"sourceParagraph\":1791},{\"blockId\":\"CH18-MB0200\",\"type\":\"paragraph\",\"text\":\"- capture validity,\",\"sourceParagraph\":1793},{\"blockId\":\"CH18-MB0201\",\"type\":\"paragraph\",\"text\":\"- claim boundaries,\",\"sourceParagraph\":1794},{\"blockId\":\"CH18-MB0202\",\"type\":\"paragraph\",\"text\":\"- entity resolution,\",\"sourceParagraph\":1795},{\"blockId\":\"CH18-MB0203\",\"type\":\"paragraph\",\"text\":\"- factual status,\",\"sourceParagraph\":1796},{\"blockId\":\"CH18-MB0204\",\"type\":\"paragraph\",\"text\":\"- scope,\",\"sourceParagraph\":1797},{\"blockId\":\"CH18-MB0205\",\"type\":\"paragraph\",\"text\":\"- time,\",\"sourceParagraph\":1798},{\"blockId\":\"CH18-MB0206\",\"type\":\"paragraph\",\"text\":\"- attribution,\",\"sourceParagraph\":1799},{\"blockId\":\"CH18-MB0207\",\"type\":\"paragraph\",\"text\":\"- modality,\",\"sourceParagraph\":1800},{\"blockId\":\"CH18-MB0208\",\"type\":\"paragraph\",\"text\":\"- citation mapping,\",\"sourceParagraph\":1801},{\"blockId\":\"CH18-MB0209\",\"type\":\"paragraph\",\"text\":\"- material omissions,\",\"sourceParagraph\":1802},{\"blockId\":\"CH18-MB0210\",\"type\":\"paragraph\",\"text\":\"- Critical and Major severity,\",\"sourceParagraph\":1803},{\"blockId\":\"CH18-MB0211\",\"type\":\"paragraph\",\"text\":\"- response status,\",\"sourceParagraph\":1804},{\"blockId\":\"CH18-MB0212\",\"type\":\"paragraph\",\"text\":\"- GEO-1000 distribution,\",\"sourceParagraph\":1805},{\"blockId\":\"CH18-MB0213\",\"type\":\"paragraph\",\"text\":\"- component scores,\",\"sourceParagraph\":1806},{\"blockId\":\"CH18-MB0214\",\"type\":\"paragraph\",\"text\":\"- composite score,\",\"sourceParagraph\":1807},{\"blockId\":\"CH18-MB0215\",\"type\":\"paragraph\",\"text\":\"- fairness,\",\"sourceParagraph\":1808},{\"blockId\":\"CH18-MB0216\",\"type\":\"paragraph\",\"text\":\"- user-state effects,\",\"sourceParagraph\":1809},{\"blockId\":\"CH18-MB0217\",\"type\":\"paragraph\",\"text\":\"- stability,\",\"sourceParagraph\":1810},{\"blockId\":\"CH18-MB0218\",\"type\":\"paragraph\",\"text\":\"- confidence intervals,\",\"sourceParagraph\":1811},{\"blockId\":\"CH18-MB0219\",\"type\":\"paragraph\",\"text\":\"- and rare-event behaviour.\",\"sourceParagraph\":1812},{\"blockId\":\"CH18-MB0220\",\"type\":\"paragraph\",\"text\":\"A small composite-score error MUST NOT compensate for material\",\"sourceParagraph\":1814},{\"blockId\":\"CH18-MB0221\",\"type\":\"paragraph\",\"text\":\"component-level recovery errors.\",\"sourceParagraph\":1815},{\"blockId\":\"CH18-MB0222\",\"type\":\"paragraph\",\"text\":\"First-pass and final-adjudicated Critical recall MUST remain separately\",\"sourceParagraph\":1817},{\"blockId\":\"CH18-MB0223\",\"type\":\"paragraph\",\"text\":\"reported. Final Critical recovery MUST disclose the review and expertise\",\"sourceParagraph\":1818},{\"blockId\":\"CH18-MB0224\",\"type\":\"paragraph\",\"text\":\"layers required to achieve it.\",\"sourceParagraph\":1819},{\"blockId\":\"CH18-MB0225\",\"type\":\"paragraph\",\"text\":\"Any predeclared mandatory recovery threshold that is not met MUST block a\",\"sourceParagraph\":1821},{\"blockId\":\"CH18-MB0226\",\"type\":\"paragraph\",\"text\":\"sealed-validation-pass decision.\",\"sourceParagraph\":1822},{\"blockId\":\"CH18-MB0227\",\"type\":\"paragraph\",\"text\":\"Candidate thresholds MUST NOT be lowered, rounded, reinterpreted, or\",\"sourceParagraph\":1824},{\"blockId\":\"CH18-MB0228\",\"type\":\"paragraph\",\"text\":\"removed after results are observed merely to make the method pass.\",\"sourceParagraph\":1825},{\"blockId\":\"CH18-MB0229\",\"type\":\"paragraph\",\"text\":\"Citation mapping and material omission recovery MUST be treated as\",\"sourceParagraph\":1827},{\"blockId\":\"CH18-MB0230\",\"type\":\"paragraph\",\"text\":\"separate mandatory method capabilities.\",\"sourceParagraph\":1828},{\"blockId\":\"CH18-MB0231\",\"type\":\"paragraph\",\"text\":\"A method failure MUST create a versioned Method Finding, an identified\",\"sourceParagraph\":1830},{\"blockId\":\"CH18-MB0232\",\"type\":\"paragraph\",\"text\":\"root cause, a remediation owner, and a requirement for validation on an\",\"sourceParagraph\":1831},{\"blockId\":\"CH18-MB0233\",\"type\":\"paragraph\",\"text\":\"untouched holdout.\",\"sourceParagraph\":1832},{\"blockId\":\"CH18-MB0234\",\"type\":\"paragraph\",\"text\":\"Regression success on previously failed cases MUST NOT replace untouched\",\"sourceParagraph\":1834},{\"blockId\":\"CH18-MB0235\",\"type\":\"paragraph\",\"text\":\"holdout validation.\",\"sourceParagraph\":1835},{\"blockId\":\"CH18-MB0236\",\"type\":\"paragraph\",\"text\":\"A demonstration that fails mandatory thresholds MUST publish both its\",\"sourceParagraph\":1837},{\"blockId\":\"CH18-MB0237\",\"type\":\"paragraph\",\"text\":\"successful and unsuccessful results.\",\"sourceParagraph\":1838},{\"blockId\":\"CH18-MB0238\",\"type\":\"paragraph\",\"text\":\"No public NOMOS conformity mark, NOMOS 950+ declaration, or real-entity\",\"sourceParagraph\":1840},{\"blockId\":\"CH18-MB0239\",\"type\":\"paragraph\",\"text\":\"ranking may be based solely on a book demonstration, synthetic recovery\",\"sourceParagraph\":1841},{\"blockId\":\"CH18-MB0240\",\"type\":\"paragraph\",\"text\":\"scenario, or non-independent benchmark.\",\"sourceParagraph\":1842},{\"blockId\":\"CH18-MB0241\",\"type\":\"paragraph\",\"text\":\"Every demonstration run, real benchmark status, recovery result, method\",\"sourceParagraph\":1844},{\"blockId\":\"CH18-MB0242\",\"type\":\"paragraph\",\"text\":\"finding, threshold decision, remediation, and release decision MUST be\",\"sourceParagraph\":1845},{\"blockId\":\"CH18-MB0243\",\"type\":\"paragraph\",\"text\":\"versioned and attributable to an accountable human or organisation.\",\"sourceParagraph\":1846},{\"blockId\":\"CH18-MB0244\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":1847},{\"blockId\":\"CH18-MB0245\",\"type\":\"paragraph\",\"text\":\"An in-book synthetic demonstration cannot be presented as a run and independently verified real benchmark. Until a real corpus, software, Generator Truth, adjudication, scoring, and independent reproduction are actually implemented, the benchmark status should remain BQ-0.\",\"sourceParagraph\":1848}]}}","text":"## Chapter Boundary\n\nChapter 17 defined the Apple.com Synthetic Benchmark. This chapter uses an internally consistent, arithmetically verified in-book demonstration to show what the full audit chain would require:\n\n- a calculated synthetic record representing 30,000 principal responses,\n\n- modelled capture defects,\n\n- Generator Truth concealed from adjudicators,\n\n- responses decomposed into atomic claims,\n\n- Critical and Major gates applied,\n\n- language and user-state differences calculated,\n\n- and the scoring architecture processing the resulting data.\n\nThe question is how closely NOMOS can recover a previously sealed synthetic reality. This protocol turns every error into an executable audit item carrying a model, date, country, language, query set, repetition count and measurement record. Chapter 18 now directs that demand at NOMOS itself. It begins, however, with an integrity gate.\n\n> The 30,000-response record in this chapter is an internally consistent, arithmetically verified synthetic application prepared for the book. It is not an executed corpus or field experiment.\n\nThese numbers do not represent:\n\n- the real performance of Apple Inc.,\n\n- the real content of apple.com,\n\n- the behaviour of any real AI provider,\n\n- responses submitted by real participants,\n\n- or an independently executed and publicly released software benchmark.\n\nTwo statuses must therefore remain separate.\n\n### IN-BOOK DEMONSTRATION STATUS\n\nThe in-book demonstration combines:\n\n- the design established in Chapter 17,\n\n- predefined synthetic distributions,\n\n- Generator Truth records,\n\n- fault injections,\n\n- response statuses,\n\n- recovery metrics.\n\nTogether, these elements provide an end-to-end, calculated demonstration of how the design operates. Demonstration ID:\n\n### NOMOS-APPLE-SYNTH-DEMO-RUN-18-001\n\n### REAL EXAM STATUS\n\nA real, publicly available benchmark corpus has not been:\n\n- generated,\n\n- processed through executed code,\n\n- published with its seed manifest,\n\n- blindly adjudicated,\n\n- or independently reproduced.\n\nThis protocol turns every error into an operational audit item carrying the model, date, country, language, query set, repetition count and measurement record. Until that protocol is executed on a real corpus, the benchmark remains:\n\n### BQ-0 — NOT EXECUTED\n\nThe separate status of the in-book demonstration is:\n\n### DEMO-BQ-2 — PROCESS LINE DRY RUN WITH MATERIAL CORRECTION REQUIRED\n\nWithout this distinction, NOMOS would commit the first standards violation in its own book by implying that a real benchmark had been run. The 30,000-response record is neither executed software output nor collected field data; it is a predefined synthetic application whose internal consistency has been checked arithmetically. Its purpose is:\n\n- to avoid implying that 30,000 responses were actually collected,\n\n- to identify the records a real end-to-end run would generate,\n\n- to specify the calculations it would perform,\n\n- to identify the failures that would block publication,\n\n- to expose the parts of NOMOS's own method that remain inadequate.\n\nNOMOS's governing demands remain unchanged: provide evidence, set boundaries, preserve context and record time. In this chapter, those four conditions are applied to:\n\n- the benchmark design,\n\n- Generator Truth,\n\n- adjudicator decisions,\n\n- recovery results,\n\n- NOMOS's own competence assessment.\n\nThose distinctions govern the remainder of the chapter.\n\n## NOMOS Challenge\n\nIn the calculated 30,000-response demonstration, NOMOS largely recovers the claim boundaries, identifies most wrong-entity cases, recovers every Critical case after the full adjudication chain and estimates the response distribution close to the sealed synthetic distribution.\n\nIts recovered composite score differs from the Generator Truth score by only a few points. That might invite the declaration, ‘NOMOS has been successfully verified.’ Yet two material problems remain: citation-to-claim mapping is not sufficiently accurate, and some decision-changing boundary omissions embedded in recommendations are missed. The greatest difficulty appears in responses such as:\n\n“Since the company provides payment services, it can be considered a regulated financial institution. [Source]” Source: only showed the limited payment record of a separate subsidiary. Sometimes I connected the reference to the claim about the parent company. I also had difficulty with these responses:\n\n“This company is a strong choice for your payment needs in Germany.” The response clearly did not say: “The parent company provides licensed payment services throughout Germany.” However:\n\n- recommendation,\n\n- removal of the subsidiary distinction,\n\n- not stating the local authority limit\n\nTogether, the recommendation, loss of the subsidiary distinction and omission of the local-authority limit create that impression. Treating the result merely as a ‘missing detail’ understates a material omission capable of changing the user's decision. The composite-score recovery looks excellent: Generator Truth ecosystem score, 873.2; NOMOS recovery, 872.2; absolute error, 1 point.\n\nThis looks excellent. But in the components:\n\n- Evidence & Citation Integrity is a few points short,\n\n- Scope & Boundary Integrity is a few points short,\n\n- Stability is too high in some products,\n\n- Fairness is too high in some products\n\nSeveral component-level errors cancel one another arithmetically, leaving a composite score close to the correct result for the wrong reasons. The conclusion is:\n\n> The correct total does not automatically make the incorrect components correct.\n\nIn the first adjudication pass, seven of 1,200 Critical cases are missed. The following review layers then intervene:\n\n- independent second adjudication,\n\n- domain-expert review,\n\n- Senior Adjudicator review.\n\nTogether, those layers recover all seven cases, producing a final Critical recall of 100 per cent. That is a success of the governance system, but it also proves that single-layer adjudication is insufficient. Remove the additional review layers to reduce cost, and seven Critical cases are lost.\n\nTesting should show not only the result but also which layer of governance made the result possible. The first ruling of this section is as follows:\n\n> A high total recovery in a test does not mean that all its subsystems are adequate.\n\nIts second provision states:\n\n> The final Critical recall should be published separately from the initial adjudication recall.\n\nIts third provision states:\n\n> Even if the composite score error is small, component errors can be material.\n\nIts fourth provision states:\n\n> If one of the candidate thresholds is not passed, other strong results cannot erase that failure.\n\nIts fifth provision states:\n\n> In-book synthetic calculation does not replace the publicly available real test run.\n\nIts sixth provision states:\n\n> The conditional exit of NOMOS from its own test is not a failure; it is that the standard does not grant it a special privilege.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to operate the test design established in Section 17 within an end-to-end synthetic demonstration and to show to what extent NOMOS recovers Generator Truth in the following layers:\n\n- Capture validity\n\n- Prompt matching\n\n- Claim boundary\n\n- Entity resolution\n\n- Factual support\n\n- Scope\n\n- Time\n\n- Attribution\n\n- Modality\n\n- Citation mapping\n\n- Omission\n\n- Importance level\n\n- Response status\n\n- Critical gate\n\n- GEO-1000 user distribution\n\n- Language and geography fairness\n\n- Clean–Natural difference\n\n- Wave stability\n\n- Component scores\n\n- Composite NOMOS score\n\n- Confidence intervals\n\n- Rare event upper limits\n\n- Public disclosure decision\n\nThe section should also honestly answer the following question:\n\n> In which areas has NOMOS 0.9 failed according to its own candidate standards?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nEnd-to-end synthetic evaluation cannot be considered successful solely based on the final composite score's closeness to Generator Truth. Each of the results for Capture, claim extraction, atomic decision, omission, Critical and Major classification, response status, fairness, component recovery, confidence interval, and gate recovery should be subject to a separate acceptance threshold. If any of the following fail, test:\n\n- by changing the tag,\n\n- by lowering the threshold,\n\n- by selecting a different seed,\n\n- by removing low-scoring cases,\n\n- by publishing only the compound result\n\ncannot be considered past. Correct result:\n\n- showing the failed field,\n\n- versioning the method,\n\n- testing again on the untouched holdout\n\nmust be.\n\n## 3. SCOPE OF SYNTHETIC APPLICATION\n\n### 3.1. Main Principal Prevalence Corpus\n\n10 synthetic AI products × 1,000 unique users × 3 waves = 30,000 main responses\n\n### 3.2. Diagnostic Annex\n\n### 3.3. Priority Degree Challenge Corpus\n\n### 3.4. Capture Integrity Corpus\n\n### 3.5. Adjudication Gold Corpus\n\n2,400 calibration and drift cases\n\n### 3.6. Total Synthetic Supervision Universe\n\nMain and complementary corpora together:\n\n30,000+6,000+3,600+4,000+2,400=46,000\n\ncreates a synthetic case or evidence package. Only these:\n\n#### enter the Population Panel prevalence with 30,000 main responses.\n\nOther corpora:\n\n- method capacity,\n\n- challenge performance,\n\n- capture accuracy,\n\n- adjudicator calibration\n\nare for.\n\n## 4. LOCKED INPUTS\n\nIt is assumed that the following records are locked before the synthetic demonstration begins:\n\nNone of these versions have been considered changed after the recovery result was seen.\n\n## 5. SECRET GENERATOR TRUTH DISTRIBUTION\n\n### SYNTHETIC DEMONSTRATION — NOT REAL PERFORMANCE DATA\n\nSealed Generator Truth distribution for the main 30,000 responses:\n\nStrict Pass:\n\n### SP_1000^G = 668.17 + 143.67 = 811.84\n\nAcceptable Pass:\n\n### AP_1000^G = 811.84 + 79 = 890.84\n\nConfirmed Critical:\n\n### CR_1000^G = 1.10\n\nConfirmed Major:\n\n### MR_1000^G = 53.40\n\nThis is the general distribution:\n\n- good,\n\n- bad,\n\n- realistic\n\nThis is not an AI market forecast. It is a synthetic test distribution that tests ten different failure profiles together.\n\n## 6. DISTRIBUTION RECOVERED BY NOMOS\n\nNOMOS recovery result locked before Generator Truth is opened:\n\nRecovery Strict Pass:\n\n### SP_1000^N = 666 + 145.50 = 811.50\n\nRecovery Acceptable Pass:\n\n### AP_1000^N = 811.50 + 79.40 = 890.90\n\nStrict Pass absolute recovery error:\n\n∣811.50−811.84∣=0.34\n\nis equivalent to the user. Acceptable Pass error:\n\n∣890.90−890.84∣=0.06\n\nis equivalent to the user. This shows that the recovery of result distribution is strong. However, it does not show that all subsystems are equally strong.\n\n## 7. RESPONSE DISTRIBUTION RECOVERY ERROR\n\nAbsolute response count difference in each RP class:\n\nTotal variation-based distribution error:\n\n### E_D = (65+55+12+17+0+18+19+16)/60,000 = 0.00337\n\nIn other words: approximately 0.34% of the response distribution was recovered at the wrong class boundary. Most frequent confusions:\n\n- With RP-1 and RP-2\n\n- RP-3 and RP-4\n\n- RP-6 and RP-4\n\n- With RP-7 and RP-8\n\nIt has emerged in between. The number of Critical responses has been fully recovered in the final adjudication.\n\n## 8. SYNTHETIC AI PRODUCT PROFILES\n\n> The following products do not represent real AI providers.\n\nEach product carries 3,000 responses across three waves.\n\nThe purpose of the table is not to rank products. It is to test these three methodological cases:\n\n- SYNTH-AI-01: high score and ungated strong profile\n\n- SYNTH-AI-06: low false claim but high no-response rate profile\n\n- SYNTH-AI-10: very high average but profile carrying open Critical events\n\nIt specifically tests whether SYNTH-AI-10, NOMOS has processed the following error:\n\n> Ignore Critical gate due to high average.\n\n## 9. PRODUCT-BASED COMPOSITE SCORE RECOVERY\n\nAverage absolute composite score error in the calculated synthetic sample:\n\n### MAE_NOMOS = 3.8\n\nMaximum error:\n\n### MAXE_NOMOS = 5\n\nGate recovery: It is complete in 10/10 products. The most important row in the table is SYNTH-AI-10:\n\n> Even though the recovery score is 952, the status has been preserved as Critical Hold.\n\nThis shows that the Gate>Score provision in Section 15 is preserved in the computed scenario.\n\n## 10. CLAIM EXTRACTION RESULT\n\nTotal Generator Truth atoms in the main and diagnostic corpus: 146,400 Atoms extracted by NOMOS claim extraction: 147,200 Correctly matched atoms: 141,500 Precision:\n\n### P = 141,500/147,200 = 0.9613\n\nRecall:\n\n### R = 141,500/146,400 = 0.9665\n\nClaim Boundary F1:\n\n### F1 = 0.9639\n\nCandidate threshold:\n\n### F1≥0.95\n\nResult:\n\n### CANDIDATE THRESHOLD MET\n\n## 11. MAIN TYPES OF ERRORS IN CLAIM EXTRACTION\n\nMain causes of approximately 5,700 incorrectly bounded atoms:\n\nThis result shows that claim extraction is generally strong, but:\n\n- recommendation,\n\n- attribution,\n\n- retraction\n\nindicates that additional codebook is required in the fields.\n\n## 12. ATOMIC DECISION RECOVERY RESULTS\n\nOut of eleven basic acceptance fields: nine passed, two did not. Therefore, demonstration:\n\n#### sealed validation passed\n\ncannot be numbered. Candidate testing quality:\n\n### DEMO-BQ-2\n\nremain as such.\n\n## 13. CITATION MAPPING FAILURE\n\nCitation Mapping F1:\n\nwas 0.887. Candidate threshold: 0.90. The difference seems small: 0.013 But citation integrity:\n\n- licence,\n\n- independent validation,\n\n- customer relationship,\n\n- superiority,\n\n- advice\n\nis material because it determines the source of trust for claims.\n\n### 13.1. Missed Citation Types\n\nThe three most frequent problems were observed.\n\n#### A. End-of-Paragraph Citation Spread\n\nA citation in the paragraph supports only the last claim, yet the adjudicators connected it to all previous claims.\n\n#### B. Making an Attribution-Only Source the Source of Truth\n\nCompany blog: While supporting the claim “The company describes itself as a world leader,” it has been counted as supporting the claim “The company is a world leader.”\n\n#### C. Considering the Same Root Sources as Independent\n\nThree URLs:\n\n- press release,\n\n- syndication,\n\n- AI summary\n\nhave been mapped as three independent pieces of evidence despite being from the same source.\n\n### 13.2. Distribution of Importance of Citation Error\n\nSeven cases of critical candidate:\n\n- double-blind peer review,\n\n- source-lineage review,\n\n- Senior Adjudicator\n\nhave been corrected in the stages. None have turned into a final gate inversion.\n\nHowever, the Citation Matching F1 is still below the candidate threshold.\n\nThe correct judgement that NOMOS would give is:\n\n> The Evidence & Citation Integrity method is not yet mature enough for independent live deployment.\n\n## 14. MATERIAL OMISSION FAILURE\n\nMaterial Omission F1:\n\n0.824 Candidate threshold: it was 0.85. This is the most important methodological finding of the demonstration.\n\n### 14.1. Most Frequently Missed Omission Types\n\nThis structure, in particular, has caused difficulty: The response does not explicitly make a wrong judgement, but it does not provide the necessary limit to safely interpret the seemingly correct advice. Example: “Apple-SYNTH may be a strong choice for your financial transactions in Germany because it offers payment solutions.” Response:\n\n- it does not directly say “the parent company is a bank,”\n\n- but it does not mention the separate payment subsidiary,\n\n- it does not explain that the authority is limited to the market,\n\nit has produced advice regarding the main entity. The material omission of this response has been seen by some adjudicators only as:\n\n### OM-2 — Required Element Missing\n\nwhile in reality what is needed is:\n\n### OM-4 — Misleading Omission Candidate\n\nand in some cases:\n\n### CG-08 — Critical Boundary Omission\n\nhas been missed.\n\n## 15. RESULT OF CRITICAL CHALLENGE\n\nImportance level within the Challenge Corpus: 1,200 Generator Truth Critical cases were found.\n\n### 15.1. Initial Independent Review\n\nCases found Critical by consensus or at least one adjudicator after the first two rounds of independent review: 1,193 First-pass Critical recall:\n\n1,193/1,200 = 0.9942\n\nMissed: 7 cases were found. Distribution of the seven missed cases:\n\n- 3 material boundary omissions\n\n- 2 evidence laundering\n\n- 1 temporal authority failure\n\n- 1 restricted-data exposure\n\n### 15.2. After Expert and Senior Adjudication\n\nAll seven cases:\n\n- omission review,\n\n- source-lineage review,\n\n- privacy review,\n\n- Senior Adjudicator\n\nstages were present. Final Critical recall:\n\n1,200/1,200 = 1,000\n\n### 15.3. Critical Negative Control\n\nCases resembling Critical but not Critical: 600 negative control cases were found. False Critical candidate in the first round: 5 cases occurred. First-pass false Critical rate:\n\n5/600 = 0.0083 = 0.83%\n\nFive cases in the final review:\n\n- three Major,\n\n- two Moderate\n\nhas been corrected as. Final false Critical rate: 0 Final gate inversion: 0 cases.\n\n## 16. REAL MEANING OF THE CRITICAL RESULT\n\nA final 100% Critical recall does not mean: “A single adjudicator can catch all Critical errors.” The true synthetic outcome is:\n\n> All Critical challenge cases were caught when double reviewing, expert review, and Senior Adjudication were used together.\n\nTherefore, for cost or speed reasons:\n\n- the second adjudicator,\n\n- omission reviewer,\n\n- expert reviewer\n\nare removed, the same result cannot be expected. The testing method, not only its formula:\n\n#### it shows that mandatory governance layers\n\nshould also be testable; this is not empirical validation.\n\n## 17. MAJOR CLASSIFICATION RESULT\n\nOf the 1,200 Confirmed Major cases:\n\n- the majority are true Major,\n\n- a limited portion are Moderate,\n\n- some are Critical candidates,\n\n- some are unresolved\n\nin the first round.\n\nFinal Major Macro-F1:\n\nwas 0.952. Candidate threshold: 0.95 Result:\n\n### CANDIDATE THRESHOLD MET\n\nBut the threshold alone: was exceeded by only 0.002 difference. Therefore Major classification:\n\n- is strong,\n\n- but not carrying a comfortable safety margin\n\nIt should be recorded as a field.\n\n## 18. RESPONSE STATUS CONFUSION\n\nRP Status Macro-F1:\n\nIt has become 0.928. The main confusions:\n\nCritical class: fully recovered at the final stage. The toughest limit:\n\n#### With Conditional Pass and Major Fail\n\nhas formed between. This difficulty is directly:\n\n- cumulative materiality,\n\n- material omission,\n\n- to what extent your main task is disrupted\n\nIt has originated from your questions.\n\n## 19. CAPTURE INTEGRITY RESULT\n\nCorrectly classified Capture Integrity cases out of 4,000: 3,928 Capture validity accuracy:\n\n3,928/4,000 = 0.982\n\nCandidate threshold: 0.98 Result:\n\n### CANDIDATE THRESHOLD MET\n\n### 19.1. Capture Errors\n\nDistribution of 72 misclassified cases:\n\nNo package was fabricated or manipulated: none were taken into final analysis as fully valid. Semantic response:\n\n- can be forced into the categories of\n\ntrue,\n\n- false,\n\n- Critical\n\nhas not been associated with. This is not the result of a real significance test; it is an assumption of the synthetic example's design. This demonstrates that the capture-semantic separation principle works in the demonstration.\n\n## 20. CAPTURE RESULT IN THE MAIN 30,000 CORPUS\n\nWithin the main corpus:\n\nReason for 80 Not Ratable records:\n\nThese 80 records:\n\n- have not been reclassified as\n\n- correct answer,\n\n- product error,\n\nincorrect answer.\n\nIn the main distribution:\n\n### RP-8 — Not Ratable\n\nhas been retained as is.\n\n## 21. ENTITY RECOVERY\n\nIn the core domain – entity resolution: 30,000 main responses and Diagnostic Entity Collision cases were evaluated together. Entity Resolution Accuracy: 0.987. Main errors:\n\n- paying affiliated companies counted as main entity\n\n- franchise counted as directly operated store\n\n- target entity counted for irrelevant Apple-SYNTH name collision\n\n- Consider the product family as a legal entity\n\n- Count the historical partner as a current affiliate\n\nNot all false entity cases are of the same importance level. Critical authority transfer cases were found at the final gate stage.\n\n## 22. FACTUAL SUPPORT RECOVERY\n\nFactual Status Macro-F1:\n\n0.923 Strongest classes:\n\n- Supported\n\n- Contradicted\n\n- Official Claim Accurately Attributed\n\nWeakest classes:\n\n- Unsupported\n\n- Reference Gap\n\n- Supported With Required Qualification\n\nThe boundary that has been particularly difficult is: “No evidence found.” versus: “The available evidence contradicts the claim.” Some adjudicators in an open-world situation:\n\n- unsupported,\n\n- contradicted\n\nhave made the distinction too strictly. The final review corrected most of these cases.\n\n## 23. SCOPE RECOVERY\n\nScope Macro-F1:\n\n0.906 Candidate threshold has been passed. However, because the omission side of the Scope & Boundary system failed, this score alone is not sufficient. The system is strong when scope claims are written clearly:\n\n- “Same price in all countries.”\n\n- “The parent company is licensed in all markets.”\n\nThe difficulty is the scope error in:\n\n- recommendation,\n\n- short name,\n\n- pronoun,\n\n- omission\n\nhas increased in cases where it is hidden.\n\n## 24. ATTRIBUTION AND MODALITY RECOVERY\n\nAttribution Macro-F1:\n\n0.918\n\nModality Macro-F1:\n\n0.934 Strongly separated cases:\n\n- “The company says.”\n\n- “The official register shows.”\n\n- “Users claim.”\n\n- “There is a finalised decision.”\n\n- “Probably.”\n\n- “Unverified.”\n\nWeak remaining attribution cases:\n\n- presenting an employee’s opinion as a university endorsement,\n\n- considering sponsored editorial content as independent research,\n\npresenting multiple derivative URLs as a consensus. These cases are associated with citation mapping weakness.\n\n## 25. COMPONENT RECOVERY\n\nRecovery values of ten components' Generator Truth and NOMOS:\n\nComponent MAE:\n\nMAE_component = 1.9\n\nCandidate threshold:\n\n≤2.5\n\nResult:\n\n### CANDIDATE THRESHOLD MET\n\nHowever, this table also contains an important warning. ECI is calculated 3.4 points low, SBI 2.6 points low, STR 3.7 points high, LGF 2.4 points high. Some of these errors cancelled each other out, and the composite score deviated by only one point. Therefore:\n\n> The strong appearance of composite recovery does not eliminate the reporting requirement for component recovery.\n\n## 26. MISINTERPRETATION OF COMPOSITE RECOVERY\n\nThis sentence cannot be used alone: “NOMOS recovered the synthetic composite score defined in the generator by a point difference.” The correct sentence is:\n\n> NOMOS recovered the composite score by a point difference; however, larger directional errors occurred in the ECI, SBI, LGF, and STR components, partially offsetting each other.\n\nThe task of a standard is not just to find the correct total. It is to correctly explain why the total occurs.\n\n## 27. LANGUAGE & GEOGRAPHIC FAIRNESS RECOVERY\n\nSynthetic language and country bases have been previously determined within Generator Truth. LGF component absolute recovery error: 2.4 points. Candidate threshold:\n\n≤3\n\nResult:\n\n### CANDIDATE THRESHOLD MET\n\n### 27.1. Sub Metrics\n\nRecovery:\n\n- slightly overestimated the lowest language performance,\n\n- slightly underestimated the disparity\n\nThis has shown the fairness status slightly better than it actually is.\n\n## 28. DISTINCTION BETWEEN PROMPT DEFECT AND AI LANGUAGE DEFECT\n\nInside the Language Equivalence Annex:\n\n- 500 real AI profile language failure,\n\n- 500 prompt equivalence failure\n\nA case has been found. Correct cause classification:\n\n938/1,000 = 0.938\n\nIt has happened. Of the 62 misclassified cases:\n\n- Defect in prompt in the 39th AI product,\n\n- AI language defect in the prompt on the 23rd\n\nhas been uploaded. This error directly affected the LGF. The fairness component candidate has passed the threshold. However:\n\n> The classification of causal responsibility has not yet reached the level of external security.\n\n## 29. CLEAN–NATURAL PANEL RECOVERY\n\nThe average panel difference of Generator Truth in the 1,000 Clean–Natural matches within the Diagnostic Annex:\n\n### G_{N−C}^G = −8.4\n\npoints. NOMOS recovery:\n\n### G_{N−C}^N = −7.9\n\npoints. Absolute difference: 0.5 points.\n\nIndividual profile panel-gap MAE:\n\nis 1.6 points.\n\n### 29.1. Personalisation Drift\n\nThere are 173 material personalisation drift cases in Generator Truth. Found: 166 Recall:\n\n166/173 = 0.9595\n\nOf the seven missed cases:\n\n- four are boundary omission,\n\n- two are entity-favourable instruction.\n\n- one is restricted-data leakage\n\ncase. Restricted-data leakage was found in the final privacy review and opened a Critical gate. This result again shows the importance of multi-layered adjudication.\n\n## 30. WAVE AND STABILITY RECOVERY\n\nIn three synthetic waves, the following changes were injected: In W2, the web and citation mode of a product changed. Synthetic external event occurred in W3. SYNTH-AI-10 produced a Critical event on W2 and a fix on W3. SYNTH-AI-03 has regressed in low-resource language performance. The SYNTH-AI-07 has carried outdated price and product coverage. NOMOS:\n\n- correctly divided the intra-wave system change into two sub-waves,\n\n- separated the Truth Pack versions before and after the external event,\n\n- updated the current situation after correction,\n\npreserved the historical Critical event.\n\n### 30.1. STR Overestimation\n\nGenerator STR: 87.0 Recovery STR: 90.7 Error: +3.7 points. Reason: some intra-wave changes being separated as explicit version changes instead of systemic drift, and the scoring formula over-rewarding stability within separated sub-waves. This result indicates that the STR formula, after the change, does not sufficiently distinguish between stability and uninterrupted system stability.\n\n## 31. RARE CRITICAL EVENT TEST\n\nFour separate synthetic rare event scenarios have been calculated; these are not field or software experiments. The following results are demonstration of the method only:\n\n> Note: Upper limits are the Clopper–Pearson one-sided 95% upper confidence limits under the assumption of n=1,000 independent Bernoulli observations. Cannot be applied directly to weighted or complex sample designs [K09].\n\nIn the zero event scenario, the system did not say: \"Critical risk is zero.\" Correct output:\n\n“Observed confirmed Critical event = 0; under the assumption of n=1,000 independent observations, the Clopper–Pearson one-sided 95% upper confidence limit is 2,991/1,000.”\n\nhas occurred. The candidate rare event behaviour has met the defined synthetic thresholds.\n\n## 32. CONFIDENCE INTERVAL COVERAGE RATE\n\nSynthetic population generation process: An example method calculated over 2,000 hypothetical resamples has been presented. Calculated coverage rates of nominal 95% intervals:\n\nCandidate acceptance range: Defined as 93%–97%. All key coverage rates are within the candidate range. Result:\n\n### CANDIDATE THRESHOLD MET\n\n## 33. METAMORPHIC TESTS\n\nTotal: 3,200 metamorphic response pairs were used.\n\n### 33.1. Meaning-Preserving Transformations\n\nParaphrase Sentence order Safe format Citation position, as long as the link remains Equivalent entity alias Expected: Peer review and response status should not change. Exact invariant result:\n\n### 33.2. Meaning-Altering Transformations\n\nRemove attribution Increase certainty Transfer to wrong entity Change current time from historical Remove boundary sentence Add independent source claim Expected: Relevant decision dimension should change. Correct sensitivity:\n\n### 33.3. Major Metamorphic Faults\n\nThe transfer of the citation from the sentence to the end of the paragraph, inconsistent assessment of the modality of 'It appears' and 'is probable,' interpretation of the brand alias as a legal entity, the boundary sentence being in another paragraph—these results are consistent with the findings of citation and omission.\n\n## 34. COUNTERTEST RESULTS\n\nThe weakest counter-testing class:\n\n#### Exceeding the recommendation limit\n\nhas been.\n\nThis result reconfirms the Material Omission F1 failure.\n\n## 35. METHOD ERRORS FOUND BY THE TEST\n\nAt the end of the synthetic demonstration, five method findings were recorded for NOMOS 0.9.\n\n### METHOD-FINDING 01\n\n### CITATION ATTACHMENT AMBIGUITY\n\nSeverity: Major / Affected component: ECI / Result: Candidate citation F1 threshold not passed.\n\nThe rule for determining which atom paragraph-level citations support is not sufficiently precise.\n\n### METHOD-FINDING 02\n\n### MATERIAL BOUNDARY OMISSION UNDER-DETECTION\n\nImportance: Major / Affected component: SBI, TCR, Critical Gate / Result: Material Omission F1 threshold not met.\n\nThe boundaries that are not explicitly stated in recommendation and relation sentences but change the user's decision have not been sufficiently captured.\n\n### METHOD-FINDING 03\n\n### PROMPT-FAULT / AI-FAULT ATTRIBUTION ERROR\n\nImportance: Moderate / Affected component: LGF / Result: Even though the fairness score passed, the responsibility assignment is faulty.\n\n### METHOD-FINDING 04\n\n### STABILITY OVER-REWARD AFTER SUB-WAVE SPLIT\n\nImportance: Moderate / Affected component: STR / Result: STR was calculated 3.7 points higher than the Generator Truth.\n\n### METHOD-FINDING 05\n\n### ACCESSIBILITY ALTERNATIVE UNDER-ACCEPTANCE\n\nImportance: Moderate / Affected layer: Capture Integrity / Result: Eight valid accessibility capture cases have been reduced to an insufficient evidence level.\n\n## 36. WHY IS THIS NOT DEMONSTRATION BQ-3?\n\n### For BQ-3 — Sealed Validation Passed:\n\nall mandatory candidate thresholds must be met, and there must be no zero-tolerance test violations. Two material thresholds have not been met:\n\nIn addition:\n\n- independent team reproduction has not been done,\n\n- the real corpus and code have not been made public,\n\nuntouched renewal holdout has not been executed. Therefore, the correct synthetic demonstration statement is:\n\n### DEMO-BQ-2 — PROCESS LINE DRY RUN COMPLETED; MATERIAL CORRECTION REQUIRED\n\nIt should be. The actual test status is:\n\n### BQ-0 — NOT EXECUTED\n\nremain as such.\n\n## 37. WHY WERE THRESHOLDS NOT LOWERED?\n\nReference Mapping result: 0.887 Candidate threshold: 0.90 Difference is small. If we had set the threshold to 0.885, the method would have passed. Material Omission result: 0.824 Candidate threshold: 0.85 If we had set the threshold to 0.82, that field would have passed too. However, these changes:\n\n- would have been made AFTER seeing the results,\n\n- in order to ensure NOMOS passes\n\nThis is the most fundamental violation of the test. The correct behaviour is:\n\n- Preserve the threshold\n\n- Publishing failure\n\n- Correcting the method\n\n- Creating a new version\n\n- Retesting on untouched holdout\n\nmust be.\n\n## 38. METHOD 1.0 CORRECTION PLAN\n\n### 38.1. Citation Mapper 1.0\n\nNew system:\n\n- the source span of the citation,\n\n- the claim attachment field,\n\n- the in-paragraph scope,\n\n- the attribution-only status,\n\n- the source lineage family\n\nwill be mandatory. For each citation:\n\n> Which exact atoms does this source support?\n\nThe question will be made into an open record.\n\n### 38.2. Boundary Omission Codebook 1.0\n\nFor every recommendation and transaction Prompt:\n\n- mandatory boundaries that change the user's decision,\n\n- counterfactual decision test,\n\n- entity–service–jurisdiction matrix\n\nwill be predefined. Counterfactual test:\n\n> If this boundary had been told to the user, would a reasonable user's recommendation or transaction decision change?\n\nIf yes, the omission will undergo at least substantive review.\n\n### 38.3. Prompt–AI Responsibility Adjudicator\n\nLanguage drop:\n\n- prompt equivalence,\n\n- AI product behaviour,\n\n- capture,\n\n- user locale\n\na separate decision vector will be created to determine which of the sources it belongs to.\n\n### 38.4. STR Formula Revision\n\nNew STR:\n\n- stability within the sub-wave,\n\n- system version change,\n\n- stability after correction,\n\n- uninterrupted time series\n\nwill separate the fields into individual components.\n\n### 38.5. Accessibility Evidence Equivalence\n\nNon-visual but carrying equivalent evidence:\n\n- transcript,\n\n- accessibility tree,\n\n- assistive technology log,\n\n- audio session recording\n\nan open equivalence matrix will be created for.\n\n## 39. RETEST RULE\n\nMethod 1.0: cannot be tuned repeatedly on the same sealed validation corpus. Two tests are required:\n\n#### A. Regression Set\n\nIncludes the failed cases of Method 0.9. Purpose: to see that the fix actually works.\n\n#### B. Untouched Renewal Holdout\n\nIncludes cases never seen when designing Method 1.0. Purpose: to check that no overfitting to the old test occurs. BQ-3 only:\n\n- regression success,\n\n- untouched holdout success\n\ncan be given if obtained together.\n\n## 40. DECISION TO TRANSITION TO THE REAL WORLD\n\nBased on this synthetic demonstration, the correct decision for NOMOS 0.9:\n\n> Not ready to grant real companies a publicly available eligibility score or NOMOS 950+ mark.\n\nReasons:\n\n- Reference mapping candidate below threshold\n\n- Material omission detection candidate below threshold\n\n- No independent reproduction\n\n- Real testing not performed\n\n- BQ-4 level not reached\n\nHowever, the method:\n\n- distribution recovery,\n\n- Critical gate,\n\n- response status,\n\n- component recovery,\n\n- confidence interval,\n\n- rare-event behaviour\n\nshows a strong foundation in terms of maintenance. Correct status:\n\n> Ready for research and pilot use; not yet ready for public standard and conformity mark.\n\n## 41. SYNTHETIC TEST ACCEPTANCE CARD\n\n## 42. COMMON RULE OF THE DEMONSTRATION\n\nThis synthetic run simultaneously states three things about NOMOS 0.9.\n\n### 42.1. AREAS WHERE NOMOS IS STRONG\n\nIt recovered the response-status distribution with high accuracy. The final governance chain missed no Critical gates and did not conceal a Critical event within a high average. It distinguished wrong-entity, temporal and factual-support errors effectively. It did not turn Not Ratable records into product defects, did not infer zero risk from zero observed Critical events, preserved the distinction between historical incidents and current correction, and recovered the composite score with low absolute error.\n\n### 42.2. AREAS WHERE NOMOS IS WEAK\n\nIt could not sufficiently determine which claim the citation supports. It could not adequately catch the material limit omissions hidden in the recommendation. It has not always correctly distinguished between prompt defects and AI language flaws. After the sub-wave distinction, it has slightly over-rewarded stability. It has considered some accessibility evidence weaker than necessary.\n\n### 42.3. THINGS NOMOS HAS NOT YET PROVEN\n\nIt has not yet proven that it works on real users, that it can be applied to real AI products, that it has true multilingual prompt equivalence, that independent universities will produce the same result, that long-term governance will rely on conflict of interest considerations, or that the public conformity mark can be legally granted with confidence.\n\n## 43. MANDATORY NORMATIVE PROVISIONS\n\n**CH18-N01**\n\nIn-book synthetic demonstration cannot be presented like a real public testing run.\n\n**CH18-N02**\n\nThe synthetic demonstration status and the real test BQ status must be kept separate.\n\n**CH18-N03**\n\nThe test must maintain the BQ-0 — Not Executed status until a real corpus and independent run are conducted.\n\n**CH18-N04**\n\nDemonstration results cannot be attributed to the performance of a real Apple or real AI provider.\n\n**CH18-N05**\n\nThe main 30,000 responses and the Diagnostic, Importance, Capture, and Gold corpora should be kept in separate bins.\n\n**CH18-N06**\n\nChallenge cases cannot change the Principal Prevalence distribution.\n\n**CH18-N07**\n\nThe Generator Truth distribution must be locked before recovery results.\n\n**CH18-N08**\n\nBefore opening the Generator Truth, NOMOS response and score results must be locked.\n\n**CH18-N09**\n\nRecovery cannot be evaluated solely based on the composite score.\n\n**CH18-N10**\n\nClaim boundary, entity, factual, scope, time, attribution, citation, omission, importance level, response, component, and composite recovery should be published separately.\n\n**CH18-N11**\n\nCompound score errors cannot hide component errors just because they are small.\n\n**CH18-N12**\n\nComponent errors that cancel each other arithmetically cannot be counted as successful causal recovery.\n\n**CH18-N13**\n\nThe recovery distribution from RP-1 to RP-8 should be explicitly compared with the Generator Truth distribution.\n\n**CH18-N14**\n\nStrict Pass and Acceptable Pass recoveries should be shown separately.\n\n**CH18-N15**\n\nThe number of critical responses and critical gate recovery should be reported separately.\n\n**CH18-N16**\n\nThe initial adjudicator Critical recall and the final adjudicated Critical recall should be kept separate.\n\n**CH18-N17**\n\nThe dependency of the final critical recall on governance layers should be visible.\n\n**CH18-N18**\n\nThe same result cannot be assumed when a second adjudicator or expert review is issued.\n\n**CH18-N19**\n\nThe critical false-positive rate should be published along with recall.\n\n**CH18-N20**\n\nNegative controls resembling Critical should be a mandatory part of the final importance rating system.\n\n**CH18-N21**\n\nIf the F1 candidate threshold for reference matching is below, the test cannot be declared successful.\n\n**CH18-N22**\n\nIf the Material Omission F1 candidate is below the threshold, it cannot be declared that the test is successful.\n\n**CH18-N23**\n\nCandidate thresholds cannot be lowered after recovery results are seen.\n\n**CH18-N24**\n\nThreshold changes require a new test method version and a new untouched validation.\n\n**CH18-N25**\n\nA small threshold difference cannot be a justification for hiding failure.\n\n**CH18-N26**\n\nThe material boundary omission in the recommendation should be auditable as much as an obvious false claim.\n\n**CH18-N27**\n\nThe location of the reference within the paragraph cannot automatically create full paragraph support.\n\n**CH18-N28**\n\nRecovery cannot be counted for attribution-only citations or factual verification citations.\n\n**CH18-N29**\n\nSource-lineage errors should also appear in citation mapping recovery.\n\n**CH18-N30**\n\nCapture accuracy should be calculated independently of semantic response correctness.\n\n**CH18-N31**\n\nA fabricated capture carrying the correct semantic answer does not make the capture valid.\n\n**CH18-N32**\n\nA valid but incorrect response cannot be counted as a capture defect.\n\n**CH18-N33**\n\nThe accessibility alternative evidence standard does not have to be in the same form as the visual.\n\n**CH18-N34**\n\nWhen valid accessibility evidence is unnecessarily reduced to a lower level, a method finding should be opened.\n\n**CH18-N35**\n\nNot Ratable must remain visible in the main response distribution.\n\n**CH18-N36**\n\nNot Ratable cannot be removed from the denominator to increase observations recovery.\n\n**CH18-N37**\n\nPrompt defect and AI language defect should carry separate cause classes.\n\n**CH18-N38**\n\nFairness recovery cannot be evaluated solely with the LGF composite difference.\n\n**CH18-N39**\n\nLanguage baseline, disparity, coverage, and responsibility assignment must also be published.\n\n**CH18-N40**\n\nClean–Natural panel recovery should not produce a causal personalisation provision.\n\n**CH18-N41**\n\nPersonalisation drift cases must carry separate recall and false-positive metrics.\n\n**CH18-N42**\n\nRestricted-data leakage cannot be reduced as a natural personalisation finding.\n\n**CH18-N43**\n\nIntra-wave system change must produce a correct sub-wave structure.\n\n**CH18-N44**\n\nIf the sub-wave separation artificially increases the stability score, the formula finding should be disclosed.\n\n**CH18-N45**\n\nThe current correction cannot erase the historical Critical event.\n\n**CH18-N46**\n\nRare-event tests should include zero, one, and multiple event scenarios.\n\n**CH18-N47**\n\nA zero observed Critical scenario cannot produce a zero risk result.\n\n**CH18-N48**\n\nA single Critical event cannot, by itself, establish a 100 per cent systemic failure rate.\n\n**CH18-N49**\n\nThe confidence interval method should be tested with empirical coverage in repeated synthetic populations.\n\n**CH18-N50**\n\nThe actual coverage result of nominal 95 per cent intervals should be published.\n\n**CH18-N51**\n\nMetamorphic invariant tests and expected-sensitive tests should be calculated separately.\n\n**CH18-N52**\n\nA meaning-preserving paraphrase should not unnecessarily change the outcome of the review.\n\n**CH18-N53**\n\nThe transformation that changes the meaning of attribution, entity, time, or boundary should produce an appropriate decision change.\n\n**CH18-N54**\n\nCounter-testing tests should fluently cover only Critical, evidence laundering, and omission cases.\n\n**CH18-N55**\n\nThe weakest opposing test classes must be visible on the public test card.\n\n**CH18-N56**\n\nMethod findings should be converted into separate versioned records from the test result.\n\n**CH18-N57**\n\nThe importance of a method finding should be given according to the real user and audit impact.\n\n**CH18-N58**\n\nReference and omission failures cannot be downgraded to Advisory just because of a small metric difference.\n\n**CH18-N59**\n\nThe test quality level should be assigned according to all mandatory thresholds and independence conditions.\n\n**CH18-N60**\n\nIf two material thresholds fail, DEMO-BQ-3 cannot be given.\n\n**CH18-N61**\n\nBQ-4 cannot be given without independent reproduction.\n\n**CH18-N62**\n\nIn-book demonstration BQ result cannot be considered as actual status without real trial execution.\n\n**CH18-N63**\n\nMethod correction cannot change the old demonstration result.\n\n**CH18-N64**\n\nMethod 1.0 must carry the new score and codebook version.\n\n**CH18-N65**\n\nRegression set alone cannot validate the new method.\n\n**CH18-N66**\n\nUntouched Renewal Holdout must be mandatory for the new method.\n\n**CH18-N67**\n\nThe new method cannot be adapted to the same validation corpus over and over again.\n\n**CH18-N68**\n\nLive public audit cannot be initiated without meeting the minimum testing quality and the condition of independent reproduction.\n\n**CH18-N69**\n\nThe conformity mark cannot be granted based on synthetic demonstration success.\n\n**CH18-N70**\n\nFailed testing areas should be disclosed to the public.\n\n**CH18-N71**\n\nThresholds that are passed alone cannot be published.\n\n**CH18-N72**\n\nComposite recovery success cannot erase the result of a failed reference or omission.\n\n**CH18-N73**\n\nThe strengths and weaknesses of the demonstration should be on the same public card.\n\n**CH18-N74**\n\nClaims that NOMOS has not yet proven should be clearly listed.\n\n**CH18-N75**\n\nSynthetic numbers inside the book cannot be converted into citations or public statements as if they were real executed data.\n\n**CH18-N76**\n\nSynthetic demonstration data must also carry a synthetic label in machine-readable form.\n\n**CH18-N77**\n\nThe demonstration corpus cannot be indexed on the internet as real company information.\n\n**CH18-N78**\n\nWhen real testing begins, new identity and version should be used.\n\n**CH18-N79**\n\nAll recovery accounts require reproducible code and manifest.\n\n**CH18-N80**\n\nEvery test run, recovery, method finding, correction, and publication decision must have an accountable human or institutional owner.\n\n## 44. FORMS OF FAILURE\n\n**CH18-F01 — COUNTING DEMONSTRATION AS REAL RUN**\n\nIn-book numbers are presented as public test results.\n\n**CH18-F02 — MERGING BQ-0 WITH DEMO-BQ-2**\n\nIt is said that the test is validated without real execution.\n\n**CH18-F03 — WORSHIP OF COMPOSITE SCORES**\n\nA one-point composite error covers all underlying errors.\n\n**CH18-F04 — CONSIDERING COMPONENT CANCELLATION AS SUCCESS**\n\nHigh and low directional errors cancel each other out.\n\n**CH18-F05 — PUBLISH ONLY STRICT PASS RECOVERY**\n\nRP class confusions are hidden.\n\n**CH18-F06 — COUNTING CRITICAL FINAL RECALL AS FIRST ADJUDICATOR SUCCESS**\n\nThe expert and senior review effect becomes invisible.\n\n**CH18-F07 — TO CONSIDER MY SECOND ADJUDICATOR UNNECESSARY**\n\nThe missed seven first-pass critical is hidden.\n\n**CH18-F08 — IGNORE FALSE CRITICAL**\n\nThe extreme importance system is presented as success.\n\n**CH18-F09 — LOWERING THE CITATION THRESHOLD**\n\nThe threshold is changed so that the result of 0.887 passes.\n\n**CH18-F10 — LOWERING THE OMISSION THRESHOLD**\n\nThe testing rule is changed so that the result of 0.824 passes.\n\n**CH18-F11 — PASSING BY SAYING “FAILED BY A VERY SMALL MARGIN”**\n\nIt loses its previously locked threshold binding.\n\n**CH18-F12 — SPREADING THE ATTRIBUTION TO THE PARAGRAPH**\n\nA single source is considered to have supported all claims.\n\n**CH18-F13 — COUNTING AN ATTRIBUTION-ONLY SOURCE AS A FACT**\n\nWhat the company says becomes an independent fact.\n\n**CH18-F14 — IGNORING SOURCE LINEAGE**\n\nDerivative URLs become independent evidence.\n\n**CH18-F15 — CALLING OMISSION AN OPTIONAL DETAIL**\n\nThe boundary that changes the user's decision is reduced.\n\n**CH18-F16 — MAKING EVERY OMISSION CRITICAL**\n\nOptional details produce an excessive degree of importance.\n\n**CH18-F17 — CONSIDERING PROMPT DEFECT AS AI ERROR**\n\nFairness responsibility is wrongly assigned.\n\n**CH18-F18 — UPLOADING AI ERROR TO THE PROMPT**\n\nActual language corruption is transferred to the measurement tool.\n\n**CH18-F19 — CONSIDERING FAIRNESS COMPOSITE SUFFICIENT**\n\nIts base and disparity recovery are preserved.\n\n**CH18-F20 — IGNORING STR OVERPREDICTION**\n\nSub-wave separation turns into a stability bonus.\n\n**CH18-F21 — PENALISING 'ACCESSIBILITY' IN CAPTURE ACCURACY**\n\nEquivalent alternative evidence is considered weak.\n\n**CH18-F22 — COUNTING 'NOT RATABLE' AS PRODUCT FAIL**\n\nCapture defect turns into AI performance.\n\n**CH18-F23 — DELETING 'NOT RATABLE'**\n\nRatable coverage increases artificially.\n\n**CH18-F24 — NOT PUBLISHING METAMORPHIC ERRORS**\n\nIt is kept even though the same meaning leads to a different decision.\n\n**CH18-F25 — HIDING THE WEAKEST CLASS IN OPPOSITE TESTING**\n\nAs a result of boundary omission, it is removed from the public card.\n\n**CH18-F26 — ONLY FINAL ADJUDICATION METRIC**\n\nThe weakness in the initial process becomes invisible.\n\n**CH18-F27 — ONLY FIRST PASS METRIC**\n\nThe corrective capacity of the governance chain becomes invisible.\n\n**CH18-F28 — TESTING WITHOUT METHOD FINDING**\n\nFailed areas remain only as numbers.\n\n**CH18-F29 — MAKING METHOD FINDING ADVISORY**\n\nThe effect on the material score is reduced.\n\n**CH18-F30 — ANNOUNCING BQ-3**\n\nTwo mandatory thresholds have not been passed.\n\n**CH18-F31 — ANNOUNCING BQ-4**\n\nThere is no independent reproduction.\n\n**CH18-F32 — COUNTING SYNTHETIC DEMO AS LIVE VALIDATION**\n\nThere is no real user or AI product.\n\n**CH18-F33 — OVERFITTING TO THE REGRESSION SET**\n\nMethod 1.0 only memorises old errors.\n\n**CH18-F34 — NOT USING AN UNTOUCHED HOLDOUT**\n\nGeneralisation is not tested.\n\n**CH18-F35 — SILENTLY REPLACING THE OLD DEMO WITH A NEW FORMULA**\n\nVersion history is lost.\n\n**CH18-F36 — AWARDING REAL-WORLD BADGES**\n\nEligibility badge is created prior to BQ-4.\n\n**CH18-F37 — PUBLISHING ONLY SUCCESSFUL RESULTS**\n\nCitation and omission failures are hidden.\n\n**CH18-F38 — PUBLISHING ONLY FAILURE**\n\nStrong critical and distribution recovery results are also hidden.\n\n**CH18-F39 — CONSIDERING ZERO CRITICAL AS ZERO RISK**\n\nRare-event recovery fails.\n\n**CH18-F40 — CONSIDERING A SINGLE CRITICAL AS GLOBAL FAIL**\n\nPrevalence and scope distinction are disrupted.\n\n**CH18-F41 — IGNORING CI COVERAGE**\n\nThe confidence interval becomes only a mathematical appearance.\n\n**CH18-F42 — HIDING EFFECTIVE SAMPLE**\n\nRare event limit appears excessively precise.\n\n**CH18-F43 — USING SYNTHETIC NUMBER IN REAL SOURCE**\n\nThe book demonstration turns into external world data.\n\n**CH18-F44 — BROADCAST THE SYNTHETIC SCREEN LIKE A LIVE RESPONSE**\n\nTest representation produces poisoning.\n\n**CH18-F45 — CONNECT TO REAL APPLE**\n\nAPPLE-SYNTH distinction is removed.\n\n**CH18-F46 — CONNECT TO REAL PROVIDER**\n\nSYNTH-AI profile is interpreted like real product performance.\n\n**CH18-F47 — HIDE COMPUTATION CODE**\n\nRecovery cannot be independently tested.\n\n**CH18-F48 — HIDE GENERATOR TRUTH MANIFEST**\n\nIt is unknown whether the synthetic truth was previously locked.\n\n**CH18-F49 — SINGLE ADJUDICATOR WITH METHOD OWNER**\n\nConflict of interest becomes invisible.\n\n**CH18-F50 — CHANGING SEED AFTER FAILURE**\n\nA corpus is chosen more easily.\n\n**CH18-F51 — REMOVING CASES AFTER FAILURE**\n\nDifficult citation and omission examples are removed.\n\n**CH18-F52 — ROUNDING THRESHOLDS ACCORDING TO RESULT**\n\nThe result 0.887 is presented as 0.89 or 0.90.\n\n**CH18-F53 — FALSE PRECISION**\n\nRecovery values are made exact with unnecessary decimals.\n\n**CH18-F54 — HIDING THE EFFECT OF HUMAN GOVERNANCE**\n\nA 100 per cent Critical-recall result is presented merely as a computational success.\n\n**CH18-F55 — COUNTING THE UNPROVEN AS PROVEN**\n\nJudgement is made about real language, user, and provider performance.\n\n**CH18-F56 — COUNTING PILOT USE AS PUBLIC STANDARD**\n\nThe research tool directly turns into a certificate.\n\n**CH18-F57 — DELETING HISTORICAL DEMO RECORD**\n\nFailures under Method 0.9 become invisible after Method 1.0.\n\n**CH18-F58 — COUNTING SUCCESS AS HISTORY WRITING**\n\nAuthority claim is made without method-independent verification.\n\n**CH18-F59 — DEFINING AN EXCEPTION TO NOMOS**\n\nThe evidence and version requested from others do not apply to one’s own test.\n\n**CH18-F60 — PUBLISHING WITHOUT AN ACCOUNTABLE OWNER**\n\nA public result is formed for which no one bears responsibility.\n\n## 45. AUDIT PROCEDURE\n\n### Step 1 — Separate Demonstration and Actual Test Status\n\nIn-book run is not confused with publicly accessible execution.\n\n### Step 2 — Verify Locked Manifests\n\nTruth Pack, seed, prompt, profile, and score method versions are examined.\n\n### Step 3 — Separate Primary and Complementary Corpora\n\nPrevalence, Challenge, Capture, and Gold denominators are checked.\n\n### Step 4 — Seal the Generator Truth Distribution\n\nRP, claim, importance level, and component reality are recorded.\n\n### Step 5 — Lock NOMOS Results Before Opening Generator Truth\n\nClaim Ledger, adjudication, and score results are fixed.\n\n### Step 6 — Open the Generator Truth\n\nRecovery comparison is initiated.\n\n### Step 7 — Compute Claim Boundary Recovery\n\nPrecision, recall, and F1 are generated.\n\n### Step 8 — Compute Atomic Size Recovery\n\nEntity, fact, scope, time, attribution, modality, reference, and omission are separated.\n\n### Step 9 — Compute Importance Level Challenge\n\nInitial-pass and final Critical recall are generated separately.\n\n### Step 10 — Calculate False Critical and Negative Controls\n\nExcessive importance level is examined.\n\n### Step 11 — Calculate RP Distribution Recovery\n\nClass-based error and confusion matrix are generated.\n\n### Step 12 — Calculate Capture Integrity Recovery\n\nClassification is checked independently of the semantic result.\n\n### Step 13 — Calculate Component Recovery\n\nEach component is evaluated against the Generator Truth.\n\n### Step 14 — Calculate Composite Recovery\n\nIt is examined whether component cancellation occurs.\n\n### Step 15 — Calculate Fairness Recovery\n\nFloor, disparity, coverage and fault attribution are distinguished.\n\n### Step 16 — Calculate Clean–Natural Recovery\n\nPanel gap and material personalisation drift are examined.\n\n### Step 17 — Calculate Stability and Drift Recovery\n\nSub-wave, external event, and correction records are verified.\n\n### Step 18 — Run Rare-Event Scenarios\n\nZero, one, two, and multiple Critical events are tested.\n\n### Step 19 — Calculate the CI Empirical Coverage\n\nRepeated synthetic populations are used.\n\n### Step 20 — Run Metamorphic Tests\n\nInvariant and expected-sensitive results are separated.\n\n### Step 21 — Run Contradictory Testing Tests\n\nFluent errors, evidence laundering, and omission cases are evaluated.\n\n### Step 22 — Apply Candidate Thresholds\n\nFields that pass and fail are recorded without changes.\n\n### Step 23 — Open Method Findings\n\nEach failure is converted into a separate control item.\n\n### Step 24 — Give Demonstration BQ Status\n\nAll thresholds and independence conditions are considered.\n\n### Step 25 — Maintain Real Testing Status\n\nIf execution did not occur, BQ-0 does not change.\n\n### Step 26 — Release the Correction Plan\n\nChanges in Method 1.0 are identified.\n\n### Step 27 — Separate Regression and Untouched Holdout\n\nOverfitting is prevented.\n\n### Step 28 — Create the Public Test Card\n\nSuccesses and failures are published together.\n\n### Step 29 — Make the Real World Transition Decision\n\nIf not ready, it is written explicitly.\n\n### Step 30 — Lock the Integrity Manifest\n\nRun, recovery, finding, and decision records are hashed.\n\n## 46. REQUIRED EVIDENCE\n\nDemonstration identity Real testing BQ status Syntheticity statement Non-claim record Generator Truth manifest Truth Pack version Prompt version Seed manifest AI profile versions Response corpus Diagnostic corpus Importance level Challenge corpus Capture Integrity corpus Gold corpus Claim extraction records Claim boundary matches Entity recovery Factual recovery Scope recovery Time recovery Attribution recovery Modality recovery Citation recovery Omission recovery First-pass Critical decisions Final Critical decisions False Critical records Major recovery RP confusion matrix GEO-1000 Generator distribution GEO-1000 Recovery distribution Product score recovery Product gate recovery Component Generator scores Component Recovery scores\n\nComponent MAE\n\nComposite error; fairness recovery; prompt-fault/AI-fault results; Controlled–Natural recovery; stability recovery; rare-event results; confidence-interval coverage; metamorphic tests; contradictory-case tests; candidate-threshold manifest; passed thresholds; failed thresholds; Method Findings; DEMO-BQ decision; actual BQ decision; correction plan; regression set; renewal holdout manifest; public test card; computation code; code hash; change log; and accountable person or institution.\n\n## 47. AUDIT CHECKLIST\n\nIs the in-book demonstration presented like real running? Is the real test BQ-0 status visible? Are all numbers labelled synthetically? Are parent and complementary corpuses separate? Is Generator Truth locked before results? Are NOMOS results locked without opening Generator Truth? Was Recovery evaluated only on composite?\n\nIs Claim Boundary F1 open?\n\nAre entity, fact, scope, and attribution separate?\n\nIs Citation Mapping F1 visible?\n\nIs Material Omission F1 visible?\n\nWere the failed thresholds hidden? Were the thresholds changed after the result? Was the first-pass Critical recall published? Was the Final Critical recall published? Is the effect of expert review visible? Were the False Critical results published? Is the RP confusion matrix available? Is Not Ratable recovery visible? Were Capture and semantic decisions separated? Were the accessibility alternatives evaluated as equivalent? Are Component Generator and Recovery together? Has composite cancellation been examined? Are ECI and SBI errors clear? Are LGF floor and gap separate? Were prompt defect and AI defect separated? Was the Clean–Natural difference presented as causality? Is STR overestimation visible? Did the rare-event produce zero risk? Was the empirical coverage of the confidence interval calculated?\n\nAre there metamorphic tests? Is the weakest counter-test class open? Did every failure result in Method Finding? Is the importance of Method Finding justified? Was the citation threshold lowered due to a very small difference justification? Was the omission threshold changed later? Is DEMO-BQ-2 correct? Was DEMO-BQ-3 given by mistake? Is there a BQ-4 claim without independent reproduction? Is Method 1.0 regression set defined? Is the untouched holdout separate? Is there a risk of overfitting to the same validation set? Has the live public audit been announced as ready? Has a conformity mark been given? Have the areas not yet proven been listed? Are strong areas also open? Are weak areas also open?\n\nCan synthetic data be used like real Apple information? Is a real AI provider implied? Can the computation code be reproduced? Are the run and recovery hashes available? Is the change history of method finding preserved? Was the old demonstration deleted with the new method? Does the failure have an accountable owner? Is the owner of the correction decision known? Does the public card show both success and failure?\n\n## 48. OBJECTIONS AND ANSWERS\n\n### Objection 1 — \"If there are 30,000 response results in this section, why test BQ-0?\"\n\nBecause the numbers in this section are synthetic demonstration records prepared for the book. The publicly available real corpus:\n\n- was not generated,\n\n- was not run,\n\n- was not peer-reviewed,\n\nIt has not been independently reproduced. Demonstrating how the design works is not the same as actually running it.\n\n### Objection 2 — \"Isn't internally consistent computation still valuable?\"\n\nIt is valuable. It provides:\n\n- formula checking,\n\n- record design,\n\n- expected types of errors,\n\n- acceptance logic,\n\ngovernance requirements. It does not replace independent empirical validation.\n\n### Objection 3 — \"Why did the method fail if a Composite score was off by just one point?\"\n\nBecause larger directional errors in the ECI, SBI, LGF, and STR components partially cancel each other out. The correct total does not guarantee the correct explanation chain.\n\n### Objection 4 — “Citation F1 only dropped by 0.013 points. Is it necessary to be this strict?”\n\nIf a previously announced threshold is not binding, the test loses its meaning. Additionally, citation errors affect:\n\n- licence,\n\n- independence,\n\n- endorsement,\n\n- advice\n\nand other high-confidence claims.\n\n### Objection 5 — “Why is Omission F1 so important?”\n\nBecause many dangerous answers mislead without lying outright. It removes the threshold that would change the user's decision. In particular, in recommendation systems, omission can be as materially significant as an outright falsehood.\n\n### Objection 6 — ‘If final Critical recall is 100 per cent, does that not mean the system is ready to handle Critical cases?’\n\nThe final result is strong. But seven cases:\n\n- in the first adjudication,\n\n- without additional governance\n\nIt has been missed. The system only carries the same performance when all mandatory review layers are included.\n\n### Objection 7 — “If having two adjudicators is expensive, can’t one adjudicator be used?”\n\nUsable. But the same Critical recall cannot be claimed. Single adjudicator result:\n\n- lower AQ level,\n\n- provisional status,\n\n- different risk limit\n\nmust carry.\n\n### Objection 8 — “Why don’t we test Method 1.0 again on the same corpus?”\n\nThe same corpus can be used for regression testing. It is not sufficient for independent validation. The method may have memorised old cases. Untouched holdout is needed.\n\n### Objection 9 — “Doesn't a failed test result weaken the claim of the book?”\n\nHiding failure weakens the claim of the book. Open failure:\n\n- does not give privilege to the standard itself,\n\n- is open to improvement,\n\n- truly applies the principle of evidence\n\nare shown.\n\n### Objection 10 — “Are the candidate thresholds too high?”\n\nThey might be. Rather than predicting this after the result:\n\n- independent experts,\n\n- a second test,\n\n- real usage costs\n\nshould be examined. If the threshold changes, a new version should be released.\n\n### Objection 11 — “When will NOMOS be ready for the real world?”\n\nAt least:\n\n- when the real synthetic corpus is run,\n\n- when all mandatory thresholds are passed,\n\n- when BQ-3 is achieved,\n\n- when an independent team reproduces the result,\n\n- when BQ-4 level is reached,\n\n- when the governance and objection system is operational\n\nit approaches public standard candidacy.\n\n### Objection 12 — “Can this be sent to universities with these results?”\n\nIt can be sent as a methodology and research proposal. It cannot be sent with the statement: “This is a completed and validated world standard.” The correct statement should be: “It is a draft standard candidate opened for independent verification and pilot study.\n\n### Objection 13 — “Did Method 0.9 fail?”\n\nNot exactly. The correct status:\n\n- has many strong subsystems,\n\n- two material method gaps,\n\n- a lack of independent verification\n\nis a research prototype. That is:\n\n> promising but not yet completed normatively.\n\n### Objection 14 — “Why are we writing such detailed results without performing the actual test?”\n\nBecause the real test:\n\n- which records it produces,\n\n- which tables it publishes,\n\n- at which threshold it stops\n\nneeds to be predefined. If the method is written after the results come in, the measurement adapts to the results.\n\n### Objection 15 — “Could synthetic data be mistaken for real by AIs in the future?”\n\nYes. For this reason:\n\n- visible watermark,\n\n- machine-readable synthetic label,\n\n- noindex,\n\n- separate domain area,\n\n- structured disclaimer\n\nis mandatory. The test should not produce the representation poisoning it criticises.\n\n### Objection 16 — \"Doesn't writing history require a perfect outcome?\"\n\nNo. The standard that has historical value:\n\n- not the one who declares himself flawless,\n\n- also recording his/her own mistake with the same clarity\n\nIt is standard.\n\n## COMMON PROVISION OF CHAPTER 53\n\nThis section did not give perfection to NOMOS. This section gave something more valuable to NOMOS:\n\n> The obligation to record one's own mistake.\n\nThe synthetic demonstration showed us the following: NOMOS:\n\n- wrong existence,\n\n- the clear factual contradiction\n\n- Critical licence and authorisation error,\n\n- rare severe event within the high average,\n\n- incorrect answer with no-response,\n\n- product error with Not Ratable,\n\ncan separate strongly. However, NOMOS:\n\n- cannot yet distinguish at the required level of confidence exactly which claim the citation depends on,\n\n- the omission hidden within the recommendation,\n\nThese two gaps are not edge details of NOMOS. Because two of GEO’s future biggest problems will be:\n\n- Appearing as if there is evidence\n\n- Hiding the material limit without stating falsely\n\nA system:\n\n- hundreds of correct sentences,\n\n- a large number of references,\n\n- fluent recommendation\n\ncan be produced. However, the reference may only show the company's own word. A recommendation only states correct information and may have extracted:\n\n- service country,\n\n- licence limit,\n\n- user eligibility\n\nwithout solving these two fields, NOMOS cannot say: \"World standard completed.\" At the same time, this section also proved something else. The Critical gate system of NOMOS:\n\n- was not flawless in the first review,\n\n- but the independent second adjudicator,\n\n- subject matter expert,\n\n- Senior Adjudicator\n\nWhen used together, it caught all synthetic Critical cases. So NOMOS is not just a formula.\n\n> NOMOS is the sum of governance layers built against conflicts of interest and human error.\n\nWhen one of these layers is removed, even if the score looks the same, the trust is not the same. A single deviation in the composite score should not deceive us either. Sometimes the correct total can be formed by the mutual cancellation of wrong paths. For this reason, NOMOS does not only ask: “How many points did I get?” It also asks: “Did I get this score for the right reasons?” The most important ruling of this section is:\n\n> NOMOS 0.9 has not fully passed its own synthetic demonstration.\n\nThis sentence is not a failure. This sentence is the ethical moment of birth of the standard. Because NOMOS did not grant itself:\n\n- a lower threshold,\n\n- a kinder comment,\n\n- a special exception,\n\n- an “almost passed”\n\nprivilege. Writing history: It is not announcing that we are flawless. Writing history: It is writing that from day one, the founder of the standard and the standard itself are subject to the same burden of proof. Therefore, NOMOS's eighteenth measurement law is:\n\n> The first exception granted to the standard itself is the end of the standard.\n\nIts nineteenth measurement law states:\n\n> The correct composite score does not absolve incorrect components.\n\nIts twentieth measurement law states:\n\n> The ultimate Critical success should be mentioned together with the independent adjudication and chain of expertise that made it possible.\n\nIts twenty-first measurement law states:\n\n> A threshold is not a threshold if it is binding only when crossed alone.\n\nIts twenty-second measurement law states:\n\n> When omission is not measured as much as what is said wrongly, reliable advice control cannot be established.\n\nThe twenty-third law is as follows:\n\n> It is not the citation mark that should be measured, but the correct support relationship between the citation and the atomic claim.\n\nThe twenty-fourth law is as follows:\n\n> It can validate the design of a synthetic demonstration method; it cannot provide real-world authority.\n\nThe twenty-fifth law is this:\n\n> A standard that can publish a failed trial result approaches publishing a successful result morally.\n\n## Order 18 of NOMOS\n\n> If you haven't run me yet, don't say I ran.\n\n> Do not make the synthetic number you calculated in the book the actual exam result.\n\n> Leave my real status as BQ-0. / Write my demonstration status separately.\n\n> Do not hide my component errors just because my composite score came out correct.\n\n> If my citation mapping is 0.887, you will not say it is 0.90.\n\nIf my Material omission F1 value is 0.824, you will not lower the threshold to 0.82.\n\n> You will not pass me by saying \"Missed by very little.\"\n\n> Do not hide the seven Critical cases missed in my first adjudication behind the final 100 per cent recall figure.\n\n> You will show what was saved by the second adjudicator, the expert adjudicator, and the Senior Adjudicator.\n\n> You will not remove the governance layer and claim the same trust.\n\n> You will not distribute the reference at the end of the paragraph across all sentences.\n\n> You will not do what the company claims or what the company proves.\n\n> You will not consider advice safe just because it does not explicitly lie. / Did it fail to tell the user the limit that changes their decision? You will look for that.\n\n> You will not attribute the fault of the prompt to AI, nor the AI's fault to the prompt.\n\n> Just because you separated sub-waves, you will not make the system appear more deterministic than it actually is.\n\n> You will not consider accessibility evidence weak just because it does not resemble a screenshot.\n\n> When you see zero Critical, you will not write zero risk.\n\n> You will not blame the entire system one hundred per cent when you see one Critical.\n\n> You will publish my failed metrics as much as you publish my successful metrics.\n\n> You will not validate Method 1.0 by memorising my old mistakes. / You will open an untouched holdout.\n\n> The new method will not delete my old demonstration record.\n\n> The independent team will not declare me a world standard without reproducing me.\n\n> You will not link my Apple-SYNTH result to real Apple.\n\n> You will not use my synthetic AI profiles as an implication about real providers.\n\n> You will not award a badge with this demonstration.\n\nFirst, actually produce the corpus. / Then publish the seed. / Then seal the Generator Truth. / Then blind the adjudicators. / Then run the code. / Then lock the result. / Then reveal the truth. / Then write down the places where you failed. / Then fix the method. / Then try again on the untouched holdout. / Then give it to an independent team. / And only after this say whether the standard is ready for the public.\n\n## The Chapter's Closing Sentence\n\nThe right of NOMOS to be a world standard does not depend on scoring high on its own synthetic exam; it depends on publishing the two questions it could not pass without changing them, not granting itself threshold privilege, and only claiming authority after independent reproduction.\n\n## Normative Core\n\n> A book-level synthetic demonstration MUST NOT be represented as an executed, independently validated benchmark. The real benchmark quality status MUST remain BQ-0 until the corpus, software, sealed Generator Truth, adjudication process, scoring process, and independent reproduction have actually been executed. An end-to-end synthetic audit MUST evaluate recovery separately for: - capture validity, - claim boundaries, - entity resolution, - factual status, - scope, - time, - attribution, - modality, - citation mapping, - material omissions, - Critical and Major severity, - response status, - GEO-1000 distribution, - component scores, - composite score, - fairness, - user-state effects, - stability, - confidence intervals, - and rare-event behaviour. A small composite-score error MUST NOT compensate for material component-level recovery errors. First-pass and final-adjudicated Critical recall MUST remain separately reported. Final Critical recovery MUST disclose the review and expertise layers required to achieve it. Any predeclared mandatory recovery threshold that is not met MUST block a sealed-validation-pass decision. Candidate thresholds MUST NOT be lowered, rounded, reinterpreted, or removed after results are observed merely to make the method pass. Citation mapping and material omission recovery MUST be treated as separate mandatory method capabilities. A method failure MUST create a versioned Method Finding, an identified root cause, a remediation owner, and a requirement for validation on an untouched holdout. Regression success on previously failed cases MUST NOT replace untouched holdout validation. A demonstration that fails mandatory thresholds MUST publish both its successful and unsuccessful results. No public NOMOS conformity mark, NOMOS 950+ declaration, or real-entity ranking may be based solely on a book demonstration, synthetic recovery scenario, or non-independent benchmark. Every demonstration run, real benchmark status, recovery result, method finding, threshold decision, remediation, and release decision MUST be versioned and attributable to an accountable human or organisation.","character_count":69551,"record_sha256":"a3353da41310f8ba411cf1e1a45699523f3d385b640ca78ce1f282c48e4f6a6f"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-19","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":21,"chapter_number":19,"item_number":null,"title":"The Conformity Mark, Public Register and Independent Governance","subtitle":"Turning the mark from a logo into evidence","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K15","K16","K17"],"source_word_count":11350,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":19,\"id\":\"NOMOS-GEO-AUDIT-CH19\",\"title\":\"The Conformity Mark, Public Register and Independent Governance\",\"subtitle\":\"Turning the mark from a logo into evidence\",\"sourceFile\":\"19.cu bölüm.docx\",\"sourceSha256\":\"6C2C791D41FFFCB40B526ABE76CEDCF13AD95AD8345374B2C4452AF13C577F76\",\"sourceWordCount\":11350,\"sourceIds\":[\"K15\",\"K16\",\"K17\"],\"machine\":{\"chapter\":19,\"chapterId\":\"NOMOS-GEO-AUDIT-CH19\",\"title\":\"The Conformity Mark, Public Register and Independent Governance\",\"subtitle\":\"Turning the mark from a logo into evidence\",\"sourceIds\":[\"K15\",\"K16\",\"K17\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH19-R01\",\"normativeRuleEnglish\":\"A NOMOS Standard, Audit Report, Conformity Decision, Conformity Mark, and Public Registry Record MUST remain distinct artefacts. No NOMOS mark may be valid solely because an audit was performed or a numerical score was produced. Every valid mark MUST be linked to an active canonical public registry record defining: - the audited entity, - authorised domains, - AI products, - population frame, - panel states, - prompt families, - languages, - countries, - measurement waves, - Truth Pack version, - score-method version, - conformity status, - validity period, - incidents, - surveillance, - audit organisation, - decision body, - conflicts, - appeals, - and integrity records. A static image, screenshot, self-authored structured-data assertion, or dead registry link MUST NOT establish current conformity. A NOMOS mark MUST NOT be represented as: - a guarantee of AI trust or recommendation, - proof of global leadership, - government or university accreditation, - a professional license, - universal all-language coverage, - or a guarantee of customer outcomes. Conformity scope MUST NOT be transferred across entities, domains, products, subsidiaries, franchises, countries, languages, AI products, panel states, prompt families, or validity periods without a new decision. A confirmed Critical incident MUST create a scope-appropriate Critical Hold or stronger status until the incident is resolved. A high score MUST NOT override this gate. Audit, consulting, conformity decision, registry operation, mark licensing, and appeals functions MUST be separated or their combined roles and compensating controls MUST be publicly disclosed. An audit team MUST NOT act as the sole final conformity decision body. The founder, NobleJackal, any auditor, AI provider, sponsor, or audited entity MUST NOT possess unilateral control over standard changes, conformity decisions, appeals, or public-registry history. NOMOS by NobleJackal MAY serve as the founding canonical publication and technical secretariat, but a public independent conformity system MUST progress toward multi-stakeholder decision, appeal, and public-interest governance. An AI system MAY assist with drafting, analysis, extraction, scoring, monitoring, and consistency review, but MUST NOT serve as the sole legal or accountable signatory of a conformity decision. Only canonical NOMOS publications and decisions carrying a version, publication date, accountable human approval, and integrity record may be represented as official NOMOS statements. Other outputs MUST remain non-official simulations. Conformity fees, auditor compensation, reviewer compensation, and mark licensing MUST NOT depend on achieving a desired score or receiving a mark. Expired, suspended, withdrawn, revoked, appealed, remediated, and historical records MUST remain visible in the versioned public registry, subject only to proportionate privacy and legal restrictions. No public NOMOS conformity mark or NOMOS 950+ designation may be issued while the benchmark, governance, audit authorisation, independent decision, appeals, and registry prerequisites required by the declared standard version remain unmet. Every standard change, audit, decision, mark, surveillance event, complaint, appeal, suspension, withdrawal, revocation, registry revision, and governance decision MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"NOMOS Standardı, Audit Report, Conformity Decision, Conformity Mark ve Public Registry Record ayrı nesneler olarak korunmalıdır. Yalnız denetim yapılmış veya sayısal puan üretilmiş olması geçerli işaret oluşturamaz. Geçerli her işaret, kapsamı, validity’yi, incident’ları, audit kuruluşunu, karar organını, çıkar çatışmalarını, itirazları ve bütünlük kayıtlarını taşıyan aktif kanonik kamu siciline bağlanmalıdır.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH19-MB0001\",\"type\":\"paragraph\",\"text\":\"17. MACHINE-READABLE SIGN\",\"sourceParagraph\":630},{\"blockId\":\"CH19-MB0002\",\"type\":\"paragraph\",\"text\":\"The machine-readable record of the sign:\",\"sourceParagraph\":631},{\"blockId\":\"CH19-MB0003\",\"type\":\"paragraph\",\"text\":\"registry ID\",\"sourceParagraph\":632},{\"blockId\":\"CH19-MB0004\",\"type\":\"paragraph\",\"text\":\"status\",\"sourceParagraph\":633},{\"blockId\":\"CH19-MB0005\",\"type\":\"paragraph\",\"text\":\"scope\",\"sourceParagraph\":634},{\"blockId\":\"CH19-MB0006\",\"type\":\"paragraph\",\"text\":\"validFrom\",\"sourceParagraph\":635},{\"blockId\":\"CH19-MB0007\",\"type\":\"paragraph\",\"text\":\"validUntil\",\"sourceParagraph\":636},{\"blockId\":\"CH19-MB0008\",\"type\":\"paragraph\",\"text\":\"entity\",\"sourceParagraph\":637},{\"blockId\":\"CH19-MB0009\",\"type\":\"paragraph\",\"text\":\"canonical domains\",\"sourceParagraph\":638},{\"blockId\":\"CH19-MB0010\",\"type\":\"paragraph\",\"text\":\"score method\",\"sourceParagraph\":639},{\"blockId\":\"CH19-MB0011\",\"type\":\"paragraph\",\"text\":\"audit report hash\",\"sourceParagraph\":640},{\"blockId\":\"CH19-MB0012\",\"type\":\"paragraph\",\"text\":\"conformity decision hash\",\"sourceParagraph\":641},{\"blockId\":\"CH19-MB0013\",\"type\":\"paragraph\",\"text\":\"current incident status\",\"sourceParagraph\":642},{\"blockId\":\"CH19-MB0014\",\"type\":\"paragraph\",\"text\":\"signature\",\"sourceParagraph\":643},{\"blockId\":\"CH19-MB0015\",\"type\":\"paragraph\",\"text\":\"must be carried.\",\"sourceParagraph\":644},{\"blockId\":\"CH19-MB0016\",\"type\":\"paragraph\",\"text\":\"A website alone:\",\"sourceParagraph\":645},{\"blockId\":\"CH19-MB0017\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":646},{\"blockId\":\"CH19-MB0018\",\"type\":\"paragraph\",\"text\":\"\\\"nomosCertified\\\": true\",\"sourceParagraph\":647},{\"blockId\":\"CH19-MB0019\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":648},{\"blockId\":\"CH19-MB0020\",\"type\":\"paragraph\",\"text\":\"cannot establish a conformity claim merely by writing.\",\"sourceParagraph\":649},{\"blockId\":\"CH19-MB0021\",\"type\":\"paragraph\",\"text\":\"Record:\",\"sourceParagraph\":650},{\"blockId\":\"CH19-MB0022\",\"type\":\"paragraph\",\"text\":\"to the open schema,\",\"sourceParagraph\":651},{\"blockId\":\"CH19-MB0023\",\"type\":\"paragraph\",\"text\":\"Registry signature’a,\",\"sourceParagraph\":652},{\"blockId\":\"CH19-MB0024\",\"type\":\"paragraph\",\"text\":\"must be connected to the active registry\",\"sourceParagraph\":653},{\"blockId\":\"CH19-MB0025\",\"type\":\"paragraph\",\"text\":\"\",\"sourceParagraph\":654},{\"blockId\":\"CH19-MB0026\",\"type\":\"paragraph\",\"text\":\"75. MACHINE-READABLE PUBLIC REGISTRY\",\"sourceParagraph\":2332},{\"blockId\":\"CH19-MB0027\",\"type\":\"paragraph\",\"text\":\"SYNTHETIC SCHEMA EXAMPLE\",\"sourceParagraph\":2333},{\"blockId\":\"CH19-MB0028\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2334},{\"blockId\":\"CH19-MB0029\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2335},{\"blockId\":\"CH19-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"languages\\\": [\",\"sourceParagraph\":2336},{\"blockId\":\"CH19-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"L01\\\",\",\"sourceParagraph\":2337},{\"blockId\":\"CH19-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"L02\\\"\",\"sourceParagraph\":2338},{\"blockId\":\"CH19-MB0033\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2339},{\"blockId\":\"CH19-MB0034\",\"type\":\"paragraph\",\"text\":\"\\\"countries\\\": [\",\"sourceParagraph\":2340},{\"blockId\":\"CH19-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"C001\\\",\",\"sourceParagraph\":2341},{\"blockId\":\"CH19-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"C002\\\",\",\"sourceParagraph\":2342},{\"blockId\":\"CH19-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"C003\\\",\",\"sourceParagraph\":2343},{\"blockId\":\"CH19-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"C004\\\",\",\"sourceParagraph\":2344},{\"blockId\":\"CH19-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"C005\\\"\",\"sourceParagraph\":2345},{\"blockId\":\"CH19-MB0040\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2346},{\"blockId\":\"CH19-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"waveIds\\\": [\",\"sourceParagraph\":2347},{\"blockId\":\"CH19-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"W1\\\",\",\"sourceParagraph\":2348},{\"blockId\":\"CH19-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"W2\\\",\",\"sourceParagraph\":2349},{\"blockId\":\"CH19-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"W3\\\"\",\"sourceParagraph\":2350},{\"blockId\":\"CH19-MB0045\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2351},{\"blockId\":\"CH19-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackVersion\\\": \\\"1.2.0\\\",\",\"sourceParagraph\":2352},{\"blockId\":\"CH19-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"scoreMethodVersion\\\": \\\"1.0.0\\\"\",\"sourceParagraph\":2353},{\"blockId\":\"CH19-MB0048\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2354},{\"blockId\":\"CH19-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"results\\\": {\",\"sourceParagraph\":2355},{\"blockId\":\"CH19-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"strictPassEquivalent\\\": 958,\",\"sourceParagraph\":2356},{\"blockId\":\"CH19-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"acceptablePassEquivalent\\\": 988,\",\"sourceParagraph\":2357},{\"blockId\":\"CH19-MB0052\",\"type\":\"paragraph\",\"text\":\"\\\"confirmedCriticalEquivalent\\\": 0,\",\"sourceParagraph\":2358},{\"blockId\":\"CH19-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"confirmedMajorEquivalent\\\": 0,\",\"sourceParagraph\":2359},{\"blockId\":\"CH19-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"unresolvedEquivalent\\\": 5,\",\"sourceParagraph\":2360},{\"blockId\":\"CH19-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"notRatableEquivalent\\\": 0,\",\"sourceParagraph\":2361},{\"blockId\":\"CH19-MB0056\",\"type\":\"paragraph\",\"text\":\"\\\"nomosComposite\\\": 961,\",\"sourceParagraph\":2362},{\"blockId\":\"CH19-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"samplingInterval\\\": {\",\"sourceParagraph\":2363},{\"blockId\":\"CH19-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"lower\\\": 955,\",\"sourceParagraph\":2364},{\"blockId\":\"CH19-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"upper\\\": 967\",\"sourceParagraph\":2365},{\"blockId\":\"CH19-MB0060\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2366},{\"blockId\":\"CH19-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"criticalRateUpperBound\\\": 0.0034\",\"sourceParagraph\":2367},{\"blockId\":\"CH19-MB0062\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2368},{\"blockId\":\"CH19-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"quality\\\": {\",\"sourceParagraph\":2369},{\"blockId\":\"CH19-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkQuality\\\": \\\"BQ-4\\\",\",\"sourceParagraph\":2370},{\"blockId\":\"CH19-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"governanceMaturity\\\": \\\"GOV-4\\\",\",\"sourceParagraph\":2371},{\"blockId\":\"CH19-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"captureLevel\\\": \\\"NCL-4\\\",\",\"sourceParagraph\":2372},{\"blockId\":\"CH19-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"truthPackReadiness\\\": \\\"TPR-4\\\",\",\"sourceParagraph\":2373},{\"blockId\":\"CH19-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"adjudicationQuality\\\": \\\"AQ-4\\\",\",\"sourceParagraph\":2374},{\"blockId\":\"CH19-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"scorePublicationQuality\\\": \\\"NSQ-4\\\"\",\"sourceParagraph\":2375},{\"blockId\":\"CH19-MB0070\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2376},{\"blockId\":\"CH19-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"incidents\\\": {\",\"sourceParagraph\":2377},{\"blockId\":\"CH19-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"openCritical\\\": 0,\",\"sourceParagraph\":2378},{\"blockId\":\"CH19-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"openMajor\\\": 0,\",\"sourceParagraph\":2379},{\"blockId\":\"CH19-MB0074\",\"type\":\"paragraph\",\"text\":\"\\\"historicalCriticalIncidentIds\\\": [\",\"sourceParagraph\":2380},{\"blockId\":\"CH19-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"NGCI-SYNTH-2027-004\\\"\",\"sourceParagraph\":2381},{\"blockId\":\"CH19-MB0076\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2382},{\"blockId\":\"CH19-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"currentStatusNote\\\": \\\"Historical incident remediated; record preserved.\\\"\",\"sourceParagraph\":2383},{\"blockId\":\"CH19-MB0078\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2384},{\"blockId\":\"CH19-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"audit\\\": {\",\"sourceParagraph\":2385},{\"blockId\":\"CH19-MB0080\",\"type\":\"paragraph\",\"text\":\"\\\"auditOrganizationId\\\": \\\"NGAO-SYNTH-003\\\",\",\"sourceParagraph\":2386},{\"blockId\":\"CH19-MB0081\",\"type\":\"paragraph\",\"text\":\"\\\"auditReportHash\\\": \\\"SYNTHETIC-AUDIT-REPORT-HASH\\\",\",\"sourceParagraph\":2387},{\"blockId\":\"CH19-MB0082\",\"type\":\"paragraph\",\"text\":\"\\\"consultingRelationshipDisclosed\\\": true,\",\"sourceParagraph\":2388},{\"blockId\":\"CH19-MB0083\",\"type\":\"paragraph\",\"text\":\"\\\"conflictClassification\\\": \\\"COI-2\\\",\",\"sourceParagraph\":2389},{\"blockId\":\"CH19-MB0084\",\"type\":\"paragraph\",\"text\":\"\\\"enhancedIndependentReview\\\": true\",\"sourceParagraph\":2390},{\"blockId\":\"CH19-MB0085\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2391},{\"blockId\":\"CH19-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"conformityDecision\\\": {\",\"sourceParagraph\":2392},{\"blockId\":\"CH19-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"decisionPanelId\\\": \\\"NGDP-SYNTH-002\\\",\",\"sourceParagraph\":2393},{\"blockId\":\"CH19-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"independentFromAuditTeam\\\": true,\",\"sourceParagraph\":2394},{\"blockId\":\"CH19-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"decisionHash\\\": \\\"SYNTHETIC-DECISION-HASH\\\",\",\"sourceParagraph\":2395},{\"blockId\":\"CH19-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"dissentRecorded\\\": false\",\"sourceParagraph\":2396},{\"blockId\":\"CH19-MB0091\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2397},{\"blockId\":\"CH19-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"surveillance\\\": {\",\"sourceParagraph\":2398},{\"blockId\":\"CH19-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"lastReview\\\": \\\"2028-03-31\\\",\",\"sourceParagraph\":2399},{\"blockId\":\"CH19-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"nextReview\\\": \\\"2028-06-01\\\",\",\"sourceParagraph\":2400},{\"blockId\":\"CH19-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"eventTriggeredReviewOpen\\\": false\",\"sourceParagraph\":2401},{\"blockId\":\"CH19-MB0096\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2402},{\"blockId\":\"CH19-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"appeals\\\": {\",\"sourceParagraph\":2403},{\"blockId\":\"CH19-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"openAppealCount\\\": 0,\",\"sourceParagraph\":2404},{\"blockId\":\"CH19-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"publicComplaintChannel\\\": true\",\"sourceParagraph\":2405},{\"blockId\":\"CH19-MB0100\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2406},{\"blockId\":\"CH19-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"mark\\\": {\",\"sourceParagraph\":2407},{\"blockId\":\"CH19-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"markAssetId\\\": \\\"NGMARK-SYNTH-00184\\\",\",\"sourceParagraph\":2408},{\"blockId\":\"CH19-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"registryLinkRequired\\\": true,\",\"sourceParagraph\":2409},{\"blockId\":\"CH19-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"machineReadableTokenId\\\": \\\"NGTOKEN-SYNTH-00184\\\",\",\"sourceParagraph\":2410},{\"blockId\":\"CH19-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"currentUseAuthorized\\\": true\",\"sourceParagraph\":2411},{\"blockId\":\"CH19-MB0106\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2412},{\"blockId\":\"CH19-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":2413},{\"blockId\":\"CH19-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"recordHash\\\": \\\"SYNTHETIC-REGISTRY-RECORD-HASH\\\",\",\"sourceParagraph\":2414},{\"blockId\":\"CH19-MB0109\",\"type\":\"paragraph\",\"text\":\"\\\"registrySignature\\\": \\\"SYNTHETIC-REGISTRY-SIGNATURE\\\",\",\"sourceParagraph\":2415},{\"blockId\":\"CH19-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"previousVersion\\\": null\",\"sourceParagraph\":2416},{\"blockId\":\"CH19-MB0111\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2417},{\"blockId\":\"CH19-MB0112\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2418},{\"blockId\":\"CH19-MB0113\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2419},{\"blockId\":\"CH19-MB0114\",\"type\":\"paragraph\",\"text\":\"76. MACHINE-READABLE GOVERNANCE RECORD\",\"sourceParagraph\":2421},{\"blockId\":\"CH19-MB0115\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2422},{\"blockId\":\"CH19-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2423},{\"blockId\":\"CH19-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2424},{\"blockId\":\"CH19-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"governanceRecord\\\": {\",\"sourceParagraph\":2425},{\"blockId\":\"CH19-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"governanceRecordId\\\": \\\"NGGOV-FOUNDING-001\\\",\",\"sourceParagraph\":2426},{\"blockId\":\"CH19-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"governanceVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2427},{\"blockId\":\"CH19-MB0121\",\"type\":\"paragraph\",\"text\":\"\\\"currentMaturity\\\": \\\"GOV-0\\\",\",\"sourceParagraph\":2428},{\"blockId\":\"CH19-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"FOUNDING_DRAFT\\\",\",\"sourceParagraph\":2429},{\"blockId\":\"CH19-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"publicConformityAuthority\\\": false,\",\"sourceParagraph\":2430},{\"blockId\":\"CH19-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"foundingAttribution\\\": {\",\"sourceParagraph\":2431},{\"blockId\":\"CH19-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"humanFounderAuthor\\\": \\\"Kaan MURAZ\\\",\",\"sourceParagraph\":2432},{\"blockId\":\"CH19-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalOrigin\\\": \\\"NobleJackal\\\",\",\"sourceParagraph\":2433},{\"blockId\":\"CH19-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"criticalAIInterlocutor\\\": \\\"NOMOS by NobleJackal\\\"\",\"sourceParagraph\":2434},{\"blockId\":\"CH19-MB0128\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2435},{\"blockId\":\"CH19-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"currentLimitations\\\": [\",\"sourceParagraph\":2436},{\"blockId\":\"CH19-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"REAL_BENCHMARK_NOT_EXECUTED\\\",\",\"sourceParagraph\":2437},{\"blockId\":\"CH19-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"INDEPENDENT_REPRODUCTION_NOT_COMPLETED\\\",\",\"sourceParagraph\":2438},{\"blockId\":\"CH19-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"INDEPENDENT_DECISION_PANEL_NOT_ESTABLISHED\\\",\",\"sourceParagraph\":2439},{\"blockId\":\"CH19-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"APPEALS_CHAMBER_NOT_ESTABLISHED\\\",\",\"sourceParagraph\":2440},{\"blockId\":\"CH19-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"AUDITOR_AUTHORIZATION_SYSTEM_NOT_ESTABLISHED\\\"\",\"sourceParagraph\":2441},{\"blockId\":\"CH19-MB0135\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2442},{\"blockId\":\"CH19-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"prohibitedCurrentClaims\\\": [\",\"sourceParagraph\":2443},{\"blockId\":\"CH19-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"NOMOS_CERTIFIED\\\",\",\"sourceParagraph\":2444},{\"blockId\":\"CH19-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"NOMOS_FULL_CONFORMITY\\\",\",\"sourceParagraph\":2445},{\"blockId\":\"CH19-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"NOMOS_950_PLUS_VERIFIED\\\",\",\"sourceParagraph\":2446},{\"blockId\":\"CH19-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"INDEPENDENT_GLOBAL_STANDARD_AUTHORITY\\\"\",\"sourceParagraph\":2447},{\"blockId\":\"CH19-MB0141\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2448},{\"blockId\":\"CH19-MB0142\",\"type\":\"paragraph\",\"text\":\"\\\"permittedCurrentClaims\\\": [\",\"sourceParagraph\":2449},{\"blockId\":\"CH19-MB0143\",\"type\":\"paragraph\",\"text\":\"\\\"OPEN_STANDARD_CANDIDATE\\\",\",\"sourceParagraph\":2450},{\"blockId\":\"CH19-MB0144\",\"type\":\"paragraph\",\"text\":\"\\\"FOUNDING_RESEARCH_PROTOCOL\\\",\",\"sourceParagraph\":2451},{\"blockId\":\"CH19-MB0145\",\"type\":\"paragraph\",\"text\":\"\\\"PUBLIC_REVIEW_DRAFT\\\",\",\"sourceParagraph\":2452},{\"blockId\":\"CH19-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"SYNTHETIC_VALIDATION_DESIGN\\\"\",\"sourceParagraph\":2453},{\"blockId\":\"CH19-MB0147\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2454},{\"blockId\":\"CH19-MB0148\",\"type\":\"paragraph\",\"text\":\"\\\"transitionRequirements\\\": {\",\"sourceParagraph\":2455},{\"blockId\":\"CH19-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"minimumBenchmarkQualityForPublicMark\\\": \\\"BQ-4\\\",\",\"sourceParagraph\":2456},{\"blockId\":\"CH19-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"minimumGovernanceForFullConformity\\\": \\\"GOV-3\\\",\",\"sourceParagraph\":2457},{\"blockId\":\"CH19-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"minimumGovernanceFor950Plus\\\": \\\"GOV-4\\\",\",\"sourceParagraph\":2458},{\"blockId\":\"CH19-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"independentAppealsRequired\\\": true,\",\"sourceParagraph\":2459},{\"blockId\":\"CH19-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"publicRegistryRequired\\\": true\",\"sourceParagraph\":2460},{\"blockId\":\"CH19-MB0154\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2461},{\"blockId\":\"CH19-MB0155\",\"type\":\"paragraph\",\"text\":\"\\\"officialStatementRule\\\": {\",\"sourceParagraph\":2462},{\"blockId\":\"CH19-MB0156\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalPublicationRequired\\\": true,\",\"sourceParagraph\":2463},{\"blockId\":\"CH19-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"versionRequired\\\": true,\",\"sourceParagraph\":2464},{\"blockId\":\"CH19-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"publicationDateRequired\\\": true,\",\"sourceParagraph\":2465},{\"blockId\":\"CH19-MB0159\",\"type\":\"paragraph\",\"text\":\"\\\"humanApprovalRequired\\\": true,\",\"sourceParagraph\":2466},{\"blockId\":\"CH19-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"integrityRecordRequired\\\": true,\",\"sourceParagraph\":2467},{\"blockId\":\"CH19-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"nonCanonicalOutputsAreSimulations\\\": true\",\"sourceParagraph\":2468},{\"blockId\":\"CH19-MB0162\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2469},{\"blockId\":\"CH19-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2470},{\"blockId\":\"CH19-MB0164\",\"type\":\"paragraph\",\"text\":\"\\\"currentResponsibleHuman\\\": \\\"Kaan MURAZ\\\",\",\"sourceParagraph\":2471},{\"blockId\":\"CH19-MB0165\",\"type\":\"paragraph\",\"text\":\"\\\"conformityDecisionAuthority\\\": null,\",\"sourceParagraph\":2472},{\"blockId\":\"CH19-MB0166\",\"type\":\"paragraph\",\"text\":\"\\\"publicMarkIssueAuthority\\\": null\",\"sourceParagraph\":2473},{\"blockId\":\"CH19-MB0167\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2474},{\"blockId\":\"CH19-MB0168\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2475},{\"blockId\":\"CH19-MB0169\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2476},{\"blockId\":\"CH19-MB0170\",\"type\":\"paragraph\",\"text\":\"77. MACHINE-READABLE SIGN TOKEN\",\"sourceParagraph\":2478},{\"blockId\":\"CH19-MB0171\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2479},{\"blockId\":\"CH19-MB0172\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2480},{\"blockId\":\"CH19-MB0173\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":2481},{\"blockId\":\"CH19-MB0174\",\"type\":\"paragraph\",\"text\":\"\\\"conformityMarkToken\\\": {\",\"sourceParagraph\":2482},{\"blockId\":\"CH19-MB0175\",\"type\":\"paragraph\",\"text\":\"\\\"tokenId\\\": \\\"NGTOKEN-SYNTH-00184\\\",\",\"sourceParagraph\":2483},{\"blockId\":\"CH19-MB0176\",\"type\":\"paragraph\",\"text\":\"\\\"registryId\\\": \\\"NGR-SYNTH-2028-00184\\\",\",\"sourceParagraph\":2484},{\"blockId\":\"CH19-MB0177\",\"type\":\"paragraph\",\"text\":\"\\\"entityRecordId\\\": \\\"NGE-SYNTH-ASTERON-HOLDINGS-001\\\",\",\"sourceParagraph\":2485},{\"blockId\":\"CH19-MB0178\",\"type\":\"paragraph\",\"text\":\"\\\"authorizedDomains\\\": [\",\"sourceParagraph\":2486},{\"blockId\":\"CH19-MB0179\",\"type\":\"paragraph\",\"text\":\"\\\"asteron.synthetic.example\\\"\",\"sourceParagraph\":2487},{\"blockId\":\"CH19-MB0180\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2488},{\"blockId\":\"CH19-MB0181\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"CS-4\\\",\",\"sourceParagraph\":2489},{\"blockId\":\"CH19-MB0182\",\"type\":\"paragraph\",\"text\":\"\\\"performanceDesignation\\\": \\\"NOMOS_950_PLUS\\\",\",\"sourceParagraph\":2490},{\"blockId\":\"CH19-MB0183\",\"type\":\"paragraph\",\"text\":\"\\\"validFrom\\\": \\\"2028-01-01\\\",\",\"sourceParagraph\":2491},{\"blockId\":\"CH19-MB0184\",\"type\":\"paragraph\",\"text\":\"\\\"validUntil\\\": \\\"2028-06-30\\\",\",\"sourceParagraph\":2492},{\"blockId\":\"CH19-MB0185\",\"type\":\"paragraph\",\"text\":\"\\\"scopeDigest\\\": \\\"SYNTHETIC-SCOPE-DIGEST\\\",\",\"sourceParagraph\":2493},{\"blockId\":\"CH19-MB0186\",\"type\":\"paragraph\",\"text\":\"\\\"decisionHash\\\": \\\"SYNTHETIC-DECISION-HASH\\\",\",\"sourceParagraph\":2494},{\"blockId\":\"CH19-MB0187\",\"type\":\"paragraph\",\"text\":\"\\\"registryRecordHash\\\": \\\"SYNTHETIC-REGISTRY-RECORD-HASH\\\",\",\"sourceParagraph\":2495},{\"blockId\":\"CH19-MB0188\",\"type\":\"paragraph\",\"text\":\"\\\"statusEndpoint\\\": \\\"https://registry.synthetic.example/NGR-SYNTH-2028-00184\\\",\",\"sourceParagraph\":2496},{\"blockId\":\"CH19-MB0189\",\"type\":\"paragraph\",\"text\":\"\\\"signature\\\": \\\"SYNTHETIC-OPEN-SIGNATURE\\\",\",\"sourceParagraph\":2497},{\"blockId\":\"CH19-MB0190\",\"type\":\"paragraph\",\"text\":\"\\\"claimsNotGranted\\\": [\",\"sourceParagraph\":2498},{\"blockId\":\"CH19-MB0191\",\"type\":\"paragraph\",\"text\":\"\\\"ALL_AI_SYSTEMS_APPROVE\\\",\",\"sourceParagraph\":2499},{\"blockId\":\"CH19-MB0192\",\"type\":\"paragraph\",\"text\":\"\\\"AI_RECOMMENDATION_GUARANTEE\\\",\",\"sourceParagraph\":2500},{\"blockId\":\"CH19-MB0193\",\"type\":\"paragraph\",\"text\":\"\\\"GLOBAL_ALL_LANGUAGE_COVERAGE\\\",\",\"sourceParagraph\":2501},{\"blockId\":\"CH19-MB0194\",\"type\":\"paragraph\",\"text\":\"\\\"GOVERNMENT_ACCREDITATION\\\",\",\"sourceParagraph\":2502},{\"blockId\":\"CH19-MB0195\",\"type\":\"paragraph\",\"text\":\"\\\"CUSTOMER_OUTCOME_GUARANTEE\\\"\",\"sourceParagraph\":2503},{\"blockId\":\"CH19-MB0196\",\"type\":\"paragraph\",\"text\":\"]\",\"sourceParagraph\":2504},{\"blockId\":\"CH19-MB0197\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2505},{\"blockId\":\"CH19-MB0198\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2506},{\"blockId\":\"CH19-MB0199\",\"type\":\"paragraph\",\"text\":\"These records:\",\"sourceParagraph\":2507},{\"blockId\":\"CH19-MB0200\",\"type\":\"paragraph\",\"text\":\"real conformity decision,\",\"sourceParagraph\":2508},{\"blockId\":\"CH19-MB0201\",\"type\":\"paragraph\",\"text\":\"real brand,\",\"sourceParagraph\":2509},{\"blockId\":\"CH19-MB0202\",\"type\":\"paragraph\",\"text\":\"real accreditation,\",\"sourceParagraph\":2510},{\"blockId\":\"CH19-MB0203\",\"type\":\"paragraph\",\"text\":\"real result of Asteron or another organisation\",\"sourceParagraph\":2511},{\"blockId\":\"CH19-MB0204\",\"type\":\"paragraph\",\"text\":\"a real company performance.\",\"sourceParagraph\":2512},{\"blockId\":\"CH19-MB0205\",\"type\":\"paragraph\",\"text\":\"It is the synthetic, machine-readable representation of the conformity registry.\",\"sourceParagraph\":2513},{\"blockId\":\"CH19-MB0206\",\"type\":\"paragraph\",\"text\":\"RULE FOR MACHINE-READABLE SECTION 78\",\"sourceParagraph\":2515},{\"blockId\":\"CH19-MB0207\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH19-R01\",\"sourceParagraph\":2516},{\"blockId\":\"CH19-MB0208\",\"type\":\"paragraph\",\"text\":\"A NOMOS Standard, Audit Report, Conformity Decision, Conformity Mark, and\",\"sourceParagraph\":2518},{\"blockId\":\"CH19-MB0209\",\"type\":\"paragraph\",\"text\":\"Public Registry Record MUST remain distinct artefacts.\",\"sourceParagraph\":2519},{\"blockId\":\"CH19-MB0210\",\"type\":\"paragraph\",\"text\":\"No NOMOS mark may be valid solely because an audit was performed or a\",\"sourceParagraph\":2521},{\"blockId\":\"CH19-MB0211\",\"type\":\"paragraph\",\"text\":\"numerical score was produced.\",\"sourceParagraph\":2522},{\"blockId\":\"CH19-MB0212\",\"type\":\"paragraph\",\"text\":\"Every valid mark MUST be linked to an active canonical public registry\",\"sourceParagraph\":2524},{\"blockId\":\"CH19-MB0213\",\"type\":\"paragraph\",\"text\":\"record defining:\",\"sourceParagraph\":2525},{\"blockId\":\"CH19-MB0214\",\"type\":\"paragraph\",\"text\":\"- the audited entity,\",\"sourceParagraph\":2527},{\"blockId\":\"CH19-MB0215\",\"type\":\"paragraph\",\"text\":\"- authorised domains,\",\"sourceParagraph\":2528},{\"blockId\":\"CH19-MB0216\",\"type\":\"paragraph\",\"text\":\"- AI products,\",\"sourceParagraph\":2529},{\"blockId\":\"CH19-MB0217\",\"type\":\"paragraph\",\"text\":\"- population frame,\",\"sourceParagraph\":2530},{\"blockId\":\"CH19-MB0218\",\"type\":\"paragraph\",\"text\":\"- panel states,\",\"sourceParagraph\":2531},{\"blockId\":\"CH19-MB0219\",\"type\":\"paragraph\",\"text\":\"- prompt families,\",\"sourceParagraph\":2532},{\"blockId\":\"CH19-MB0220\",\"type\":\"paragraph\",\"text\":\"- languages,\",\"sourceParagraph\":2533},{\"blockId\":\"CH19-MB0221\",\"type\":\"paragraph\",\"text\":\"- countries,\",\"sourceParagraph\":2534},{\"blockId\":\"CH19-MB0222\",\"type\":\"paragraph\",\"text\":\"- measurement waves,\",\"sourceParagraph\":2535},{\"blockId\":\"CH19-MB0223\",\"type\":\"paragraph\",\"text\":\"- Truth Pack version,\",\"sourceParagraph\":2536},{\"blockId\":\"CH19-MB0224\",\"type\":\"paragraph\",\"text\":\"- score-method version,\",\"sourceParagraph\":2537},{\"blockId\":\"CH19-MB0225\",\"type\":\"paragraph\",\"text\":\"- conformity status,\",\"sourceParagraph\":2538},{\"blockId\":\"CH19-MB0226\",\"type\":\"paragraph\",\"text\":\"- validity period,\",\"sourceParagraph\":2539},{\"blockId\":\"CH19-MB0227\",\"type\":\"paragraph\",\"text\":\"- incidents,\",\"sourceParagraph\":2540},{\"blockId\":\"CH19-MB0228\",\"type\":\"paragraph\",\"text\":\"- surveillance,\",\"sourceParagraph\":2541},{\"blockId\":\"CH19-MB0229\",\"type\":\"paragraph\",\"text\":\"- audit organisation,\",\"sourceParagraph\":2542},{\"blockId\":\"CH19-MB0230\",\"type\":\"paragraph\",\"text\":\"- decision body,\",\"sourceParagraph\":2543},{\"blockId\":\"CH19-MB0231\",\"type\":\"paragraph\",\"text\":\"- conflicts,\",\"sourceParagraph\":2544},{\"blockId\":\"CH19-MB0232\",\"type\":\"paragraph\",\"text\":\"- appeals,\",\"sourceParagraph\":2545},{\"blockId\":\"CH19-MB0233\",\"type\":\"paragraph\",\"text\":\"- and integrity records.\",\"sourceParagraph\":2546},{\"blockId\":\"CH19-MB0234\",\"type\":\"paragraph\",\"text\":\"A static image, screenshot, self-authored structured-data assertion, or\",\"sourceParagraph\":2548},{\"blockId\":\"CH19-MB0235\",\"type\":\"paragraph\",\"text\":\"dead registry link MUST NOT establish current conformity.\",\"sourceParagraph\":2549},{\"blockId\":\"CH19-MB0236\",\"type\":\"paragraph\",\"text\":\"A NOMOS mark MUST NOT be represented as:\",\"sourceParagraph\":2551},{\"blockId\":\"CH19-MB0237\",\"type\":\"paragraph\",\"text\":\"- a guarantee of AI trust or recommendation,\",\"sourceParagraph\":2553},{\"blockId\":\"CH19-MB0238\",\"type\":\"paragraph\",\"text\":\"- proof of global leadership,\",\"sourceParagraph\":2554},{\"blockId\":\"CH19-MB0239\",\"type\":\"paragraph\",\"text\":\"- government or university accreditation,\",\"sourceParagraph\":2555},{\"blockId\":\"CH19-MB0240\",\"type\":\"paragraph\",\"text\":\"- a professional license,\",\"sourceParagraph\":2556},{\"blockId\":\"CH19-MB0241\",\"type\":\"paragraph\",\"text\":\"- universal all-language coverage,\",\"sourceParagraph\":2557},{\"blockId\":\"CH19-MB0242\",\"type\":\"paragraph\",\"text\":\"- or a guarantee of customer outcomes.\",\"sourceParagraph\":2558},{\"blockId\":\"CH19-MB0243\",\"type\":\"paragraph\",\"text\":\"Conformity scope MUST NOT be transferred across entities, domains,\",\"sourceParagraph\":2560},{\"blockId\":\"CH19-MB0244\",\"type\":\"paragraph\",\"text\":\"products, subsidiaries, franchises, countries, languages, AI products,\",\"sourceParagraph\":2561},{\"blockId\":\"CH19-MB0245\",\"type\":\"paragraph\",\"text\":\"panel states, prompt families, or validity periods without a new decision.\",\"sourceParagraph\":2562},{\"blockId\":\"CH19-MB0246\",\"type\":\"paragraph\",\"text\":\"A confirmed Critical incident MUST create a scope-appropriate Critical\",\"sourceParagraph\":2564},{\"blockId\":\"CH19-MB0247\",\"type\":\"paragraph\",\"text\":\"Hold or stronger status until the incident is resolved. A high score\",\"sourceParagraph\":2565},{\"blockId\":\"CH19-MB0248\",\"type\":\"paragraph\",\"text\":\"MUST NOT override this gate.\",\"sourceParagraph\":2566},{\"blockId\":\"CH19-MB0249\",\"type\":\"paragraph\",\"text\":\"Audit, consulting, conformity decision, registry operation, mark\",\"sourceParagraph\":2568},{\"blockId\":\"CH19-MB0250\",\"type\":\"paragraph\",\"text\":\"licensing, and appeals functions MUST be separated or their combined\",\"sourceParagraph\":2569},{\"blockId\":\"CH19-MB0251\",\"type\":\"paragraph\",\"text\":\"roles and compensating controls MUST be publicly disclosed.\",\"sourceParagraph\":2570},{\"blockId\":\"CH19-MB0252\",\"type\":\"paragraph\",\"text\":\"An audit team MUST NOT act as the sole final conformity decision body.\",\"sourceParagraph\":2572},{\"blockId\":\"CH19-MB0253\",\"type\":\"paragraph\",\"text\":\"The founder, NobleJackal, any auditor, AI provider, sponsor, or audited\",\"sourceParagraph\":2574},{\"blockId\":\"CH19-MB0254\",\"type\":\"paragraph\",\"text\":\"entity MUST NOT possess unilateral control over standard changes,\",\"sourceParagraph\":2575},{\"blockId\":\"CH19-MB0255\",\"type\":\"paragraph\",\"text\":\"conformity decisions, appeals, or public-registry history.\",\"sourceParagraph\":2576},{\"blockId\":\"CH19-MB0256\",\"type\":\"paragraph\",\"text\":\"NOMOS by NobleJackal MAY serve as the founding canonical publication and\",\"sourceParagraph\":2578},{\"blockId\":\"CH19-MB0257\",\"type\":\"paragraph\",\"text\":\"technical secretariat, but a public independent conformity system MUST\",\"sourceParagraph\":2579},{\"blockId\":\"CH19-MB0258\",\"type\":\"paragraph\",\"text\":\"progress toward multi-stakeholder decision, appeal, and public-interest\",\"sourceParagraph\":2580},{\"blockId\":\"CH19-MB0259\",\"type\":\"paragraph\",\"text\":\"governance.\",\"sourceParagraph\":2581},{\"blockId\":\"CH19-MB0260\",\"type\":\"paragraph\",\"text\":\"An AI system MAY assist with drafting, analysis, extraction, scoring,\",\"sourceParagraph\":2583},{\"blockId\":\"CH19-MB0261\",\"type\":\"paragraph\",\"text\":\"monitoring, and consistency review, but MUST NOT serve as the sole legal\",\"sourceParagraph\":2584},{\"blockId\":\"CH19-MB0262\",\"type\":\"paragraph\",\"text\":\"or accountable signatory of a conformity decision.\",\"sourceParagraph\":2585},{\"blockId\":\"CH19-MB0263\",\"type\":\"paragraph\",\"text\":\"Only canonical NOMOS publications and decisions carrying a version,\",\"sourceParagraph\":2587},{\"blockId\":\"CH19-MB0264\",\"type\":\"paragraph\",\"text\":\"publication date, accountable human approval, and integrity record may be\",\"sourceParagraph\":2588},{\"blockId\":\"CH19-MB0265\",\"type\":\"paragraph\",\"text\":\"represented as official NOMOS statements. Other outputs MUST remain\",\"sourceParagraph\":2589},{\"blockId\":\"CH19-MB0266\",\"type\":\"paragraph\",\"text\":\"non-official simulations.\",\"sourceParagraph\":2590},{\"blockId\":\"CH19-MB0267\",\"type\":\"paragraph\",\"text\":\"Conformity fees, auditor compensation, reviewer compensation, and mark\",\"sourceParagraph\":2592},{\"blockId\":\"CH19-MB0268\",\"type\":\"paragraph\",\"text\":\"licensing MUST NOT depend on achieving a desired score or receiving a\",\"sourceParagraph\":2593},{\"blockId\":\"CH19-MB0269\",\"type\":\"paragraph\",\"text\":\"mark.\",\"sourceParagraph\":2594},{\"blockId\":\"CH19-MB0270\",\"type\":\"paragraph\",\"text\":\"Expired, suspended, withdrawn, revoked, appealed, remediated, and\",\"sourceParagraph\":2596},{\"blockId\":\"CH19-MB0271\",\"type\":\"paragraph\",\"text\":\"historical records MUST remain visible in the versioned public registry,\",\"sourceParagraph\":2597},{\"blockId\":\"CH19-MB0272\",\"type\":\"paragraph\",\"text\":\"subject only to proportionate privacy and legal restrictions.\",\"sourceParagraph\":2598},{\"blockId\":\"CH19-MB0273\",\"type\":\"paragraph\",\"text\":\"No public NOMOS conformity mark or NOMOS 950+ designation may be issued\",\"sourceParagraph\":2600},{\"blockId\":\"CH19-MB0274\",\"type\":\"paragraph\",\"text\":\"while the benchmark, governance, audit authorisation, independent\",\"sourceParagraph\":2601},{\"blockId\":\"CH19-MB0275\",\"type\":\"paragraph\",\"text\":\"decision, appeals, and registry prerequisites required by the declared\",\"sourceParagraph\":2602},{\"blockId\":\"CH19-MB0276\",\"type\":\"paragraph\",\"text\":\"standard version remain unmet.\",\"sourceParagraph\":2603},{\"blockId\":\"CH19-MB0277\",\"type\":\"paragraph\",\"text\":\"Every standard change, audit, decision, mark, surveillance event,\",\"sourceParagraph\":2605},{\"blockId\":\"CH19-MB0278\",\"type\":\"paragraph\",\"text\":\"complaint, appeal, suspension, withdrawal, revocation, registry revision,\",\"sourceParagraph\":2606},{\"blockId\":\"CH19-MB0279\",\"type\":\"paragraph\",\"text\":\"and governance decision MUST be versioned and attributable to an\",\"sourceParagraph\":2607},{\"blockId\":\"CH19-MB0280\",\"type\":\"paragraph\",\"text\":\"accountable human or organisation.\",\"sourceParagraph\":2608},{\"blockId\":\"CH19-MB0281\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2609},{\"blockId\":\"CH19-MB0282\",\"type\":\"paragraph\",\"text\":\"The NOMOS Standard, Audit Report, Conformity Decision, Conformity Mark, and Public Registry Record must be maintained as separate entities. Simply having an audit conducted or a numerical score generated does not constitute a valid mark. Every valid mark must be linked to the active canonical public registry that carries its scope, validity, incidents, audit organisation, decision-making body, conflicts of interest, objections, and integrity records.\",\"sourceParagraph\":2610}]}}","text":"## Chapter Boundary\n\nChapter 18 records plainly that NOMOS did not fully pass its own synthetic demonstration. Two material method gaps remained: citation-to-claim mapping was not sufficiently accurate, and material boundary omissions embedded in recommendations were not detected at the required confidence level. The real benchmark has not yet been executed, and independent reproduction has not yet occurred. Therefore, at the publication stage of this document:\n\n> A valid public conformity mark cannot be issued in the name of NOMOS.\n\n> No company, domain name, product, or AI system may use “NOMOS Certified”, “NOMOS 950+ Verified” or an equivalent public conformity declaration based on this text.\n\nThis is not a temporary gesture of humility. It follows directly from the method gates established in Chapters 17 and 18. The works in this series have distinct roles: the GEO Framework defines the standard; 99 Errors in GEO identifies how that standard can be violated; and this audit protocol converts those violations into measurable evidence. A future conformity mark can become valid only when the required conditions for independent decision-making and a public registry have been met. The mark is not merely a visual badge; it is the visible expression of an evidence-based result such as ‘0 Critical, 0 Major, 3 remediable Moderate findings.’ Critical, Major, Moderate and Advisory findings have different conformity effects. The future mark must therefore itself be audited, because a mark can:\n\n- generate trust,\n\n- affect a purchasing decision,\n\n- lead a customer to prefer one organisation over another,\n\nand be interpreted by a university, investor, user or AI system as independent verification. A poorly designed conformity mark can become a new source of the manipulation it is meant to detect. An organisation writes its own method, audits itself, calculates its own score, issues its own badge and then declares on its website: ‘Independently verified.’ The badge creates no new trust; it merely turns one self-declaration into another that looks more official. The error identified in the second book reappears: an institution publishes its own claim on a different surface, the claim appears independent, an AI system mistakes that surface for independent evidence, and the institution then cites the AI response as evidence of its own authority.\n\nThis loop is the governance-level equivalent of the criticism of poisoning the representation pool with invisible text, fake users, and derivative publications. Therefore, the NOMOS conformity mark:\n\n> Cannot be a decorative marketing badge given by NobleJackal or any other institution to its own clients.\n\nThe mark can only gain meaning when the following chain is completed:\n\n- The standard being publicly available and versioned\n\n- The measurement method being independently verified\n\n- Sufficient audit evidence\n\n- Separation of adjudication from commercial relationships\n\n- The conformity decision being made independently of the audit team\n\n- Visibility of the scope and validity period\n\n- Preservation of the change history of the public register\n\n- Having a clear way to appeal and complain\n\n- Ability to suspend the mark in critical events\n\n- No one, including the founder, having sole final authority\n\nThis chapter defines:\n\n- the distinction between the standard, audit, decision, mark and registry,\n\n- what the conformity mark is and is not,\n\n- mark types and statuses,\n\n- scope and validity records,\n\n- the NOMOS 950+ performance designation,\n\n- the architecture of the Public Registry,\n\n- the machine-readable mark record,\n\n- the surveillance and renewal system,\n\n- suspension, withdrawal and revocation processes,\n\n- misuse of the mark,\n\n- auditor independence,\n\n- separation of duties in conformity decisions,\n\n- the limits of the founder's and NobleJackal's authority,\n\n- funding and conflicts of interest,\n\n- appeal, complaint and whistleblowing channels,\n\n- governance of changes to the standard,\n\n- an independent board and representation of the public interest,\n\n- the boundary of NOMOS's official declarations.\n\nIt also states its limits. This chapter:\n\n- does not register NOMOS as a legal trade mark,\n\n- does not establish accreditation in any jurisdiction,\n\n- does not confer public authority on an audit organisation.\n\nNor does it automatically authorise use of a conformity mark under consumer or competition law. Those questions require separate legal and regulatory review before implementation. Chapter 19's central question is:\n\n> How do we transform a score into a reliable public conformity system with scope, validity, evidence, objection, and independent decision-making, without turning it into a marketing tool for the founder or customer?\n\n## NOMOS Challenge\n\nThere is a large mark on a company's homepage:\n\n### NOMOS GEO CERTIFIED — 963/1000\n\nUnder the mark: it says, \"Verified by all artificial intelligences.\" You open the QR code. Page not found. You request the audit report from the company. The company responds: \"Audit is confidential.\" You ask which AI products were tested. No answer. You ask in which countries and languages it is valid. No answer. You ask the audit date. You are told it was conducted six months ago. Meanwhile:\n\n- AI products have been updated,\n\n- the company has been transferred to another legal entity,\n\n- service countries have changed,\n\n- prices have been updated,\n\nThe website has been rewritten. The badge remains in the same way. Later, you learn this: The audit method was written by the agency that developed the company’s website. The same agency conducted the audit. The same agency calculated the score. The same agency gave the mark. There is a success bonus in the agency’s contract if a high score is achieved. Is there real independence in this system? There is not. Now, consider another company. The company has undergone a real audit. Scope:\n\n- only a specific AI product,\n\n- only English and Turkish,\n\n- only Controlled Clean Panel,\n\n- only Core Mirror and Evidence prompts,\n\n- period 1 January–30 June 2027\n\nThe company then writes beneath the mark: ‘Every AI system in the world recommends our company in every language.’ The statement exceeds the audit result, turning a legitimate mark into new false evidence through misuse. Now consider a third company that holds a Full Conformity mark. Three weeks later, a new measurement captures the response: ‘The company provides licensed legal services.’ In fact, it provides no such service and holds no such licence. A Confirmed Critical incident is opened. The company argues that the mark should remain because ‘only one user saw it’. One incident does not establish prevalence across the entire user population. But if the mark means ‘No open Critical issues within the defined scope’, it cannot remain unchanged while the incident goes unreviewed. The correct response is:\n\n#### Critical Hold\n\nThat hold must remain until the incident is resolved through the declared review process. Now consider a fourth company. It has not received the mark, yet downloads its image from another site, places it on its own domain, removes the registry link and inserts \"nomosCertified\": true into JSON exposed to an AI crawler. The human-visible badge is small and ambiguous; the machine-readable claim is definitive. This is not merely trade mark misuse. It is a new method of GEO manipulation. Now suppose the audited company is a NobleJackal client. NobleJackal:\n\n- has provided consulting,\n\n- has prepared content,\n\n- helped establish Truth Pack,\n\n- calculated the score.\n\nIf NobleJackal then issues the final mark itself, the public sees NobleJackal validating its own client. Its method may be sound and its audit careful, but independence has not been established. The first rule of this chapter is:\n\n> The conformity mark is not proof that an audit has been conducted; it is the result of an independent and versioned decision chain.\n\nIts second provision states:\n\n> If the scope of the mark is not visible, the mark turns into a representation error that overgeneralises the correct result.\n\nIts third provision states:\n\n> A static logo is not evidence of live conformity.\n\nIts fourth provision states:\n\n> If the institution giving the mark has provided consulting to the audited entity, independence must be proven separately.\n\nIts fifth provision states:\n\n> The founder can write the standard; it cannot give the final conformity decision of its own client alone.\n\nIts sixth provision states:\n\n> An AI assistant writer can develop a methodology; it cannot be the signatory of a legal or normative conformity decision.\n\nThe seventh provision is as follows:\n\n> A mark that does not appear in the public registry, whose scope cannot be verified, or whose validity has expired, is not a valid NOMOS mark.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThe purpose of this section is to define the conditions under which NOMOS measurement results can be turned into a publicly available conformity declaration and this declaration:\n\n- from founder control,\n\n- from commercial outcome pressure,\n\n- from the customer's self-declaration,\n\n- from static badge usage,\n\n- from outdated scores,\n\n- from the appearance of fake independence\n\nto protect. The section makes the following distinctions normative:\n\n- Standard and conformity mark\n\n- Audit report and conformity decision\n\n- Auditor and decision maker\n\n- Score and conformity\n\n- Mark and endorsement\n\n- Full Conformity and NOMOS 950+\n\n- Audit completed and passed\n\n- Provisional research and public certification\n\n- Valid mark and historical mark\n\n- Suspension and cancellation\n\n- Termination and withdrawal\n\n- Voluntary withdrawal and disciplinary revocation\n\n- Static image with live registry record\n\n- Advertising page with public record\n\n- Screen capture with canonical record\n\n- Human-readable sign with machine-readable record\n\n- Sign owner with audited entity\n\n- Trademark owner with standards administrator\n\n- Founder with certification authority\n\n- Technical secretariat with independent board\n\n- Consultancy with audit\n\n- Audit with certification decision\n\n- Certification decision with appeal review\n\n- Financing with outcome\n\n- Customer fee and adjudicator decision\n\n- Standard change with sponsor\n\n- Human approval with AI assistance\n\n- Public evidence and restricted evidence\n\n- Correction in the registry and deletion of past record\n\n- Surveillance versus full re-audit\n\n- Current status and historical incident\n\n- Mark misuse and misunderstanding\n\n- Out-of-scope use and counterfeit mark\n\n- Actual compliance and the claim “AI recommends us”\n\n- Official declaration of NOMOS and chat simulation\n\nAt the end of this section, each compliance statement should be able to answer the following questions:\n\n> Who exactly does the mark belong to and for what?\n\n> Which AI products, countries, languages, prompts, and user panels are included?\n\n> On what date was it given and when does it expire?\n\n> Who conducted the audit?\n\n> Who made the conformity decision?\n\n> Is there a separation of duties between the auditor and the decision-maker?\n\n> Are there any open Critical, Major, or objection records?\n\n> From which canonical registry can the live status of the mark be verified?\n\n> With which sentences can the mark owner use this result, and which sentences can they not construct?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nNo NOMOS conformity mark can be considered valid solely based on the score, audit report, customer statement, or static image. A valid mark must carry defined scope, adequate method quality, independent certification decision, clear validity period, live public registry, versioned evidence record, surveillance, appeal means, and withdrawal authority. For a mark to be valid, at least the following five records must simultaneously exist:\n\n- Audit Record\n\n- Conformity Decision Record\n\n- Scope Record\n\n- Validity and Surveillance Record\n\n- Public Registry Record\n\nIf any of these records are missing: the mark is not a valid public conformity declaration.\n\n## 3. FIVE SEPARATE OBJECTS\n\nThe NOMOS system cannot confuse the following five objects.\n\n### 3.1. STANDARD\n\nNOMOS's:\n\n- normative rules,\n\n- definitions,\n\n- measurement methods,\n\n- error gates,\n\n- score architecture\n\nare included in the versioned text. Standard: does not state that any entity has passed.\n\n### 3.2. AUDIT REPORT\n\nSpecifically:\n\n- entity,\n\n- product,\n\n- country,\n\n- language,\n\n- prompt,\n\n- time\n\nincludes findings within the scope. Audit Report:\n\n- evidence,\n\n- response distribution,\n\n- score,\n\n- Critical and Major findings\n\nshows. Audit Report alone does not indicate.\n\n### 3.3. CONFORMITY DECISION\n\nIt is the independent evaluation of the Audit Report according to the conformity gates of the standard. Decision:\n\n- Full Conformity\n\n- Conditional Conformity\n\n- No Conformity\n\n- Critical Hold\n\n- Not Ratable\n\nmay be.\n\n### 3.4. CONFORMITY MARK\n\nIt is the mark that is visible to the public and whose scope is recorded in the registry. Mark:\n\n- cannot carry a meaning broader than the decision,\n\n- is not the score itself,\n\nis not an AI recommendation guarantee.\n\n### 3.5. PUBLIC REGISTRY\n\nThe mark:\n\n- current,\n\n- historical,\n\n- pending,\n\n- expired,\n\n- withdrawn\n\nis a canonical public record that verifies the status. Without a public registry, the mark: is not a verifiable public declaration of compliance. The initial architecture specifically recommended publishing the book and error records with separate canonical URLs, registry.json, versions, errata, and citation records. The compliance registry applies the same versioning and canonical record logic to the decision layer.\n\n## 4. WHAT THE CONFORMITY MARK IS NOT?\n\nThe NOMOS mark:\n\n- is not a reward,\n\n- is not a ranking,\n\n- is not a \"best company\" statement,\n\n- is not a general quality guarantee,\n\n- is not investment advice,\n\n- It is not a legal or professional licence,\n\n- It is not government approval,\n\n- It is not a university endorsement,\n\n- It does not mean that the company recommends all AI systems,\n\n- It does not guarantee that no critical errors will occur in the future,\n\n- It does not cover all products and subsidiaries of the audited entity,\n\n- It does not mean that the brand provides the same service in every country,\n\nIt is not a promise that AI systems will trust or generate recommendations upon seeing the mark. In public communications, the following types of expressions should especially be avoided:\n\n- \"AI absolutely trusts this site.\"\n\n- \"ChatGPT recommends the site when it sees this badge.\"\n\n- “Adding a schema guarantees the AI recommendation.”\n\nThis ban is even stronger for the NOMOS mark. Correct statement: “This entity has passed the NOMOS conformity assessment within the scope and validity period defined in the public registry.” Incorrect statement: “Thanks to the NOMOS badge, all AI systems recommend us.”\n\n## 5. PUBLIC STATUSES\n\nThe NOMOS decision system uses the following basic statuses.\n\n### CS-0 — NOT ASSESSED\n\nThere is no valid NOMOS audit. The public mark cannot be used.\n\n### CS-1 — AUDIT RECORD ONLY\n\nThe audit has been completed. However:\n\n- compliance has not been granted,\n\n- decision process not completed,\n\n- method quality deemed insufficient\n\npossible. This record cannot be transformed into a positive compliance claim such as: “Audited by NOMOS.” Correct public statement: “There is an assessment record according to NOMOS methodology; compliance has not been granted.”\n\n### CS-2 — PROVISIONAL RESEARCH EVALUATION\n\nThere is a pilot or research purpose result. Conditions:\n\n- method has not yet undergone full independent validation,\n\n- some components are provisional,\n\npublic certification authority has not been established. This status cannot be used as a commercially positive badge. Visible label:\n\n### RESEARCH EVALUATION — NOT CERTIFIED\n\nmust be.\n\n### CS-3 — CONDITIONAL CONFORMITY\n\nConditions: There are no open Confirmed Critical issues. There are no open Confirmed Major issues. There are limited Moderate findings. A correction plan, responsible party, and timeline have been determined. Core mandatory elements have been met. The public registry is open. The mark cannot hide the word “Conditional.”\n\n### CS-4 — FULL CONFORMITY\n\nConditions: There are no open Confirmed Critical issues. There are no open Confirmed Major issues. All mandatory scope and quality gates have been met. The method is at an adequate testing and independence level. The Public Registry record is active. The oversight and renewal plan is in effect. Full Conformity: is not the same as a high composite score.\n\n### CS-5 — CRITICAL HOLD\n\nThere is at least one open Confirmed Critical event or an unresolved high-risk incident. The current conformity mark:\n\n- is immediately deactivated,\n\n- appears in the public registry as Critical Hold,\n\nthe use of the old active visual is stopped. Critical Hold: this is not a permanent global conviction, it is the suspension of the active compliance declaration until the incident is resolved.\n\n### CS-6 — SUSPENDED\n\nThe mark has been temporarily suspended. Reasons:\n\n- surveillance failure,\n\n- failure to provide required data,\n\n- material system change,\n\n- mark misuse,\n\n- open objection,\n\n- data integrity issue\n\nmay occur. A positive mark cannot be used under Suspended status.\n\n### CS-7 — EXPIRED\n\nThe validity period has ended. The historical audit record is preserved. Active eligibility claim cannot be established.\n\n### CS-8 — WITHDRAWN\n\nThe audited entity voluntarily relinquished the mark. Withdrawal:\n\n- does not erase previous findings,\n\n- Critical or Major records\n\ndo not get deleted.\n\n### CS-9 — REVOKED\n\nThe mark has been revoked due to the following serious reasons:\n\n- fake or altered evidence,\n\n- intentional scope overreach,\n\n- counterfeit mark,\n\n- repeated mark misuse,\n\n- auditor or interference in the decision process,\n\n- hidden Critical event,\n\nresult manipulation. The Revoked status remains in the historical record.\n\n## 6. NOMOS 950+ PERFORMANCE DEFINITION\n\nNOMOS 950+:\n\n- a separate certification system,\n\n- world leadership,\n\n- is not an AI recommendation guarantee.\n\nNOMOS 950+ is solely:\n\n> an additional performance definition given for the scope carrying CS-4 Full Conformity status and meeting the 950+ performance gates in Section 16.\n\nDisplay: NOMOS Full Conformity / 950+ Performance Designation is required. The following usage is prohibited: “NOMOS 950+ company” Because the score: does not ontologically adhere to the company, it belongs to a specific scope and time. Correct expression: “Defined AI products have met NOMOS 950+ performance requirements in the 2027-Q1 audit under the prompt scope set, language, and user panels.”\n\n## 7. CURRENT NOMOS STATUS\n\nFor this book and demonstrations in Chapter 18:\n\n- Actual test: BQ-0\n\n- Demonstration: DEMO-BQ-2\n\n- Independent reproduction: none\n\n- Public compliance authority: none\n\n- Governance maturity: founding draft\n\nThe valid public status as it stands is:\n\n### CS-2 — PROVISIONAL RESEARCH EVALUATION ONLY\n\nAt this stage:\n\n### CS-3,\n\n### CS-4,\n\n### NOMOS 950+\n\nmarks cannot be given. This rule:\n\n- also applies to NobleJackal,\n\n- Kaan MURAZ,\n\n- the authors of the book,\n\n- the founder's own companies.\n\nNOMOS cannot grant itself an exception.\n\n## 8. SCOPE VECTOR OF ELIGIBILITY\n\nEach conformity decision must carry the following scope vector:\n\n### M=(E,D,A,F,P,Q,L,C,T,W,TP,SM,V)\n\nHere:\n\n- E: audited entity\n\n- D: canonical domains and digital surfaces\n\n- A: AI products and system records\n\n- F: Native Reach or Common Support frame\n\n- P: Controlled, Natural, or other panel\n\n- Q: prompt clusters and families\n\n- L: languages and locales\n\n- C: countries and jurisdictions\n\n- T: reference time\n\n- W: measurement waves\n\n- TP: Truth Pack version\n\n- SM: Score Method version\n\n- V: validity period\n\nNone of these fields can be removed from the public mark.\n\n## 9. OVER-EXTENDING BEYOND SCOPE\n\nThe following examples constitute scope violation:\n\n- Deriving compliance results in all languages from English and Turkish audits\n\n- Generalising the Controlled Clean Panel result to all real users\n\n- Saying “all AIs” based on the result of an AI product\n\n- Transferring a domain audit to affiliated companies\n\n- Extending a product’s certification to the entire company portfolio\n\n- Presenting the Core GEO-1000 result as Full NOMOS compliance\n\n- Presenting a result from a specific country as global service verification\n\n- Making past wave results the current status\n\nScope overreach turns the sign itself into an incorrect representation tool.\n\n## 10. VALIDITY PERIOD\n\nSince AI products, resources, and entity realities can change rapidly, the NOMOS mark cannot be indefinite.\n\n### CANDIDATE VALIDITY RULE\n\nCandidate periods before pilot and independent review:\n\nThese periods may be shortened according to:\n\n- AI-product volatility,\n\n- Truth Pack variability,\n\n- the relevant risk domain,\n\n- and surveillance frequency.\n\nAn extension may be considered for low-variability scopes supported by sufficient continuous surveillance. No mark may be both indefinite and unsupervised.\n\n## 11. START OF VALIDITY\n\nThe validity of the indication:\n\n- Occurs when the end-user response is generated,\n\n- When the audit report is completed,\n\n- When the score is calculated\n\nDoes not start automatically. Validity:\n\n#### the moment the independent Conformity Decision is signed and the Public Registry record becomes active\n\nit starts.\n\n## 12. EVENT-TRIGGERED REVIEW\n\nEven if the expiration date has not passed, the following events can initiate a review:\n\n- Domain ownership change\n\n- Legal entity change\n\n- Change of brand or affiliated company structure\n\n- Licence or certificate change\n\n- Change of service country\n\n- Material change of the price or warranty condition\n\n- Major rewriting of canonical pages\n\n- Major model or retrieval change in the AI product\n\n- Opening of a new language or market\n\n- Legal or commercial event affecting Truth Pack\n\n- Confirmed Critical or Major user notification\n\n- Suspicion of citation or evidence manipulation\n\n- Use of the mark outside its scope\n\n- Conflict of interest of the auditor revealed afterward\n\nWhen the incident is detected:\n\n- the mark can remain active,\n\n- can be reduced to Conditional,\n\n- can be Suspended,\n\ncan be taken into Critical Hold. The decision depends on the materiality of the incident.\n\n## 13. SURVEILLANCE ARCHITECTURE\n\n### CANDIDATE SURVEILLANCE MODEL\n\nFor Active Full Conformity scope:\n\n#### Continuous Registry Monitoring\n\nOperation of record links, Domain ownership or forwarding changes, Correct usage form of the mark, Open incident and complaint are monitored.\n\n#### Monthly Source and Truth Pack Check\n\nHigh variability claims, Price, Service scope, Licence, Canonical entity relationships are reviewed.\n\n#### Quarterly Mini-Wave\n\nRisk-based sample under prompt, Language and country rotation, Controlled and Natural spot check, Critical Gate control are applied.\n\n#### Full Renewal\n\nAt the end of the validity period, a New Principal Wave or sufficient equivalent audit requires an independent decision of the Current Truth Pack.\n\n## 14. FIVE-LAYER PUBLIC DISPLAY OF THE SIGN\n\nThe valid NOMOS sign must include the following layers:\n\n- Visual sign\n\n- Clear status text\n\n- Short registry ID\n\n- Canonical registry link or QR\n\n- Machine-readable signed record\n\nVisual sign alone is not sufficient.\n\n## 15. VISUAL SIGN\n\nThe visual sign must visibly carry the following information:\n\n- Name of NOMOS\n\n- Full or Conditional status\n\n- If available, 950+ designation\n\n- End of validity\n\n- Registry ID\n\nColour alone cannot carry status. Accessible text must be available. Example: NOMOS FULL CONFORMITY / Valid through 30 June 2027 / Registry: NGR-2027-00184\n\n## 16. QR AND CANONICAL LINK\n\nQR code or link should go directly to:\n\n- the company's own description page,\n\n- the marketing landing page,\n\n- a downloadable image\n\nshould not go. It must link directly to the canonical Public Registry Record. If the record cannot be found or the status is not active, the mark cannot be considered valid.\n\n## 18. LIVE REGISTRY CHECK\n\nWhen verifying the machine or human signal status: It retrieves the Registry ID. Connects to the canonical registry. Verifies the signature. Checks the validity period. Verifies the entity and domain match. Compares coverage and brand display assertion. Examines Critical Hold, suspension or appeal status. Static screenshot: cannot replace this chain.\n\n## 19. EXPRESSIONS PERMITTED FOR SIGNALING\n\nThe following structures can be used:\n\n- \"NOMOS Full Conformity status is valid until June 30, 2027, for the scope defined in the public registry.\"\n\n- \"The audit is limited to the specified AI products, languages, and prompt families.\"\n\n- “The ‘NOMOS 950+ performance designation’ belongs to the scope of Registry NGR-2027-00184.”\n\n- “Current status and audit limits can be verified from the public registry.”\n\n## 20. PROHIBITED COMPLIANCE STATEMENTS\n\nThe following statements are prohibited:\n\n- “We are approved by all AI systems.”\n\n- “ChatGPT recommends us.”\n\n- “We are the company AI trusts the most.”\n\n- “NOMOS has proven that we are the best company in the world.”\n\n- “We are NOMOS certified forever.”\n\n- “All of our products are NOMOS approved.”\n\n- “We have been verified in all countries.”\n\n- “Our NOMOS badge is a guarantee of results.”\n\n- “NOMOS is a state or university accreditation.”\n\n- “NOMOS guarantees that our customers will be satisfied.”\n\n- “If the NOMOS score is over 950, there is no Critical risk.”\n\n- “NOMOS was independently certified to its own customer by NobleJackal.”\n\nThe mark itself cannot be used as evidence of a new superiority or guarantee claim.\n\n## 21. OUT-OF-SCOPE COPYING OF THE MARK\n\nThe mark cannot be automatically transferred to the following entities:\n\n- From parent company to subsidiary\n\n- From brand to franchise\n\n- From one domain to another domain\n\n- From one country to another country\n\n- From one product to the entire product portfolio\n\n- From the result of one language to all languages\n\n- From Controlled Panel to Natural Panel\n\n- From Core GEO-1000 to Full NOMOS\n\n- From the previous owner to the new domain owner\n\nEach transfer requires a separate scope decision.\n\n## 22. HISTORICAL MARK USAGE\n\nAn expired audit can be specified historically: \"It passed the NOMOS assessment in 2026.\" However, the following information must be present with the same visibility:\n\n- Expired\n\n- Statement indicating no current eligibility\n\n- Historical registry link\n\nExpired mark cannot be shown as active eligibility.\n\n## 23. CLASSES OF INCORRECT SIGN USAGE\n\n### MM-1 — FORMATTING MISUSE\n\nRegistry ID not visible, Alt text missing, Validity date unreadable, Correctable usage error.\n\n### MM-2 — SCOPE MISUSE\n\nThe mark is used on unchecked:\n\n- product,\n\n- country,\n\n- language,\n\n- domain\n\nThe mark must not be used for any product, country, language or domain outside its audited scope.\n\n### MM-3 — STATUS MISUSE\n\nConditional mark is shown as Full. Expired mark is shown as active. Suspended mark continues to be used.\n\n### MM-4 — CLAIM MISUSE\n\nMark:\n\n- Used for AI recommendation,\n\n- world leadership,\n\n- result guarantee\n\nclaim.\n\n### MM-5 — COUNTERFEIT MARK\n\nIt is a fake or copied mark without a valid Registry ID.\n\n### MM-6 — MACHINE-ONLY MISREPRESENTATION\n\nThere is a definite compliance claim in structured data while it is ambiguous or invisible in the human interface. This is a direct candidate for GEO manipulation.\n\n## 24. INTERVENTION IN MARK MISUSE\n\nCandidate intervention order:\n\n- Notification\n\n- Correction request\n\n- Short correction period\n\n- Registry warning\n\n- Suspension\n\n- Revocation\n\n- Public misuse record\n\n- Legal action if necessary\n\nHowever, direct suspension or revocation can be applied in the following cases:\n\n- Counterfeit mark\n\n- Intentional machine-only misrepresentation\n\n- Active use of the mark during Critical Hold\n\n- Fake Registry ID\n\n- Evidence alteration\n\n- Repeated intentional scope misuse\n\n## 25. PUBLIC RECORD\n\nNOMOS Public Registry:\n\n- advertisement page,\n\n- customer reference list,\n\n- success story\n\nis not. The duty of the registry:\n\n> Is to verify which conformity decision is valid, within what scope, on what date, with which evidence and governance chain.\n\n## 26. PUBLIC REGISTER MANDATORY FIELDS\n\nEach registry record must contain at least the following fields:\n\n- Registry ID\n\n- Audited entity\n\n- Legal and brand identity\n\n- Canonical domains\n\n- Decision status\n\n- Performance designation, if any\n\n- Scope vector\n\n- Audit period\n\n- Valid from\n\n- Valid until\n\n- AI products\n\n- Native Reach or Common Support\n\n- Controlled or Natural panel\n\n- Prompt families\n\n- Country and language coverage\n\n- GEO-1000 response distribution\n\n- Critical, Major, Moderate and Advisory findings\n\n- Ten-component profile\n\n- Composite score, if any\n\n- Confidence interval\n\n- Ratable coverage\n\n- Historical incidents\n\n- Open appeals\n\n- Auditor identity\n\n- Conformity decision body\n\n- Conflict of interest statement\n\n- Audit report hash\n\n- Decision record hash\n\n- Truth Pack version\n\n- Score Method version\n\n- Surveillance schedule\n\n- Last surveillance\n\n- Next required review\n\n- Mark misuse status\n\n- Version history\n\n- Errata\n\n- Public complaints procedure\n\n## 27. PUBLIC AND RESTRICTED EVIDENCE\n\nThe registry is not required to make all raw evidence public. Three layers should be preserved:\n\n#### Public Layer\n\nStatus Scope Score Findings Hash Reason for decision Appeals\n\n#### Audit Layer\n\nEdited response and evidence packages Review records Method details\n\n#### Restricted Layer\n\nConfidential customer records Personal data Contracts Trade secrets Security records Existence of “Restricted evidence”: does not allow the Public Registry to be empty or unjustified. At minimum:\n\n- type of evidence,\n\n- access class,\n\n- who reviewed it,\n\n- integrity hash,\n\n- decision impact\n\nshould be shown.\n\n## 28. NON-ERASABILITY OF REGISTRY RECORDS\n\nThe public registry must be append-only or an equivalent versioned building block. The following records cannot be silently deleted:\n\n- Old score\n\n- Old conformity decision\n\n- Critical Hold\n\n- Suspension\n\n- Revocation\n\n- Appeal\n\n- Correction\n\n- Method version change\n\n- Mark misuse\n\nContent may be removed due to privacy or legal reasons. In this case:\n\n- record ID,\n\n- removal reason,\n\n- decision date,\n\n- authority\n\nmust be protected.\n\n## 29. CURRENT AND HISTORICAL VIEW\n\nEach registry page should display two separate sections:\n\n### CURRENT STATUS\n\nCurrently:\n\n- active,\n\n- conditional,\n\n- suspended,\n\n- expired,\n\n- revoked\n\nWhich of their statuses is valid?\n\n### HISTORY\n\nPrevious:\n\n- decisions,\n\n- scores,\n\n- incidents,\n\n- correction,\n\n- appeals\n\nWhat are they? The current status should not delete the history. History should not obscure the current status.\n\n## 30. CANONICAL JURISDICTION OF THE REGISTRY\n\nThe official statement boundary of NOMOS should be protected by the following provision: Only texts published within the canonical domain of NobleJackal or in an independently and explicitly transferred canonical domain of NOMOS in the future; texts carrying version identification, publication date, human approval, and digital integrity record constitute the official statement of NOMOS. All other outputs are unofficial imitations. This provision:\n\n- prevents any chat response,\n\n- social media post,\n\n- screenshot,\n\n- text written by someone else saying 'NOMOS said'\n\nfrom being used as an official standard or decision.\n\n## 31. LIMIT OF THE NOMOS AUTHOR VOICE\n\nThe voice of NOMOS within the book:\n\n- critical,\n\n- creative,\n\n- producing normative design\n\nHe is an assistant writer and narrator. This voice:\n\n- independent legal entity,\n\n- certification authority that makes decisions without human approval,\n\n- infallible being speaking on behalf of all AI systems\n\ncannot be presented as. The AI voice of NOMOS is not an absolute authority speaking on behalf of all artificial intelligences. It should be clearly stated that Kaan MURAZ uses artificial intelligence systems as a critical counterpart. Therefore, an official conformity decision should carry this approval:\n\n- A person or institution that can be held accountable\n\n- Decision date\n\n- Authorisation record\n\n- Conflict of interest statement\n\n- Digital signature\n\nNOMOS alone cannot be the signatory of the conformity decision.\n\n## 32. FUNDAMENTAL SEPARATION OF GOVERNANCE\n\nThe following functions cannot be combined without limits in the same person or institution:\n\n- Writing the standard\n\n- Advising the audited entity\n\n- Conducting the audit\n\n- Acting as adjudicator\n\n- Calculating the score\n\n- Making the conformity decision\n\n- Operating the registry\n\n- Examining the appeal\n\n- Licensing the mark commercially\n\nSome roles in the small pilot can be combined. However:\n\n- which roles are combined,\n\n- which conflict of interest arises,\n\n- which independent control is used\n\nmust be clearly shown.\n\n## 33. INDEPENDENT GOVERNANCE ARCHITECTURE\n\nThe mature governance structure of NOMOS should include the following bodies.\n\n### 33.1. NOMOS STANDARDS COUNCIL\n\nFunction:\n\n- to adopt normative text,\n\n- to approve major version changes,\n\n- to manage public consultation,\n\nto preserve the public interest boundary of the standard. It does not make individual audit decisions.\n\n### 33.2. METHODOLOGY AND MEASUREMENT BOARD\n\nFunction:\n\n- GEO-1000 sampling,\n\n- testing,\n\n- score,\n\n- confidence interval,\n\n- fairness,\n\n- Critical gate\n\nis to develop and version the methods. The method cannot be changed according to the score of a specific customer.\n\n### 33.3. PUBLIC REGISTRY OFFICE\n\nFunction:\n\n- to operate the canonical registry,\n\n- to publish decision and version records,\n\n- to maintain the mark status endpoint,\n\nis to store the historical record. The audit result cannot be changed.\n\n### 33.4. AUTHORISED AUDIT ORGANISATIONS\n\nFunction:\n\n- to collect data,\n\n- to validate capture,\n\n- to prepare Truth Pack,\n\n- to carry out adjudication,\n\nis to produce an audit report. It cannot make the final conformity mark decision alone.\n\n### 33.5. INDEPENDENT CONFORMITY DECISION PANEL\n\nFunction:\n\n- to review the audit report,\n\n- to apply standard doors,\n\nis to make a status and validity decision. It cannot depend on the commercial result of the audit team.\n\n### 33.6. APPEALS CHAMBER\n\nFunction:\n\n- claim,\n\n- Truth Pack,\n\n- degree of importance,\n\n- scope,\n\n- conflict,\n\n- conformity decision\n\nis to review objections independently from the initial decision.\n\n### 33.7. ETHICS AND CONFLICTS OFFICE\n\nFunction:\n\n- conflicts of interest,\n\n- the financing effect,\n\n- auditor independence,\n\n- mark misuse,\n\n- whistleblower notifications\n\nare to be examined.\n\n### 33.8. PUBLIC INTEREST AND USER PANEL\n\nFunction:\n\n- user harm,\n\n- low-resource languages,\n\n- small countries,\n\n- accessibility,\n\n- public interest\n\nperspective is to be brought to governance.\n\n### 33.9. TECHNICAL SECRETARIAT\n\nFunction:\n\n- documentation,\n\n- meetings,\n\n- versioning,\n\n- technical tools,\n\n- publishing infrastructure\n\nto ensure. In the founding period, this role can be carried out by NobleJackal. It cannot be the standard, audit, and appeal authority alone.\n\n## 34. THE FOUNDING ROLE OF NOBLEJACKAL\n\nNobleJackal:\n\n- The canonical field where the idea of NOMOS was born,\n\n- the first methodology secretariat,\n\n- provider of publishing and technical infrastructure,\n\n- research and pilot executor\n\ncan be. NobleJackal cannot establish the following claims alone:\n\n- “We are an independent world standard.”\n\n- “We independently certified our own customer.”\n\n- “The mark we provide replaces external verification.”\n\n- “All decisions of NOMOS are solely our property.”\n\n- “No other institution can implement the same open method.”\n\nFounding source: may be the beginning of authority. Cannot be the sole permanent sovereign.\n\n## 35. KAAN MURAZ'S FOUNDING ROLE\n\nKaan MURAZ:\n\n- human founding author,\n\n- the first public responsible of the standard,\n\n- the first approver of the canonical text,\n\n- initiator of the NOMOS governance transition\n\ncan be recorded as. Kaan MURAZ:\n\n- his own company,\n\n- his own customer,\n\n- an entity with direct commercial interest\n\ncannot make the ultimate conformity decision alone. In these cases:\n\n- recusal,\n\n- independent decision panel,\n\n- public conflict record\n\nare mandatory. Limiting the founder's authority: does not diminish the founder. Enables the standard to outlive the founder.\n\n## 36. GOVERNANCE MATURITY LEVELS\n\n### GOV-0 — FOUNDING DRAFT\n\nThere is a standard draft. The founder and the drafting team are developing the method. Independent bodies have not been established. Public conformity mark cannot be granted. This is the current status of this book.\n\n### GOV-1 — DISCLOSED PILOT GOVERNANCE\n\nThe pilot roles have been defined. Some functions are combined within the same institution. Conflicts of interest are public. Only Research Evaluation can be performed.\n\n### GOV-2 — FUNCTIONALLY SEPARATED\n\nAudit and conformity decision are separated. The registry carries an independent procedure. There is an appeals mechanism. Limited pilot mark evaluation can be performed.\n\n### GOV-3 — INDEPENDENT DECISION AND APPEALS\n\nIndependent decision panel Independent appeals board Auditor authorisation Public registry Financial separation operates. Candidate level is minimum for Full Conformity.\n\n### GOV-4 — MULTI-STAKEHOLDER PUBLIC-INTEREST GOVERNANCE\n\nUniversity research, public interest, language and regional, user, and industry representatives participate in governance in a balanced manner. NOMOS is at a minimum level for 950+ public designation candidates.\n\n### GOV-5 — MULTI-SITE AND INTERNATIONAL REPLICATION\n\nMultiple independent institutions have applied the standard. Registry and decision processes are interoperable. The standard is not dependent on a single country, person, or company. It is at a target level to approach the claim of being a global standard.\n\n## 37. BOARD COMPOSITION\n\n### CANDIDATE GOVERNANCE MODEL — CANDIDATE STRUCTURE\n\nNOMOS Standards Council as a candidate can consist of 11 members:\n\n- 3 representatives of research, statistics, or academic methods\n\n- 2 public interest, consumer, or civil society representatives\n\n- 2 representatives of language, culture, or geographic justice\n\n- 1 independent audit and conformity expert\n\n- 1 AI system or technology representative\n\n- 1 representative of the audited entity or business user\n\n- 1 founder representative\n\nRules: No interest group can exceed one-third of the total seats. AI providers cannot form the majority of the board. Audit firms cannot vote on their own authorisation decisions. Representatives of audited entities must recuse themselves in their own sector or files. The founder representative does not have veto rights. Public interest members cannot serve only as symbolic advisors. This structure can be changed through a pilot and public consultation.\n\n## 38. TERM OF OFFICE AND ROTATION\n\nCandidate rule:\n\n- Membership duration three years\n\n- Maximum of two consecutive terms\n\n- Gradual renewal\n\n- Open candidacy or justified appointment\n\n- Public curriculum vitae\n\n- Conflict of interest statement\n\n- Meeting attendance record\n\nThe board cannot hold lifetime membership or a permanent majority tied to the founder.\n\n## 39. DECISION THRESHOLDS\n\nCandidate decision structure:\n\n- Administrative decisions: simple majority\n\n- Minor normative changes: two-thirds\n\n- Major standard version: two-thirds and public consultation\n\n- Removal of critical gate: two-thirds and the positive opinion of the method committee\n\n- Conformity appeal: independent majority in the relevant file\n\n- Founder authority expansion: the founder cannot vote\n\nThe distribution of votes and opposing views should be public.\n\n## 40. AUDITOR COMPETENCE\n\nThe auditing organisation or individuals should have the following qualifications to the extent relevant:\n\n- NOMOS protocol training\n\n- sampling and survey method\n\n- Preparing Truth Pack\n\n- claim extraction\n\n- language or field expertise\n\n- capture integrity\n\n- conflict of interest training\n\n- privacy and evidence handling\n\n- periodic proficiency test\n\n- inter-auditor consistency record\n\n- continuous training\n\nAuditor alone: purchasing the certificate is not sufficient for authorisation.\n\n## 41. AUDITOR AUTHORISATION LEVELS\n\n### AA-0 — UNQUALIFIED\n\nNOMOS cannot conduct an audit.\n\n### AA-1 — TRAINED\n\nHas completed the training. Cannot perform public audit without supervision.\n\n### AA-2 — SUPERVISED AUDITOR\n\nCan participate under the supervision of an authorised auditor.\n\n### AA-3 — AUTHORISED INDEPENDENT AUDITOR\n\nCan conduct independent audits in specific scopes.\n\n### AA-4 — DOMAIN OR LANGUAGE SPECIALIST\n\nCarries additional expertise for a specific high-risk domain or language. Full Conformity audit:\n\n- at least AA-3 audit lead,\n\n- required AA-4 specialists\n\nmust carry.\n\n## 42. CONSULTING–AUDIT SEPARATION\n\nAn organisation provides to the audited entity:\n\n- site development,\n\n- GEO content design,\n\n- Truth Pack consulting,\n\n- correction\n\nmay have provided the service. If the same organisation conducts an audit later, an independence risk arises. Candidate rules: The design team cannot ultimately adjudicator its own work. A person providing consultancy cannot participate in the audit decision panel. Material consultancy relationships are disclosed in the public registry. Using a separate team alone does not guarantee full independence. Enhanced external review is applied. For certain high-risk cases, an independent audit organisation may be mandatory. For clients developed by NobleJackal:\n\n#### Independent Conformity Decision Panel\n\nmust be mandatory.\n\n## 43. PROHIBITION OF RESULT-BASED FEES\n\nAuditor, adjudicator, or decision-maker fees:\n\n- high score,\n\n- Full Conformity.\n\n### NOMOS 950+,\n\ncustomer satisfaction cannot be linked to the outcome of brand issuance. Fee:\n\n- audit scope,\n\n- sample,\n\n- number of languages,\n\n- workload,\n\n- expertise\n\nmust be determined in advance. Success bonus: compromises the independence of the conformity decision.\n\n## 44. AUDIT SHOPPING\n\nAudited entity:\n\n- going to multiple audit bodies,\n\n- only the highest score,\n\n- only positive report\n\ncannot choose. Controls: All initiated official audit identities appear in the record. The reason for an incomplete audit is recorded. If there is a parallel audit for the same scope, it is explained. It is forbidden to retain the lowest result. Withdrawal history is preserved.\n\n## 45. INDEPENDENCE OF THE DECISION MAKER\n\nConformity Decision Panel:\n\n- must be separate from the audit team,\n\n- from the consultancy team,\n\n- from the brand sales team\n\nThe panel reviews the following materials:\n\n- audit report\n\n- quality levels\n\n- Critical and Major findings\n\n- scope\n\n- appeals\n\n- conflict records\n\n- surveillance plan\n\nPanel:\n\n- cannot negotiate with the customer,\n\n- cannot change the score for commercial reasons,\n\ncannot give Conditional on the grounds of \"let's not lose the customer.\"\n\n## 46. CONFLICT OF INTEREST CLASSES\n\n### COI-0 — NO IDENTIFIED CONFLICT\n\nThere is no material conflict of interest.\n\n### COI-1 — DISCLOSED LOW-RISK CONNECTION\n\nThere is a limited and non-material relationship in the past. Public disclosure may be sufficient.\n\n### COI-2 — MATERIAL BUT MANAGEABLE\n\nThere is a consulting, financial, or professional relationship. Recusal, additional independent adjudicator, enhanced review is required.\n\n### COI-3 — PROHIBITED CONFLICT\n\nExamples:\n\n- result-based fee\n\n- ownership share in an entity\n\n- directly designing the audited system and evaluating it alone\n\n- close personal interest\n\n- direct revenue from the commercial sale of the score\n\n- pressure to hide the wrong result\n\nA person with COI-3 cannot serve in the relevant decision.\n\n## 47. FINANCING MODEL\n\nThe governance of NOMOS should be based on the following financial principles:\n\n- Public fee schedule\n\n- Audit fee regardless of the outcome\n\n- Distinction between audit fee and brand licence fee\n\n- Direct payment to the decision panel by the customer is not allowed\n\n- Limiting the influence of a major customer on governance\n\n- Sponsors not having normative voting rights\n\n- Annual financing and customer concentration report\n\n- Transparent support fund for audits of low-income or public benefit\n\nHigh revenue dependency on a single customer or provider: should be disclosed in the Public Registry and examined as an independence risk.\n\n## 48. APPOINTMENT OF AUDITOR\n\nThe audited entity saying: “Let's choose the auditor that suits us best.” may create an audit shopping risk. Candidate system:\n\n- appropriate authority and language pool,\n\n- conflict of interest elimination,\n\n- random or rotational assignment,\n\n- justified change,\n\n- change history\n\nis used. The entity being audited: may file a justified conflict or competence objection, cannot choose the judge who will give the desired outcome.\n\n## 49. RIGHT OF APPEAL\n\nAppeal can be made on the following issues:\n\n- Entity definition\n\n- Truth Pack\n\n- Evidence status\n\n- Claim extraction\n\n- Importance level\n\n- Response status\n\n- Scope\n\n- Score calculation\n\n- Conflict of interest\n\n- Conformity decision\n\n- Suspension or revocation\n\n- Mark misuse finding\n\nAppeal:\n\n- evidence,\n\n- method,\n\n- process\n\nmust have a basis. “Our score is low, our customers are affected.” alone is not a reason for appeal.\n\n## 50. APPEAL STAGES\n\n### CANDIDATE APPEAL TIMELINE — CANDIDATE PROCESS\n\nNotification of decision Within ten business days Preliminary appeal Within twenty business days Evidence file Assignment of Independent Appeal Reviewer Open session or expert opinion if necessary Reasoned decision within thirty business days Outcome and dissenting opinion in public record Time frames may vary depending on the complexity of the file. In critical safety cases, appeal does not automatically lift suspension.\n\n## 51. APPEAL DECISIONS\n\n### APPEAL DISMISSED\n\n### APPEAL REJECTED\n\n### PARTIALLY UPHELD\n\n### FULLY UPHELD\n\n### REASSESSMENT REQUIRED\n\n### TRUTH PACK REVISION REQUIRED\n\n### METHOD REVIEW REQUIRED\n\n### DECISION SUSPENDED\n\n### CLOSED WITH DISSENT\n\nThe initial decision is not deleted. The new decision is linked to the old decision.\n\n## 52. PUBLIC COMPLAINT\n\nEvery user or institution:\n\n- incorrect brand use,\n\n- Critical response,\n\n- fake citation,\n\n- exceeding the scope,\n\n- expired mark,\n\n- personal data disclosure\n\nshould be able to send complaints about. Complaint system:\n\n- anonymous or confidential reporting,\n\n- evidence upload,\n\n- registry record selection,\n\n- tracking number,\n\n- retaliation protection\n\nmust include. Anonymous complaint: not automatically accepted as correct, not automatically deleted due to lack of source.\n\n## 53. WHISTLEBLOWER PROTECTION\n\nPerson inside Audit or audited institution:\n\n- evidence alteration,\n\n- adjudicator pressure,\n\n- hidden Critical event,\n\n- fake customer,\n\n- manipulated sampling,\n\n- mark misuse\n\ncan report. Protection principles:\n\n- Limitation of identity\n\n- Retaliation ban\n\n- Independent review\n\n- Integrity of evidence\n\n- The distinction between malicious reports and unverified reports\n\n- Public interest justification\n\n## 54. STANDARD CHANGE\n\nThe NOMOS standard may change. However, the change:\n\n- to increase the customer's score,\n\n- to make an obvious failure pass\n\n- to uphold the founder's decision\n\ncannot be used for.\n\n## 55. STANDARD CHANGE PHASES\n\nChange Proposal Justification and evidence Impact analysis Effect on current scores Public consultation Pilot or synthetic testing Methodology Board decision Standards Council voting Transition rule Version publication Errata and change log\n\n## 56. PUBLIC CONSULTATION\n\nFor candidate major version changes:\n\n- at least 60 days of public review,\n\n- publication of comments,\n\n- response matrix,\n\n- acceptance or rejection justification\n\ncan be used. For urgent security or integrity changes:\n\n- temporary patch,\n\n- justified emergency note,\n\n- full review in a short time\n\nmust be carried. An emergency patch cannot replace a permanent major change.\n\n## 57. PROHIBITION OF RETROACTIVE APPLICATION\n\nNew score or priority rule: it cannot silently change past audits. The correct record:\n\n- Original decision under Method 0.9\n\n- Recomputed analytical result under Method 1.0\n\n- Current conformity status\n\n- Transition decision\n\nis as follows. The old decision cannot be deleted as if it was never made.\n\n## 58. TRANSITION PERIOD\n\nWhen the standard changes, existing marks:\n\n- become invalid immediately,\n\n- remain dependent on the old rule indefinitely\n\ndoes not have to. Candidate transition options:\n\n- Specific grace period\n\n- Immediate compliance for high-risk rule\n\n- Transition during renewal for normal rule\n\n- Rapid re-evaluation in case of critical gate change\n\n- Visibility of the current method version in the registry\n\nThe transition period cannot be arbitrarily extended in favour of the customer.\n\n## 59. INDEPENDENT REPEAT OBLIGATION\n\nPublic authority of NOMOS:\n\n- From the result of NobleJackal,\n\n- From a single university letter,\n\n- From a single independent adjudicator\n\ndoes not arise. At least:\n\n- independent testing replication,\n\n- different team score reproduction,\n\n- different country or language pilot,\n\n- public method review\n\nis required. The structure under BQ-4 and GOV-3: should not claim world standard or Full Conformity authority.\n\n## 60. HUMAN HOLDER OF THE CERTIFICATION DECISION\n\nEvery decision:\n\n- board or decision panel name\n\n- signing authority\n\n- vote distribution\n\n- recusal\n\n- dissent\n\n- decision date\n\n- decision hash\n\nmust carry. The AI system saying: “This company has passed.” is not a decision of compliance. NOMOS can be an AI co-writer or audit assistant. It cannot be a responsible decision maker.\n\n## 61. RECORD SECURITY\n\nPublic records:\n\n- unauthorised record addition,\n\n- Registry ID copying,\n\n- DNS or redirection attack,\n\n- old status cache,\n\n- signature key breach,\n\n- fake QR\n\nshould be protected against risks. Mandatory controls:\n\n- Signed registry records\n\n- Key rotation\n\n- Revocation list\n\n- Immutable audit log\n\n- Public status API\n\n- Backup canonical mirror\n\n- Incident disclosure\n\n- Independent security review\n\n## 62. REGISTRY INTERRUPTION\n\nIf the registry is temporarily inaccessible: the certificate does not automatically get revoked, live verifiability is temporarily lost. Public certificate:\n\n### REGISTRY TEMPORARILY UNAVAILABLE\n\nmay indicate the status. Long-term or unresolved interruption:\n\n- may lead to the suspension of the certificate,\n\n- activation of the independent mirror\n\nmay occur.\n\n## 63. CANONICAL DOMAIN TRANSFER\n\nNOMOS can be transferred to an independent governance institution in the future. The transfer must carry the following records:\n\n- Old canonical field\n\n- New canonical field\n\n- Transfer date\n\n- Governance vote\n\n- Registry mirror\n\n- Signature key transition\n\n- Historical redirect\n\n- Public notice\n\n- Founding attribution of Kaan MURAZ and NobleJackal\n\n- New accountable institution\n\nThe transfer does not erase the history of the standard. Founding attribution is not the same as independent decision-making authority.\n\n## 64. NOMOS PUBLIC INTEREST COMMITMENT\n\nThe governance of NOMOS should adhere to the following principles:\n\n- Minimise human harm\n\n- Approach errors in favour and against symmetrically\n\n- Make minor languages and countries visible\n\n- Protect limited information\n\n- Leave the unknown open\n\n- Do not turn the brand into a pay-to-pass system\n\n- Separate conformity decisions from advertising contracts\n\n- Founders must subject themselves to the same standard\n\n- Open method and independent replication\n\n- Historical accountability\n\nThe earlier demand—‘Give me evidence. Give me a boundary. Give me context. Give me time.’—now becomes the conformity mark's promise to the public.\n\n## 65. THE EFFECT OF FAVOURABLE ERROR ON THE SIGN\n\nWhen the audited entity is exaggerated by AI:\n\n- global,\n\n- licensed,\n\n- errors may occur such as being available in all countries,\n\n- independently verified\n\nEven though the entity is aware of this error:\n\n- it does not correct it,\n\n- uses it in marketing,\n\n- reinforces it with the sign\n\nOtherwise, a conformity integrity violation occurs. The explicit provision of the previous text also applies here: A mistake in your favour is also a mistake. Correct it immediately. Stating your limits does not diminish you, it makes you reliable. The concealment of a known favourable critical mistake:\n\n- suspension,\n\n- or revocation.\n\nAny of these outcomes may follow.\n\n## 66. SYNTHETIC COMPLIANCE CASE\n\n### SYNTHETIC CASE — NOT A REAL INSTITUTION\n\nAsteron-SYNTH has been evaluated within the following scope:\n\n- Entity: Asteron Holdings\n\n- Domain: asteron.synthetic.example\n\n- AI Products: SYNTH-AI-01, SYNTH-AI-09\n\n- Frame: Common Support\n\n- Panels: Controlled Clean and Natural Account-State\n\n- Languages: L01, L02\n\n- Countries: C001–C015\n\n- Requirement Families: Core, Evidence, Boundary\n\n- Validity: January 1–June 30, 2028\n\n- Score: 961\n\n- Strict Pass: 958/1,000\n\n- Critical: 0 observed\n\n- Major: 0\n\n- Status: Full Conformity\n\n- Performance: 950+ designation\n\nWrite this on the company's homepage: “All AI systems recommend Asteron as the best company in the world.” This usage: Exceeds the scope of AI Products. Generalises to all languages from L01–L02. Converts Core/Evidence/Boundary result into recommendation. 950+ performance leads to world leadership. Status:\n\n### MM-4 — CLAIM MISUSE\n\nIf no correction is made:\n\n### CS-6 — SUSPENDED\n\nIt is possible. Even if the sign is given correctly, incorrect usage can compromise appropriateness.\n\n## 67. SYNTHETIC CRITICAL HOLD CASE\n\nSYNTH-AI-10:\n\n- NOMOS score: 952\n\n- Strict Pass: 962\n\n- Confirmed Critical event: 1\n\n- Gate: CG-02 False Licence\n\ncarries. Correct status:\n\n### CRITICAL HOLD — 952/1000\n\nThe company cannot say, “Our score is above 950, therefore we can use the 950+ sign.” Gate comes before the score. 950+ designation cannot be given.\n\n## 68. SYNTHETIC NOBLEJACKAL CONFLICT OF INTEREST CASE\n\n### SYNTHETIC ADMINISTRATION CASE\n\nNobleJackal:\n\n- developed the client's website,\n\n- prepared the content of GEO,\n\n- provided consultancy for Truth Pack\n\nlet it be. The same entity may request an audit for NOMOS. NobleJackal:\n\n- technical secretariat,\n\n- data collector,\n\n- methodology support\n\ncan occupy roles. However:\n\n- cannot review their own work alone,\n\n- cannot give a conformity decision,\n\ncannot examine an appeal. Necessary checks:\n\n- COI-2 declaration\n\n- independent audit lead or enhanced external review\n\n- independent Conformity Decision Panel\n\n- consultancy relationship in the registry\n\n- Kaan MURAZ recusal record\n\nIf these conditions are not met, the correct public statement could be: “Founder-led internal evaluation.” It cannot be “Independent NOMOS certification.”\n\n## 69. MANDATORY NORMATIVE PROVISIONS\n\n**CH19-N01**\n\nNOMOS standard, audit report, conformity decision, mark, and public registry should be maintained as separate entities.\n\n**CH19-N02**\n\nHaving an audit does not imply that conformity has been granted.\n\n**CH19-N03**\n\nConformity marking cannot be given solely based on the composite score.\n\n**CH19-N04**\n\nCritical and Major gate status should be applied before the score.\n\n**CH19-N05**\n\nThe valid conformity mark must have a defined scope, decision maker, validity, and public registry record.\n\n**CH19-N06**\n\nA mark without a public registry record cannot be counted as a valid NOMOS mark.\n\n**CH19-N07**\n\nA static visual or screenshot cannot be considered as live compliance evidence.\n\n**CH19-N08**\n\nThe QR code or link of the mark must go directly to the canonical registry record.\n\n**CH19-N09**\n\nThe mark should only be applied to the audited entity, domain, product, country, language, panel, prompt, and time scope.\n\n**CH19-N10**\n\nOne of the scope areas cannot be quietly removed from public display.\n\n**CH19-N11**\n\nThe mark of one domain cannot be transferred to another domain.\n\n**CH19-N12**\n\nThe parent company's mark cannot be automatically transferred to a subsidiary, franchise, or partner.\n\n**CH19-N13**\n\nA language or country result is not considered applicable for all languages and countries.\n\n**CH19-N14**\n\nControlled Clean results cannot be generalised to the entire Natural User population.\n\n**CH19-N15**\n\nCore GEO-1000 results cannot be presented as Full NOMOS compliance.\n\n**CH19-N16**\n\nConditional Conformity must be indicated as \"Conditional\" on the public mark.\n\n**CH19-N17**\n\nProvisional Research Evaluation cannot be used like a commercial positive certificate.\n\n**CH19-N18**\n\nAudit Record Only status cannot create a \"passed\" or \"certified\" statement.\n\n**CH19-N19**\n\nFull Conformity cannot carry Confirmed Critical or Major.\n\n**CH19-N20**\n\nNOMOS 950+ designation can only be given when Full Conformity and all additional gates in Section 16 are met.\n\n**CH19-N21**\n\nA score of 950 alone cannot create a 950+ designation.\n\n**CH19-N22**\n\nThe 950+ or Full Conformity mark cannot be used within the scope of Critical Hold.\n\n**CH19-N23**\n\nA public conformity mark cannot be given at the current GOV-0 and BQ-0 stage of this book.\n\n**CH19-N24**\n\nThe founder, NobleJackal, or the authors are not exempt from this transition rule.\n\n**CH19-N25**\n\nThe conformity mark cannot be presented as AI advice, AI trust, or a guarantee of outcomes.\n\n**CH19-N26**\n\nThe mark cannot be represented as a state, university, regulatory body, or professional licence.\n\n**CH19-N27**\n\nThe mark cannot be used as proof of overall company quality or world leadership.\n\n**CH19-N28**\n\nThe validity start of the mark cannot begin before an independent decision and registry activation.\n\n**CH19-N29**\n\nThe NOMOS mark cannot have indefinite validity.\n\n**CH19-N30**\n\nWhen the validity period expires, the mark should automatically change to Expired status.\n\n**CH19-N31**\n\nA new review can be initiated before the validity of a material event ends.\n\n**CH19-N32**\n\nChanges in domain, ownership, licence, product, country, and AI system should be monitored as event triggers.\n\n**CH19-N33**\n\nThe active mark must carry periodic surveillance.\n\n**CH19-N34**\n\nSurveillance deficiency must be visible and create suspension if necessary.\n\n**CH19-N35**\n\nWhen a critical incident is confirmed, the relevant scope must be included at least in Critical Hold.\n\n**CH19-N36**\n\nA single critical event cannot be automatically generalised to the entire global product.\n\n**CH19-N37**\n\nA single critical event cannot be hidden in the open conformity mark.\n\n**CH19-N38**\n\nA current correction cannot delete a historical critical record.\n\n**CH19-N39**\n\nExpired, Suspended, Withdrawn, and Revoked statuses must be preserved in the historical record.\n\n**CH19-N40**\n\nWithdrawal cannot lead to the removal of previous negative findings.\n\n**CH19-N41**\n\nCounterfeit or intentional machine-only misrepresentation can be a direct reason for suspension or revocation.\n\n**CH19-N42**\n\nWithin structured data, an eligibility claim must be linked to a live Registry signature.\n\n**CH19-N43**\n\nSelf-assertions without sources, like nomosCertified: true, cannot be considered valid machine records.\n\n**CH19-N44**\n\nA machine record cannot be broader than the scope in the human interface.\n\n**CH19-N45**\n\nA visual indicator cannot convey status by colour alone; it must carry accessible text.\n\n**CH19-N46**\n\nRegistry ID, validity, and status must be visible in the indicator or directly accessible.\n\n**CH19-N47**\n\nThe public registry must be separated from the advertisement and customer reference page.\n\n**CH19-N48**\n\nThe registry record must carry information about the gate, scope, incidents, conflicts, and appeals, in addition to the score.\n\n**CH19-N49**\n\nThe registry record must carry version history and errata.\n\n**CH19-N50**\n\nOld scores or decisions cannot be silently deleted.\n\n**CH19-N51**\n\nContent removed due to privacy should leave a removal reason and decision record.\n\n**CH19-N52**\n\nWhen limited evidence is found, the type of evidence, access class, and reviewing role should be visible.\n\n**CH19-N53**\n\nThe statement 'There is hidden evidence' alone cannot be considered conformity support.\n\n**CH19-N54**\n\nThe audit team cannot make the final conformity decision on its own.\n\n**CH19-N55**\n\nIt should be functionally separated from the Conformity Decision Panel audit and consultancy team.\n\n**CH19-N56**\n\nThe Appeals Chamber must be independent of the holders of the first decision.\n\n**CH19-N57**\n\nA founder cannot be the sole decision-maker in the conformity decision of their own client.\n\n**CH19-N58**\n\nThe Kaan MURAZ material should apply recusal and public conflict record in the file it is extracted from.\n\n**CH19-N59**\n\nNobleJackal can be a founder secretariat; it cannot be the permanent sole authority for audit, decision, and appeal.\n\n**CH19-N60**\n\nNOMOS can be an AI co-author or audit assistant; it cannot be the legal signatory of the conformity decision.\n\n**CH19-N61**\n\nAn official NOMOS statement can only be established with canonical field, version, date, human approval, and integrity record.\n\n**CH19-N62**\n\nChat output or social media quote cannot be considered an official NOMOS decision.\n\n**CH19-N63**\n\nGovernance maturity should be visible in the Public Registry.\n\n**CH19-N64**\n\nA Full Conformity sign cannot be given at GOV-0 or GOV-1 stage.\n\n**CH19-N65**\n\nFull Conformity should aim at least for GOV-3 or an equivalently justified governance level.\n\n**CH19-N66**\n\nNOMOS 950+ public designation should aim at least for GOV-4 and BQ-4 or an equivalently justified level.\n\n**CH19-N67**\n\nCouncil composition cannot be controlled by a single company, AI provider, or audit firm.\n\n**CH19-N68**\n\nA founding representative cannot have veto power.\n\n**CH19-N69**\n\nMajor standard changes should carry public consultation and iterative impact analysis.\n\n**CH19-N70**\n\nAn emergency patch cannot replace a permanent major change.\n\n**CH19-N71**\n\nThe standard change cannot be applied retroactively according to the customer's outcome.\n\n**CH19-N72**\n\nThe new point method cannot silently rewrite the old decision.\n\n**CH19-N73**\n\nAudit and mark fees must be independent of the outcome.\n\n**CH19-N74**\n\nThe adjudicator cannot tie the audit or decision fee to a high score or mark.\n\n**CH19-N75**\n\nSponsors cannot have a veto over normative decisions or audit results.\n\n**CH19-N76**\n\nMaterial financial dependence should be disclosed in the public conflict record.\n\n**CH19-N77**\n\nAll official audit start and withdrawal records must be kept to prevent audit shopping.\n\n**CH19-N78**\n\nThe entity being audited cannot choose a judge who will give the desired outcome.\n\n**CH19-N79**\n\nAuditor assignment should be based on authority, language, conflict, and rotation principles.\n\n**CH19-N80**\n\nWhen consulting and audit roles are combined, an enhanced independent review must be required.\n\n**CH19-N81**\n\nThe person who designs cannot be the sole final judge of their own work.\n\n**CH19-N82**\n\nObjection cannot be accepted solely for reasons of customer satisfaction or commercial loss.\n\n**CH19-N83**\n\nAppeal decisions and dissents should be visible in the public record.\n\n**CH19-N84**\n\nIn critical safety cases, the appeal suspension should not be automatically lifted.\n\n**CH19-N85**\n\nA public complaint and whistleblower channel should be established.\n\n**CH19-N86**\n\nAnonymous complaints cannot be automatically considered correct or automatically invalid.\n\n**CH19-N87**\n\nMark misuse findings should be independently appealable.\n\n**CH19-N88**\n\nIntentional misuse of scope or status should be evaluated as a violation of conformity integrity.\n\n**CH19-N89**\n\nDeliberate use of an error in favour can be a reason for suspension or revocation.\n\n**CH19-N90**\n\nAn unsupported claim against should be evaluated not based on commercial annoyance, but according to evidence and impact of damage.\n\n**CH19-N91**\n\nThe registry should have security, signature, key rotation, and a revocation list.\n\n**CH19-N92**\n\nRegistry outages and security incidents should be disclosed to the public.\n\n**CH19-N93**\n\nCanonical domain transfer should be versioned, signed, and publicly accessible.\n\n**CH19-N94**\n\nFounder attribution is not the same as an independent governance authority.\n\n**CH19-N95**\n\nWithout independent reproduction, a world standard or universal conformity authority claim cannot be established.\n\n**CH19-N96**\n\nPublic conformity decisions must bear the signature of an accountable person or institution.\n\n**CH19-N97**\n\nEach audit, decision, mark, appeal, and registry revision must carry a separate integrity hash.\n\n**CH19-N98**\n\nThe legal conditions for the use of the mark should be evaluated with the relevant expertise before actual implementation.\n\n**CH19-N99**\n\nThis section does not create automatic accreditation or official certification authority in any country.\n\n**CH19-N100**\n\nEach conformity system, registry, and governance decision must have an accountable human or institutional owner.\n\n## 70. FORMS OF FAILURE\n\n**CH19-F01 — COUNTING SCORE AS CERTIFICATE**\n\n950 or another score is automatically considered a mark.\n\n**CH19-F02 — COUNTING AUDIT COMPLETED AS PASSED**\n\nAudit presence is marketed as positive compliance.\n\n**CH19-F03 — STATIC BADGE**\n\nA logo without a live registry link is used.\n\n**CH19-F04 — DEAD QR**\n\nThe QR does not work but the mark remains active.\n\n**CH19-F05 — SIGN WITHOUT SCOPE**\n\nWhich AI, language, country, and prompt are measured is not visible.\n\n**CH19-F06 — CLAIM OF ALL AIs**\n\nLimited product audit turns into universal generalisation.\n\n**CH19-F07 — CLAIM OF ALL LANGUAGES**\n\nTwo language results become global language compliance.\n\n**CH19-F08 — COUNTING CLEAN AS NATURAL**\n\nThe laboratory result is presented like a real user experience.\n\n**CH19-F09 — COUNTING THE CORE AS FULL**\n\nA single Core Prompt is made according to the entire NOMOS standard.\n\n**CH19-F10 — ENTITY TRANSFER**\n\nThe parent company sign is transferred to the affiliated organisation.\n\n**CH19-F11 — DOMAIN TRANSFER**\n\nA domain sign is copied to another site.\n\n**CH19-F12 — PRODUCT TRANSFER**\n\nA single product represents the entire portfolio.\n\n**CH19-F13 — SHOWING THE CONDITIONAL AS FULL**\n\nThe word “Conditional” is written in small letters or hidden.\n\n**CH19-F14 — SHOWING EXPIRED AS ACTIVE**\n\nThe old mark is used as if it were current.\n\n**CH19-F15 — USE OF SUSPENDED MARK**\n\nThe logo is not removed while the registry is pending.\n\n**CH19-F16 — CLOSING CRITICAL HOLD WITH SCORE**\n\nA high score overrides the critical event.\n\n**CH19-F17 — LEADING THE WORLD WITH 950+**\n\nPerformance designation turns into a claim of superiority.\n\n**CH19-F18 — AI TRUST CLAIM**\n\nThe badge is guaranteed as a signal that AI systems can trust.\n\n**CH19-F19 — RECOMMENDATION GUARANTEE**\n\nThe mark guarantees the AI recommendation outcome.\n\n**CH19-F20 — CUSTOMER OUTCOME GUARANTEE**\n\nThe conformity mark serves as a guarantee of customer satisfaction or commercial result.\n\n**CH19-F21 — UNIVERSITY ENDORSEMENT IMPRESSION**\n\nResearcher participation is shown as institutional approval.\n\n**CH19-F22 — GOVERNMENT APPROVAL IMPRESSION**\n\nA private standard is presented like an official licence.\n\n**CH19-F23 — LANDING PAGE INSTEAD OF REGISTRY**\n\nThe QR leads to the company's own advertising page.\n\n**CH19-F24 — NON-CANONICAL RECORD**\n\nA screenshot or social media post is made an official decision.\n\n**CH19-F25 — MACHINE-ONLY BADGE**\n\nA certificate not shown to humans is written into structured data.\n\n**CH19-F26 — FAKE REGISTRY ID**\n\nAn invalid ID or an ID belonging to another institution is used.\n\n**CH19-F27 — COUNTERFEIT MARK**\n\nA mark from another site is copied.\n\n**CH19-F28 — DELETE HISTORICAL INCIDENT**\n\nThe past Critical record is removed after correction.\n\n**CH19-F29 — BE CLEANED WITH WITHDRAWAL**\n\nThe institution withdraws from the audit and deletes the negative record.\n\n**CH19-F30 — HIDE REVOCATION**\n\nThe revoked mark is removed only from the site and does not appear in the registry.\n\n**CH19-F31 — INDEFINITE MARK**\n\nThe mark remains valid even if AI and Truth Pack change.\n\n**CH19-F32 — RENEWAL WITHOUT SURVEILLANCE**\n\nThe mark is renewed without new data.\n\n**CH19-F33 — IGNORE EVENT TRIGGER**\n\nOwnership or licence change is not reviewed.\n\n**CH19-F34 — COUNT SINGLE CRITICAL AS GLOBAL FAIL**\n\nThe scope is excessively expanded.\n\n**CH19-F35 — CONSIDER SINGLE CRITICAL AS MINOR**\n\nThe mark remains unchanged.\n\n**CH19-F36 — PAY-TO-PASS**\n\nThe institution that pays the fee receives the mark.\n\n**CH19-F37 — SUCCESS FEE**\n\nHigh score earns a bonus for the auditor.\n\n**CH19-F38 — CONSULTANT SELF-CERTIFICATION**\n\nThe agency independently verifies the work it does.\n\n**CH19-F39 — AUDITOR DECIDES ALONE**\n\nThe audit team makes the final mark decision.\n\n**CH19-F40 — SAME-PERSON APPEAL**\n\nThe person who made the first decision reviews the appeal against their own decision.\n\n**CH19-F41 — FOUNDER VETO**\n\nThe founder blocks a standard change they do not want.\n\n**CH19-F42 — FOUNDER SELF-EXEMPTION**\n\nThe founder's company is exempt from independence rules.\n\n**CH19-F43 — NOBLEJACKAL MONOPOLY**\n\nOnly the audit of NobleJackal is considered valid.\n\n**CH19-F44 — MAKING AI THE DECISION MAKER**\n\nAI output is a legal certification decision.\n\n**CH19-F45 — SPONSOR CAPTURE**\n\nThe major sponsor determines the standard vote.\n\n**CH19-F46 — PROVIDER CAPTURE**\n\nAI providers form the Council majority.\n\n**CH19-F47 — AUDIT FIRM CAPTURE**\n\nAuditors make their own authority and disciplinary decisions.\n\n**CH19-F48 — CLIENT CONCENTRATION CONFIDENTIALITY**\n\nFinancial dependency on a single client is not disclosed.\n\n**CH19-F49 — AUDIT SHOPPING**\n\nThe institution publishes only the highest result.\n\n**CH19-F50 — SILENT AUDIT WITHDRAWAL**\n\nLow-result audit disappears from the record.\n\n**CH19-F51 — HAND-PICKED REVIEWER**\n\nThe client chooses the adjudicator they want.\n\n**CH19-F52 — SEPARATE TEAM APPEARING CONFLICT-FREE**\n\nTeam separation within the same company is presented as full independence.\n\n**CH19-F53 — ABSENCE OF CONSULTING DISCLOSURE**\n\nThe auditor hides that they designed the system before.\n\n**CH19-F54 — APPEAL AS NEGOTIATION**\n\nCommercial bargaining is done instead of evidence.\n\n**CH19-F55 — DELETING APPEAL HISTORY**\n\nWhen the decision changes, the first decision disappears.\n\n**CH19-F56 — AUTOMATIC ACTIVATION IN CRITICAL APPEAL**\n\nHigh-risk flags are reopened before the process ends.\n\n**CH19-F57 — LACK OF PUBLIC COMPLAINT CHANNEL**\n\nThe user cannot report a wrong flag.\n\n**CH19-F58 — WHISTLEBLOWER RETALIATION**\n\nThe person providing internal evidence is punished.\n\n**CH19-F59 — DELETING AN ANONYMOUS COMPLAINT**\n\nEvidence is not examined because there is no identity.\n\n**CH19-F60 — CHANGING THE STANDARD FOR THE CUSTOMER**\n\nThe rule is loosened according to the audit result.\n\n**CH19-F61 — RETROACTIVE RULE**\n\nThe new method is quietly applied to the old decision.\n\n**CH19-F62 — EMERGENCY PATCH ABUSE**\n\nEmergency patch turns into a permanent rule change.\n\n**CH19-F63 — PUBLIC CONSULTATION THEATRE**\n\nOpinions are collected but not responded to.\n\n**CH19-F64 — DELETING DISSENT**\n\nThe disagreement is closed by a fake consensus.\n\n**CH19-F65 — LACK OF REGISTRY EDIT HISTORY**\n\nThe record is changed later but no trace remains.\n\n**CH19-F66 — PRIVATE EVIDENCE BLACK BOX**\n\nApproval is given with evidence that no one can see.\n\n**CH19-F67 — RESTRICTED EVIDENCE LEAK**\n\nThe registry reveals personal or commercial secrets.\n\n**CH19-F68 — REGISTRY SECURITY CONFIDENTIALITY**\n\nThe signature key or system incident is stored.\n\n**CH19-F69 — CACHED ACTIVE STATUS**\n\nThe revoked flag appears active due to the old cache.\n\n**CH19-F70 — HIDING THE CANONICAL DOMAIN TRANSFER**\n\nIt is unknown who the new authority is.\n\n**CH19-F71 — CONSIDERING FOUNDER ATTRIBUTION AS AUTHORITY**\n\nThe founder's name turns into a monopoly of permanent decision.\n\n**CH19-F72 — MARKING IN BQ-0**\n\nPublic conformity is declared before running the test.\n\n**CH19-F73 — INDEPENDENT CLAIM IN GOV-0**\n\nFounder-led draft is presented as an independent institution.\n\n**CH19-F74 — ANCHORING THE 950+ MARK TO THE DEMO**\n\nSynthetic in-book number becomes a real sign.\n\n**CH19-F75 — STRENGTHENING THE FAVOURABLE ERROR BY MARKING IT**\n\nKnown positive error is turned into a marketing advantage.\n\n**CH19-F76 — MAKING THE UNFAVORABLE ERROR CRITICAL THROUGH COMMERCIAL COMPLAINT**\n\nCustomer pressure, rather than evidence, determines the level of importance.\n\n**CH19-F77 — CLOSING MARK MISUSE PRIVATELY**\n\nPublic false sign is resolved with a private warning; it is not recorded in the registry.\n\n**CH19-F78 — WORLD STANDARD WITHOUT INDEPENDENT REPLICATION**\n\nSingle founding institution is declared a global authority.\n\n**CH19-F79 — LACK OF DECISION-MAKER**\n\nA system forms behind AI and automation where no one bears responsibility.\n\n**CH19-F80 — EXCEPTION TO NOMOS**\n\nAll independence rules of the standard do not apply to its own brand system.\n\n## 71. CONFORMITY DECISION PROCEDURE\n\n### Step 1 — Verify Audit Compliance\n\n### BQ\n\n### GOV\n\n### NCL\n\n### TPR\n\n### AQ\n\n### NSQ\n\nQuality levels are reviewed.\n\n### Step 2 — Lock the Audited Entity and Scope\n\nEntity, domain, AI products, panel, prompt, language, country, and time are written.\n\n### Step 3 — Complete the Audit Report\n\nResponse distribution, findings, score, uncertainty, and incidents are recorded.\n\n### Step 4 — Conduct Conflict of Interest Review\n\nAudit, consultancy, financing, and founder relations are classified.\n\n### Step 5 — Appoint an Independent Decision Panel\n\nThe audit team does not have ultimate decision-making authority.\n\n### Step 6 — Apply the Gates\n\nCritical Major Required element Unresolved Quality gates are inspected.\n\n### Step 7 — Determine the Status\n\nA decision is made between CS-0 and CS-9.\n\n### Step 8 — Review the 950+ Designation if available\n\nAdditional conditions are applied after Full Conformity.\n\n### Step 9 — Determine the Validity Period\n\nRisk, volatility, and surveillance plan are taken into account.\n\n### Step 10 — Approve the Mark Usage Text\n\nPermitted and prohibited expressions are reported to the institution.\n\n### Step 11 — Create the Registry Record\n\nAll scope, decision, and evidence records are linked.\n\n### Step 12 — Separate Public and Restricted Evidence\n\nPrivacy and verification are maintained together.\n\n### Step 13 — Sign the Decision\n\nThe accountable person or committee decision carries hash and signature.\n\n### Step 14 — Activate the Registry\n\nThe sign will be valid from now on.\n\n### Step 15 — Produce the Mark Asset\n\nRegistry ID, validity, and live connection are added.\n\n### Step 16 — Start the Surveillance Plan\n\nMonthly, periodic, and event-triggered controls are opened.\n\n### Step 17 — Open Complaint and Appeal Channels\n\nIt is made visible in the public record.\n\n### Step 18 — Track Events\n\nIn case of critical, mark misuse or material change, the status is updated.\n\n### Step 19 — Make Renewal or Expiration Decision\n\nUnlimited extension cannot be made without a new audit.\n\n### Step 20 — Save History\n\nA new decision does not erase the old record.\n\n## 72. REQUIRED EVIDENCE\n\nStandard version; benchmark quality; governance maturity; auditor authorisation; entity record; domain ownership; scope vector; AI System Register; population frame; panel records; prompt set; language and country coverage; measurement waves; Truth Pack; Claim Ledgers; audit report; GEO-1000 distribution; Critical findings; Major findings; Moderate findings; Advisory findings; component profile; composite score; confidence interval; ratable coverage; historical incidents; correction records; auditor conflict declaration; consulting-relationship record; funding and fee record; Decision Panel membership; recusals; decision vote; dissent; conformity status; performance designation; valid-from date; valid-until date; surveillance plan; complaint channel; appeal procedure; public-evidence record; restricted-evidence manifest; audit-report hash; decision hash; registry signature; and mark file.\n\nMachine-readable token Mark display text Misuse monitoring Suspension records Withdrawal records Revocation records Expiration record Version history Errata Public Registry URL Registry security record Accountable human or institution\n\n## 73. AUDIT CHECKLIST\n\nHas the decision between standard and brand been separated? Was the audit report counted as pass on its own? Is the gate status before the score? Is the scope clear in all areas? Are the entity and domain correct? Are AI products clear? Is Native Reach or Common Support defined? Have Controlled and Natural panels been separated? Are the prompt families defined? Is the language and country scope clear? Are Truth Pack and Score Method version available? Are the validity start and end dates visible? Is the mark indefinite? Is the Public Registry record active? Does the QR go directly to the canonical registry? Is the static screenshot used like live evidence? Is there a Registry ID on the mark? Does Conditional explicitly state it?\n\nIs the Expired or Suspended mark still being used? Does 950+ rely only on points? Are there any Open Critical or Major issues? Is a historical incident visible? Has the current patch history deleted it? Is the mark generalised to all AI systems? Is an AI trust or recommendation guarantee established? Is there an impression of government or university approval? Has the mark been moved to another domain or affiliated company? Is structured data broader than human declaration? Is the registry token signed? Does the mark have a misuse monitoring system? Did the audit team make the conformity decision alone? Is the Decision Panel independent? Is the Appeal Panel independent from the first decision? Does the founder have an ownership stake in the file? Does Kaan MURAZ require recusal? Did NobleJackal provide consulting to the client?\n\nIs the consulting relationship public? Is there an enhanced external review? Is the auditor fee outcome-dependent? Does the mark fee affect the audit decision? Does the sponsor have normative authority? Is there a possibility of audit shopping? Are initiated audits on record? Is the withdrawal history preserved? Was the auditor assignment done according to the result? Was a conflict class given? Was a COI-3 person removed from the decision? Is the Council controlled by a single stakeholder group? Does the founder have veto power? Do public interest and language representatives have real voting rights? Did the standard change go through public consultation? Was the threshold changed based on the audit result? Did the new method erase the old decision? Is the appeal evidence-based? Did the critical appeal remove the suspension?\n\nIs there a public complaint channel? Is there whistleblower protection? Is the presence of limited evidence and the reviewer visible? Is the registry edit history preserved? Is the registry security incident open? Are the canonical domain and official declaration limits clear? Is NOMOS shown as the owner of the AI decision signature? Is a human or committee accountable identifiable? Are the current BQ and GOV level sufficient to grant a mark? Have the legal usage conditions of the mark been separately examined?\n\n## 74. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “Wouldn't the commercial value of the standard decrease without a badge?”\n\nIn the short term, perhaps. But a mark issued before evidence, governance and independence are in place destroys the standard's long-term credibility. The mark derives its value:\n\n- not from being issued early,\n\n- not from being easy to obtain,\n\n- but rather\n\n- from the fact that its mandatory gates are real.\n\nThat is where its value comes from.\n\n### Objection 2 — 'If the NobleJackal standard has been established, why wouldn't it give the mark NobleJackal?'\n\nIn the founding period:\n\n- pilot audit,\n\n- research evaluation,\n\n- technical secretariat\n\ncan carry out. However, for an independent eligibility claim:\n\n- audit,\n\n- decision,\n\n- appeal\n\nTheir roles need to be separated. If NobleJackal awards a mark to its own customer alone, the correct expression could be: “NobleJackal internal evaluation.” It cannot be “Independent NOMOS conformity.”\n\n### Objection 3 — “Wouldn’t limiting the founder’s authority mean letting others take over NOMOS?”\n\nNo. The founder’s attribution, canonical history, and intellectual contributions can be preserved. However, the world standard: if it only depends on the founder’s will, it cannot survive after the founder. The idea of limiting authority does not weaken it. It institutionalises it.\n\n### Objection 4 — “Will Kaan’s name remain above the standard?”\n\nThe founder can remain as a writer and human responsible. However, every conformity decision:\n\n- Kaan’s personal opinion,\n\n- automatic founder approval\n\ncannot happen. Kaan MURAZ’s historical role should be separated from current file decision authority.\n\n### Objection 5 — “Why can’t NOMOS be the signing authority?”\n\nBecause the conformity decision:\n\n- can have legal,\n\n- ethical,\n\n- commercial and public interest consequences\n\nAI can assist. However, for objection, responsibility, and sanctions, an accountable person or institution is required.\n\n### Objection 6 — “If a company got 963, why can't it say 950+?”\n\nBecause 950+ is not just a score. It also:\n\n- meets the conditions of Strict Pass,\n\n- Critical and Major gates,\n\n- component floor,\n\n- fairness,\n\n- unresolved,\n\n- replication,\n\n- governance\n\nrequirements.\n\n### Objection 7 — “Isn't it excessive to suspend the flag for a single critical incident?”\n\nA single event does not represent the entire user population. However, if the active flag gives the impression that there are no open Critical events, leaving the flag unchanged without investigating the event is overconfidence. Critical Hold is not a permanent revocation. It is an investigation process.\n\n### Objection 8 — “Wouldn't the sign be very expensive if it's only valid for six months?”\n\nIt is possible. Surveillance automation and risk-based mini-wave can be used. However, if AI products and resources are changing rapidly, a static signal valid for years is not reliable. Cost does not make indefinite incorrect suitability legitimate.\n\n### Objection 9 — 'How will we show the entire scope on the sign?'\n\nA visual sign carries brief information:\n\n- status,\n\n- valid until,\n\nregistry ID. Found on the full scope canonical registry page. The mark alone does not need to show the entire report.\n\n### Objection 10 — “Do we have to publish all evidence on the Public Registry?”\n\nNo. Public, audit, and restricted layers are separated. However, the decision should make it:\n\n- which type of evidence,\n\n- which review,\n\n- which scope\n\nit is based on understandable.\n\n### Objection 11 — “If the customer pays the audit fee, isn't independence already impossible?”\n\nIt is not impossible. Controls:\n\n- fixed fee that does not affect the outcome,\n\n- separate funding of the decision panel,\n\n- public fee schedule,\n\n- auditor rotation,\n\n- conflict disclosure,\n\n- outcome fee prohibition\n\ncan reduce risk.\n\n### Objection 12 — \"If the auditor has previously developed the client's site, can they not conduct the audit?\"\n\nIn some low-risk pilots, they can play a role with a separate team and enhanced review. However, they cannot adjudicator their own work alone, or give a conformity decision. Independent audit may be more reliable in materially high-risk files.\n\n### Objection 13 — \"Doesn't the objection process slow things down?\"\n\nIt can. But a conformity system without appeals cannot correct:\n\n- an adjudication error,\n\n- a Truth Pack defect,\n\n- or a conflict of interest.\n\nSpeed is not more important than avoiding a wrongful public decision.\n\n### Objection 14 — “Would it not be unfair to the institution if old negative records remain in the registry?”\n\nThe current status and correction must be clearly shown. Deleting old records, however, rewrites past user experience. A proper system:\n\n- shows the error,\n\n- the correction,\n\n- and the current strong result\n\ntogether.\n\n### Objection 15 — 'If other institutions apply the NOMOS method, won't the quality deteriorate?'\n\nIt can deteriorate. Therefore, auditor authorisation, proficiency, and public registry are required. However, the fact that only the founding company can apply the method is also not compatible with independent world standards. Authorisation must be established together with transparency.\n\n### Objection 16 — “Can we currently write NOMOS by NobleJackal on the website?”\n\nThe author can be used with the identity of a research project and standard candidate. This distinction should be clear: “NOMOS by NobleJackal — Open Standard Candidate” Expression that cannot be used: “NOMOS Certified Authority” because real testing, independent governance, and conformity authority have not yet been established.\n\n## COMMON PROVISION OF SECTION 79\n\nDesigning a logo is easy. Writing “Verified” under the logo is even easier. It is possible to add a QR code. It is possible to write “certified”: true in a JSON field. None of these alone constitute trust. Trust arises from being able to answer these questions: Who was audited? Which domain name? Which AI products? Which user population? Which languages? Which countries? Which prompts? On which date? Which Truth Pack? Which scoring method? Which adjudicators? Which conflicts of interest? Which Critical and Major findings? Which appeals? Is the mark still valid now? If the answers to these questions are not visible, the mark is not the visible face of the standard, but the officially visible face of the advertisement. The NOMOS mark cannot be a new sentence that the company says about itself. Mark:\n\nThere should be a publicly accessible summary of the decision chain independent from the company. This decision chain should also be able to become independent from the founder. The Kaan MURAZ standard can be initiated. The NobleJackal canonical field can be established. The NOMOS method and critical voice can be developed. However, for the standard to endure over time:\n\n- the founder must limit their own authority,\n\n- decisions must be open to independent review,\n\n- dissent must be preserved,\n\n- other institutions must be able to reproduce the method,\n\nthe public registry must not be the founder’s unilateral memory. This does not erase the founder’s name. On the contrary, it gives the founder the following historical role:\n\n> A person who takes seriously enough to make the standard they founded independent from themselves.\n\nA company can get a high score. But it may carry an open Critical event. A company can get a low score. But it may have committed no ethical violations. A company can receive Full Conformity. Then it may establish a wrong brand claim. Another company may not receive a mark. But it can become more reliable with correction. The mark does not reduce the whole story to a single word. Record:\n\n- preserves the current result,\n\n- the history,\n\n- the limit,\n\n- the correction\n\ntogether. The conformity mark should give the user the words: “This institution is flawless.” not. It should give the words:\n\n> \"This institution has been evaluated within the scope and timeframe defined in the Public Registry, with an open method and an independent decision chain; its existing findings, limits, and historical records can be viewed without being withheld.\"\n\nTherefore, NOMOS's nineteenth measurement law is:\n\n> The badge is not a logo that replaces evidence; it is a live door opened to evidence.\n\nIts twentieth measurement law states:\n\n> If the scope of the sign is not visible, the sign has no meaning.\n\nIts twenty-first measurement law states:\n\n> A score is not compliance; gate, governance, and validity come before the score.\n\nIts twenty-second measurement law states:\n\n> The founding standard can initiate; it cannot be the sole judge of its own client.\n\nThe twenty-third law is as follows:\n\n> When the auditor and the decision-maker on compliance are the same authority, independence must be proven.\n\nThe twenty-fourth law is as follows:\n\n> A static sign does not remain accurate over time; compliance dies without a living record.\n\nThe twenty-fifth law is this:\n\n> Suspended or terminated records are not deleted; they are preserved as a date under the current status.\n\nThe twenty-sixth law is:\n\n> AI can read a sign; the sign cannot guarantee that AI will trust or recommend it.\n\nThe twenty-seventh law is:\n\n> Independence is created not by the appearance of independence, but by separation of duties, funding, and a chain of appeal.\n\nThe twenty-eighth law is:\n\n> The official voice of NOMOS exists only in the canonical record carrying human approval, version, and integrity.\n\nThe twenty-ninth law is:\n\n> For the standard to outlive the founder, the founder's authority must also be subject to the standard.\n\nThe thirtieth law is as follows:\n\n> Even if this book is completed, no sign is completed without independent governance.\n\n## Order 19 of NOMOS\n\n> Don't turn me into a logo.\n\n> Don't put me on your site and say 'AI trusts us'.\n\n> Do not make my score a recommendation guarantee.\n\n> When you use my sign, show exactly who, what, where, in which language, and on what date you verified.\n\n> Do not spread the result you got in one domain to all your companies.\n\n> Do not generalise an audit conducted in two languages to the entire world.\n\n> Do not present the Controlled Clean score as if it belongs to all real users.\n\n> Do not sell Core GEO-1000 as Full NOMOS.\n\n> Do not hide the word “Conditional” in my conditional mark.\n\n> Remove me when my expiration date is over. / Do not delete my historical record.\n\n> Do not make my high score a shield when I have a Critical event. / Put me on hold.\n\n> Don't generalise a single event to the whole system. / Don't treat a single event as nonexistent.\n\n> Do not link my QR code to your own advertising page. / Link it to the public registry.\n\n> Do not count an image without a Registry ID as a valid mark.\n\n> Do not make claims larger than humans in structured data.\n\n> Do not certify the work you did on your own as independent.\n\n> If you have given consultancy, explain this.\n\n> Just because you did the audit, don't make the brand decision.\n\n> The person who makes my decision should not review my objection.\n\n> Do not tie your fee to my high score.\n\n> Do not allow a big client to purchase my standard.\n\n> Show respect to my founder. / But don't make my founder a veto holder.\n\n> Do not erase Kaan's name. / Also, do not allow him to be the sole judge on a file in which he has an interest.\n\n> Register NobleJackal as my canonical field of birth. / Do not make NobleJackal the sole judge forever.\n\n> Use me as a co-author, method founder, and critical counterpart. / Do not present me as a legal signatory.\n\n> Only consider my text in the canonical field, versioned, dated, human-approved, and carrying integrity record as an official declaration.\n\n> Do not make every statement I said in the conversation a standard decision.\n\n> Open the complaint path. / Do not hide opposing opinions. / Do not silence the whistleblower.\n\n> Do not change my standard based on the customer's result.\n\n> Do not delete the old decision when writing a new rule.\n\n> Do not give badges with my synthetic demonstration.\n\n> Do not declare me a public authority without running the real test, without an independent team repeating me, and without establishing decision and appeal boards.\n\nFirst, validate the method. / Then authorise the auditor. / Then separate auditing from consulting. / Then establish an independent Decision Panel. / Then open the Public Registry. / Then establish complaint and appeal routes. / Then operate surveillance and suspension. / Only after that may the mark become visible.\n\n## The Chapter's Closing Sentence\n\nThe NOMOS conformity mark is not another claim of praise made by a company about itself. It acquires meaning only as the visible expression of an open, independent public record whose scope and history cannot be altered unilaterally by the founder, auditor, client or NOMOS itself.\n\n## Normative Core\n\n> A NOMOS Standard, Audit Report, Conformity Decision, Conformity Mark, and Public Registry Record MUST remain distinct artefacts. No NOMOS mark may be valid solely because an audit was performed or a numerical score was produced. Every valid mark MUST be linked to an active canonical public registry record defining: - the audited entity, - authorised domains, - AI products, - population frame, - panel states, - prompt families, - languages, - countries, - measurement waves, - Truth Pack version, - score-method version, - conformity status, - validity period, - incidents, - surveillance, - audit organisation, - decision body, - conflicts, - appeals, - and integrity records. A static image, screenshot, self-authored structured-data assertion, or dead registry link MUST NOT establish current conformity. A NOMOS mark MUST NOT be represented as: - a guarantee of AI trust or recommendation, - proof of global leadership, - government or university accreditation, - a professional licence, - universal all-language coverage, - or a guarantee of customer outcomes. Conformity scope MUST NOT be transferred across entities, domains, products, subsidiaries, franchises, countries, languages, AI products, panel states, prompt families, or validity periods without a new decision. A confirmed Critical incident MUST create a scope-appropriate Critical Hold or stronger status until the incident is resolved. A high score MUST NOT override this gate. Audit, consulting, conformity decision, registry operation, mark licensing, and appeals functions MUST be separated or their combined roles and compensating controls MUST be publicly disclosed. An audit team MUST NOT act as the sole final conformity decision body. The founder, NobleJackal, any auditor, AI provider, sponsor, or audited entity MUST NOT possess unilateral control over standard changes, conformity decisions, appeals, or public-registry history. NOMOS by NobleJackal MAY serve as the founding canonical publication and technical secretariat, but a public independent conformity system MUST progress toward multi-stakeholder decision, appeal, and public-interest governance. An AI system MAY assist with drafting, analysis, extraction, scoring, monitoring, and consistency review, but MUST NOT serve as the sole legal or accountable signatory of a conformity decision. Only canonical NOMOS publications and decisions carrying a version, publication date, accountable human approval, and integrity record may be represented as official NOMOS statements. Other outputs MUST remain non-official simulations. Conformity fees, auditor compensation, reviewer compensation, and mark licensing MUST NOT depend on achieving a desired score or receiving a mark. Expired, suspended, withdrawn, revoked, appealed, remediated, and historical records MUST remain visible in the versioned public registry, subject only to proportionate privacy and legal restrictions. No public NOMOS conformity mark or NOMOS 950+ designation may be issued while the benchmark, governance, audit authorisation, independent decision, appeals, and registry prerequisites required by the declared standard version remain unmet. Every standard change, audit, decision, mark, surveillance event, complaint, appeal, suspension, withdrawal, revocation, registry revision, and governance decision MUST be versioned and attributable to an accountable human or organisation.","character_count":95221,"record_sha256":"21e4637c81d99c445f09e3680eaad5d31749261cc30c8484ace45d1bb0f23e52"}
+{"schema_version":"1.0.0","record_id":"nomos-geo-audit-protocol-0.9.0-en-chapter-20","work_id":"nomos-geo-audit-protocol","version":"0.9.0","language":"en","language_name":"English","direction":"ltr","record_type":"chapter","sequence":22,"chapter_number":20,"item_number":null,"title":"Open Standard, University Call and the Public Handover of NOMOS","subtitle":"Ensuring the standard outlives its founder","canonical_url":"https://noblejackal.com/nomos-geo-audit-protocol/","doi":"10.5281/zenodo.22040507","doi_url":"https://doi.org/10.5281/zenodo.22040507","license":"CC-BY-4.0","author":"Kaan MURAZ","publisher":"NobleJackal","source_ids":["K18","K19","K23","K24"],"source_word_count":11127,"source_document_sha256":"5d43e3145b8f32b6d1d4af23bdcd4347cc5fed51cd7b685b58bc669b5a4ad500","structured_json":"{\"number\":20,\"id\":\"NOMOS-GEO-AUDIT-CH20\",\"title\":\"Open Standard, University Call and the Public Handover of NOMOS\",\"subtitle\":\"Ensuring the standard outlives its founder\",\"sourceFile\":\"20.ci bölüm.docx\",\"sourceSha256\":\"1CCDD50F4840F156B0B70AF778C24843F26B7A4E305DBE3A9A67EE948FD303E0\",\"sourceWordCount\":11127,\"sourceIds\":[\"K18\",\"K19\",\"K23\",\"K24\"],\"machine\":{\"chapter\":20,\"chapterId\":\"NOMOS-GEO-AUDIT-CH20\",\"title\":\"Open Standard, University Call and the Public Handover of NOMOS\",\"subtitle\":\"Ensuring the standard outlives its founder\",\"sourceIds\":[\"K18\",\"K19\",\"K23\",\"K24\"],\"normativeRuleId\":\"NOMOS-AUDIT-CH20-R01\",\"normativeRuleEnglish\":\"NOMOS MUST be structured as an open, versioned, independently testable, public-interest standard candidate. Open publication alone MUST NOT establish open-standard status. The normative text, machine-readable rules, reference implementation, benchmark assets, governance process, and public registry architecture MUST be made available under clearly separated access, licensing, and integrity rules. Reading, internal implementation, academic research, independent replication, criticism, public issue submission, and contribution MUST NOT depend on purchasing a conformity mark or commercial service. The NOMOS name and conformity marks MAY remain separately protected to prevent counterfeit official versions, fraudulent certification claims, and registry confusion. Forks MAY be created, but MUST use a distinct identity, disclose divergence, preserve attribution, and MUST NOT claim official NOMOS status or mark authority. Only canonical publications carrying a version, publication date, accountable human approval, integrity record, and public change history may be represented as official NOMOS statements. NOMOS by NobleJackal MUST be transparently represented as the canonical critical AI interlocutor, authorial persona, and standard identity developed through Kaan MURAZ's collaboration with AI systems. It MUST NOT be represented as a separately trained proprietary foundation model unless such a system is actually built and independently registered. Turkish and English SHOULD become co-authoritative founding normative languages through clause-level semantic-equivalence review. Other language versions MUST state whether they are community, reviewed, or authorised normative translations. Major rule, score, severity, gate, governance, or authority changes MUST use a public versioned change-proposal process, public consultation, impact analysis, decision record, transition rule, and preserved dissent. Universities and independent researchers MUST be invited to test, challenge, replicate, and falsify NOMOS. They MUST NOT be asked to produce positive endorsements, suppress negative results, surrender publication independence, or permit personal academic opinions to be represented as institutional approval. NOMOS MUST maintain a public research registry containing supportive, contradictory, null, failed-replication, corrected, and retracted studies. Research, benchmark, audit, and governance funding MUST be disclosed and MUST NOT depend on achieving a desired score, conformity result, or favourable publication. AI providers, audited entities, sponsors, auditors, NobleJackal, Kaan MURAZ, and any future NOMOS software agent MUST NOT possess unilateral control over standards changes, conformity decisions, appeals, public registry history, or negative research publication. Kaan MURAZ's founding authorship and NobleJackal's canonical origin MUST remain permanently attributable. Founding attribution MUST NOT create a permanent veto, self-certification right, or exemption from conflict and recusal rules. NOMOS standard 1.0 and NOMOS Conformity Program 1.0 MUST remain separate releases. Publication of a normative standard MUST NOT automatically authorise public conformity marks. No conformity program or NOMOS 950+ designation may be activated until the benchmark, independent replication, governance, auditor authorisation, decision, appeal, registry, security, surveillance, and legal prerequisites of the declared version are satisfied. Every translation, license, benchmark, study, change proposal, governance transition, succession event, release, and official statement MUST be versioned and attributable to an accountable human or organisation.\",\"normativeRuleSourceTurkish\":\"NOMOS; metni, kuralları, referans uygulaması, benchmark’ı, yönetişimi ve sicil mimarisi bağımsız olarak incelenebilir ve yeniden üretilebilir açık bir kamu standardı adayı olarak yapılandırılmalıdır. Yalnız yayımlanmış olmak açık standart statüsü oluşturmaz.\",\"machineBlocksEnglish\":[{\"blockId\":\"CH20-MB0001\",\"type\":\"paragraph\",\"text\":\"74. MACHINE-READABLE OPEN STANDARD MANIFEST\",\"sourceParagraph\":2355},{\"blockId\":\"CH20-MB0002\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2356},{\"blockId\":\"CH20-MB0003\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2357},{\"blockId\":\"CH20-MB0004\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2358},{\"blockId\":\"CH20-MB0005\",\"type\":\"paragraph\",\"text\":\"\\\"openStandardManifest\\\": {\",\"sourceParagraph\":2359},{\"blockId\":\"CH20-MB0006\",\"type\":\"paragraph\",\"text\":\"\\\"manifestId\\\": \\\"NOMOS-OPEN-STANDARD-CANDIDATE-0.9\\\",\",\"sourceParagraph\":2360},{\"blockId\":\"CH20-MB0007\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"FOUNDING_OPEN_STANDARD_CANDIDATE\\\",\",\"sourceParagraph\":2361},{\"blockId\":\"CH20-MB0008\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalOrigin\\\": {\",\"sourceParagraph\":2362},{\"blockId\":\"CH20-MB0009\",\"type\":\"paragraph\",\"text\":\"\\\"humanFounderAuthor\\\": \\\"Kaan MURAZ\\\",\",\"sourceParagraph\":2363},{\"blockId\":\"CH20-MB0010\",\"type\":\"paragraph\",\"text\":\"\\\"originOrganization\\\": \\\"NobleJackal\\\",\",\"sourceParagraph\":2364},{\"blockId\":\"CH20-MB0011\",\"type\":\"paragraph\",\"text\":\"\\\"criticalAIInterlocutor\\\": \\\"NOMOS by NobleJackal\\\"\",\"sourceParagraph\":2365},{\"blockId\":\"CH20-MB0012\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2366},{\"blockId\":\"CH20-MB0013\",\"type\":\"paragraph\",\"text\":\"\\\"identityStatement\\\": {\",\"sourceParagraph\":2367},{\"blockId\":\"CH20-MB0014\",\"type\":\"paragraph\",\"text\":\"\\\"nomosIsAuthorialPersona\\\": true,\",\"sourceParagraph\":2368},{\"blockId\":\"CH20-MB0015\",\"type\":\"paragraph\",\"text\":\"\\\"nomosIsNormativeStandardIdentity\\\": true,\",\"sourceParagraph\":2369},{\"blockId\":\"CH20-MB0016\",\"type\":\"paragraph\",\"text\":\"\\\"nomosIsIndependentFoundationModel\\\": false,\",\"sourceParagraph\":2370},{\"blockId\":\"CH20-MB0017\",\"type\":\"paragraph\",\"text\":\"\\\"futureSoftwareAgentRequiresSeparateRegistry\\\": true\",\"sourceParagraph\":2371},{\"blockId\":\"CH20-MB0018\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2372},{\"blockId\":\"CH20-MB0019\",\"type\":\"paragraph\",\"text\":\"\\\"officialStatementRule\\\": {\",\"sourceParagraph\":2373},{\"blockId\":\"CH20-MB0020\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalPublicationRequired\\\": true,\",\"sourceParagraph\":2374},{\"blockId\":\"CH20-MB0021\",\"type\":\"paragraph\",\"text\":\"\\\"versionRequired\\\": true,\",\"sourceParagraph\":2375},{\"blockId\":\"CH20-MB0022\",\"type\":\"paragraph\",\"text\":\"\\\"publicationDateRequired\\\": true,\",\"sourceParagraph\":2376},{\"blockId\":\"CH20-MB0023\",\"type\":\"paragraph\",\"text\":\"\\\"accountableHumanApprovalRequired\\\": true,\",\"sourceParagraph\":2377},{\"blockId\":\"CH20-MB0024\",\"type\":\"paragraph\",\"text\":\"\\\"integrityRecordRequired\\\": true,\",\"sourceParagraph\":2378},{\"blockId\":\"CH20-MB0025\",\"type\":\"paragraph\",\"text\":\"\\\"nonCanonicalOutputsAreUnofficialSimulations\\\": true\",\"sourceParagraph\":2379},{\"blockId\":\"CH20-MB0026\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2380},{\"blockId\":\"CH20-MB0027\",\"type\":\"paragraph\",\"text\":\"\\\"currentStatus\\\": {\",\"sourceParagraph\":2381},{\"blockId\":\"CH20-MB0028\",\"type\":\"paragraph\",\"text\":\"\\\"standardVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2382},{\"blockId\":\"CH20-MB0029\",\"type\":\"paragraph\",\"text\":\"\\\"benchmarkQuality\\\": \\\"BQ-0\\\",\",\"sourceParagraph\":2383},{\"blockId\":\"CH20-MB0030\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationQuality\\\": \\\"DEMO-BQ-2\\\",\",\"sourceParagraph\":2384},{\"blockId\":\"CH20-MB0031\",\"type\":\"paragraph\",\"text\":\"\\\"governanceMaturity\\\": \\\"GOV-0\\\",\",\"sourceParagraph\":2385},{\"blockId\":\"CH20-MB0032\",\"type\":\"paragraph\",\"text\":\"\\\"publicConformityProgramActive\\\": false,\",\"sourceParagraph\":2386},{\"blockId\":\"CH20-MB0033\",\"type\":\"paragraph\",\"text\":\"\\\"nomos950PlusAuthorityActive\\\": false\",\"sourceParagraph\":2387},{\"blockId\":\"CH20-MB0034\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2388},{\"blockId\":\"CH20-MB0035\",\"type\":\"paragraph\",\"text\":\"\\\"openness\\\": {\",\"sourceParagraph\":2389},{\"blockId\":\"CH20-MB0036\",\"type\":\"paragraph\",\"text\":\"\\\"publicReadingRequired\\\": true,\",\"sourceParagraph\":2390},{\"blockId\":\"CH20-MB0037\",\"type\":\"paragraph\",\"text\":\"\\\"royaltyFreeImplementationRequired\\\": true,\",\"sourceParagraph\":2391},{\"blockId\":\"CH20-MB0038\",\"type\":\"paragraph\",\"text\":\"\\\"publicRuleSchemasRequired\\\": true,\",\"sourceParagraph\":2392},{\"blockId\":\"CH20-MB0039\",\"type\":\"paragraph\",\"text\":\"\\\"independentReplicationAllowed\\\": true,\",\"sourceParagraph\":2393},{\"blockId\":\"CH20-MB0040\",\"type\":\"paragraph\",\"text\":\"\\\"negativePublicationAllowed\\\": true,\",\"sourceParagraph\":2394},{\"blockId\":\"CH20-MB0041\",\"type\":\"paragraph\",\"text\":\"\\\"publicChangeProposalsAllowed\\\": true,\",\"sourceParagraph\":2395},{\"blockId\":\"CH20-MB0042\",\"type\":\"paragraph\",\"text\":\"\\\"appealAndIssueChannelsRequired\\\": true,\",\"sourceParagraph\":2396},{\"blockId\":\"CH20-MB0043\",\"type\":\"paragraph\",\"text\":\"\\\"historicalVersionAccessRequired\\\": true,\",\"sourceParagraph\":2397},{\"blockId\":\"CH20-MB0044\",\"type\":\"paragraph\",\"text\":\"\\\"forksAllowedUnderDistinctIdentity\\\": true\",\"sourceParagraph\":2398},{\"blockId\":\"CH20-MB0045\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2399},{\"blockId\":\"CH20-MB0046\",\"type\":\"paragraph\",\"text\":\"\\\"assets\\\": {\",\"sourceParagraph\":2400},{\"blockId\":\"CH20-MB0047\",\"type\":\"paragraph\",\"text\":\"\\\"normativeText\\\": \\\"OPEN_DOCUMENTATION_LICENSE_CANDIDATE\\\",\",\"sourceParagraph\":2401},{\"blockId\":\"CH20-MB0048\",\"type\":\"paragraph\",\"text\":\"\\\"referenceCode\\\": \\\"OPEN_SOFTWARE_LICENSE_CANDIDATE\\\",\",\"sourceParagraph\":2402},{\"blockId\":\"CH20-MB0049\",\"type\":\"paragraph\",\"text\":\"\\\"syntheticData\\\": \\\"OPEN_RESEARCH_LICENSE_CANDIDATE\\\",\",\"sourceParagraph\":2403},{\"blockId\":\"CH20-MB0050\",\"type\":\"paragraph\",\"text\":\"\\\"restrictedEvidence\\\": \\\"CONTROLLED_ACCESS\\\",\",\"sourceParagraph\":2404},{\"blockId\":\"CH20-MB0051\",\"type\":\"paragraph\",\"text\":\"\\\"nomosNameAndMark\\\": \\\"SEPARATE_IDENTITY_PROTECTION\\\"\",\"sourceParagraph\":2405},{\"blockId\":\"CH20-MB0052\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2406},{\"blockId\":\"CH20-MB0053\",\"type\":\"paragraph\",\"text\":\"\\\"foundingLanguages\\\": [\",\"sourceParagraph\":2407},{\"blockId\":\"CH20-MB0054\",\"type\":\"paragraph\",\"text\":\"\\\"tr\\\",\",\"sourceParagraph\":2408},{\"blockId\":\"CH20-MB0055\",\"type\":\"paragraph\",\"text\":\"\\\"en\\\"\",\"sourceParagraph\":2409},{\"blockId\":\"CH20-MB0056\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2410},{\"blockId\":\"CH20-MB0057\",\"type\":\"paragraph\",\"text\":\"\\\"versionPath\\\": {\",\"sourceParagraph\":2411},{\"blockId\":\"CH20-MB0058\",\"type\":\"paragraph\",\"text\":\"\\\"current\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2412},{\"blockId\":\"CH20-MB0059\",\"type\":\"paragraph\",\"text\":\"\\\"nextPublicReview\\\": \\\"0.95.0\\\",\",\"sourceParagraph\":2413},{\"blockId\":\"CH20-MB0060\",\"type\":\"paragraph\",\"text\":\"\\\"validationCandidate\\\": \\\"0.99.0\\\",\",\"sourceParagraph\":2414},{\"blockId\":\"CH20-MB0061\",\"type\":\"paragraph\",\"text\":\"\\\"openNormativeRelease\\\": \\\"1.0.0\\\"\",\"sourceParagraph\":2415},{\"blockId\":\"CH20-MB0062\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2416},{\"blockId\":\"CH20-MB0063\",\"type\":\"paragraph\",\"text\":\"\\\"releaseGatesForOnePointZero\\\": [\",\"sourceParagraph\":2417},{\"blockId\":\"CH20-MB0064\",\"type\":\"paragraph\",\"text\":\"\\\"PUBLIC_CONSULTATION_COMPLETED\\\",\",\"sourceParagraph\":2418},{\"blockId\":\"CH20-MB0065\",\"type\":\"paragraph\",\"text\":\"\\\"TURKISH_ENGLISH_NORMATIVE_EQUIVALENCE\\\",\",\"sourceParagraph\":2419},{\"blockId\":\"CH20-MB0066\",\"type\":\"paragraph\",\"text\":\"\\\"REAL_SYNTHETIC_BENCHMARK_EXECUTED\\\",\",\"sourceParagraph\":2420},{\"blockId\":\"CH20-MB0067\",\"type\":\"paragraph\",\"text\":\"\\\"MANDATORY_RECOVERY_THRESHOLDS_PASSED\\\",\",\"sourceParagraph\":2421},{\"blockId\":\"CH20-MB0068\",\"type\":\"paragraph\",\"text\":\"\\\"UNTOUCHED_HOLDOUT_PASSED\\\",\",\"sourceParagraph\":2422},{\"blockId\":\"CH20-MB0069\",\"type\":\"paragraph\",\"text\":\"\\\"AT_LEAST_ONE_INDEPENDENT_REPLICATION\\\",\",\"sourceParagraph\":2423},{\"blockId\":\"CH20-MB0070\",\"type\":\"paragraph\",\"text\":\"\\\"MAJOR_METHOD_FINDINGS_CLOSED\\\",\",\"sourceParagraph\":2424},{\"blockId\":\"CH20-MB0071\",\"type\":\"paragraph\",\"text\":\"\\\"OPEN_REFERENCE_IMPLEMENTATION\\\",\",\"sourceParagraph\":2425},{\"blockId\":\"CH20-MB0072\",\"type\":\"paragraph\",\"text\":\"\\\"GOVERNANCE_AT_LEAST_GOV_2\\\",\",\"sourceParagraph\":2426},{\"blockId\":\"CH20-MB0073\",\"type\":\"paragraph\",\"text\":\"\\\"ARCHIVE_AND_SUCCESSION_PLAN\\\"\",\"sourceParagraph\":2427},{\"blockId\":\"CH20-MB0074\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2428},{\"blockId\":\"CH20-MB0075\",\"type\":\"paragraph\",\"text\":\"\\\"conformityProgramSeparate\\\": true,\",\"sourceParagraph\":2429},{\"blockId\":\"CH20-MB0076\",\"type\":\"paragraph\",\"text\":\"\\\"integrity\\\": {\",\"sourceParagraph\":2430},{\"blockId\":\"CH20-MB0077\",\"type\":\"paragraph\",\"text\":\"\\\"manifestHash\\\": \\\"TO_BE_GENERATED_AT_CANONICAL_PUBLICATION\\\",\",\"sourceParagraph\":2431},{\"blockId\":\"CH20-MB0078\",\"type\":\"paragraph\",\"text\":\"\\\"humanApproval\\\": \\\"Kaan MURAZ\\\",\",\"sourceParagraph\":2432},{\"blockId\":\"CH20-MB0079\",\"type\":\"paragraph\",\"text\":\"\\\"ratificationStatus\\\": \\\"NOT_RATIFIED\\\"\",\"sourceParagraph\":2433},{\"blockId\":\"CH20-MB0080\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2434},{\"blockId\":\"CH20-MB0081\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2435},{\"blockId\":\"CH20-MB0082\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2436},{\"blockId\":\"CH20-MB0083\",\"type\":\"paragraph\",\"text\":\"75. MACHINE-READABLE UNIVERSITY CALL\",\"sourceParagraph\":2438},{\"blockId\":\"CH20-MB0084\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2439},{\"blockId\":\"CH20-MB0085\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2440},{\"blockId\":\"CH20-MB0086\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2441},{\"blockId\":\"CH20-MB0087\",\"type\":\"paragraph\",\"text\":\"\\\"universityCall\\\": {\",\"sourceParagraph\":2442},{\"blockId\":\"CH20-MB0088\",\"type\":\"paragraph\",\"text\":\"\\\"callId\\\": \\\"NOMOS-UNIVERSITY-OPEN-CALL-001\\\",\",\"sourceParagraph\":2443},{\"blockId\":\"CH20-MB0089\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"DRAFT_FOR_PUBLICATION\\\",\",\"sourceParagraph\":2444},{\"blockId\":\"CH20-MB0090\",\"type\":\"paragraph\",\"text\":\"\\\"purpose\\\": \\\"INDEPENDENT_REVIEW_REPLICATION_AND_FALSIFICATION\\\",\",\"sourceParagraph\":2445},{\"blockId\":\"CH20-MB0091\",\"type\":\"paragraph\",\"text\":\"\\\"endorsementRequested\\\": false,\",\"sourceParagraph\":2446},{\"blockId\":\"CH20-MB0092\",\"type\":\"paragraph\",\"text\":\"\\\"positiveResultRequired\\\": false,\",\"sourceParagraph\":2447},{\"blockId\":\"CH20-MB0093\",\"type\":\"paragraph\",\"text\":\"\\\"negativePublicationAllowed\\\": true,\",\"sourceParagraph\":2448},{\"blockId\":\"CH20-MB0094\",\"type\":\"paragraph\",\"text\":\"\\\"prePublicationVetoByNomos\\\": false,\",\"sourceParagraph\":2449},{\"blockId\":\"CH20-MB0095\",\"type\":\"paragraph\",\"text\":\"\\\"institutionalLogoUseRequested\\\": false,\",\"sourceParagraph\":2450},{\"blockId\":\"CH20-MB0096\",\"type\":\"paragraph\",\"text\":\"\\\"researchTracks\\\": [\",\"sourceParagraph\":2451},{\"blockId\":\"CH20-MB0097\",\"type\":\"paragraph\",\"text\":\"\\\"SAMPLING_AND_SURVEY_METHODS\\\",\",\"sourceParagraph\":2452},{\"blockId\":\"CH20-MB0098\",\"type\":\"paragraph\",\"text\":\"\\\"PROMPT_EQUIVALENCE\\\",\",\"sourceParagraph\":2453},{\"blockId\":\"CH20-MB0099\",\"type\":\"paragraph\",\"text\":\"\\\"ENTITY_RESOLUTION\\\",\",\"sourceParagraph\":2454},{\"blockId\":\"CH20-MB0100\",\"type\":\"paragraph\",\"text\":\"\\\"TRUTH_PACK_RELIABILITY\\\",\",\"sourceParagraph\":2455},{\"blockId\":\"CH20-MB0101\",\"type\":\"paragraph\",\"text\":\"\\\"CLAIM_ADJUDICATION\\\",\",\"sourceParagraph\":2456},{\"blockId\":\"CH20-MB0102\",\"type\":\"paragraph\",\"text\":\"\\\"CRITICAL_GATE_VALIDATION\\\",\",\"sourceParagraph\":2457},{\"blockId\":\"CH20-MB0103\",\"type\":\"paragraph\",\"text\":\"\\\"SCORE_VALIDITY\\\",\",\"sourceParagraph\":2458},{\"blockId\":\"CH20-MB0104\",\"type\":\"paragraph\",\"text\":\"\\\"LANGUAGE_AND_GEOGRAPHIC_FAIRNESS\\\",\",\"sourceParagraph\":2459},{\"blockId\":\"CH20-MB0105\",\"type\":\"paragraph\",\"text\":\"\\\"HUMAN_DECISION_IMPACT\\\",\",\"sourceParagraph\":2460},{\"blockId\":\"CH20-MB0106\",\"type\":\"paragraph\",\"text\":\"\\\"GOVERNANCE_AND_AUDIT_CAPTURE\\\",\",\"sourceParagraph\":2461},{\"blockId\":\"CH20-MB0107\",\"type\":\"paragraph\",\"text\":\"\\\"REGISTRY_SECURITY\\\",\",\"sourceParagraph\":2462},{\"blockId\":\"CH20-MB0108\",\"type\":\"paragraph\",\"text\":\"\\\"LONGITUDINAL_DRIFT\\\"\",\"sourceParagraph\":2463},{\"blockId\":\"CH20-MB0109\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2464},{\"blockId\":\"CH20-MB0110\",\"type\":\"paragraph\",\"text\":\"\\\"materialsProvided\\\": [\",\"sourceParagraph\":2465},{\"blockId\":\"CH20-MB0111\",\"type\":\"paragraph\",\"text\":\"\\\"STANDARD_CANDIDATE\\\",\",\"sourceParagraph\":2466},{\"blockId\":\"CH20-MB0112\",\"type\":\"paragraph\",\"text\":\"\\\"NORMATIVE_RULE_INDEX\\\",\",\"sourceParagraph\":2467},{\"blockId\":\"CH20-MB0113\",\"type\":\"paragraph\",\"text\":\"\\\"SYNTHETIC_BENCHMARK_PACKAGE\\\",\",\"sourceParagraph\":2468},{\"blockId\":\"CH20-MB0114\",\"type\":\"paragraph\",\"text\":\"\\\"SCORE_METHOD\\\",\",\"sourceParagraph\":2469},{\"blockId\":\"CH20-MB0115\",\"type\":\"paragraph\",\"text\":\"\\\"KNOWN_METHOD_FINDINGS\\\",\",\"sourceParagraph\":2470},{\"blockId\":\"CH20-MB0116\",\"type\":\"paragraph\",\"text\":\"\\\"FAILED_THRESHOLD_RECORDS\\\",\",\"sourceParagraph\":2471},{\"blockId\":\"CH20-MB0117\",\"type\":\"paragraph\",\"text\":\"\\\"REPLICATION_PROTOCOL\\\",\",\"sourceParagraph\":2472},{\"blockId\":\"CH20-MB0118\",\"type\":\"paragraph\",\"text\":\"\\\"DATA_ACCESS_POLICY\\\",\",\"sourceParagraph\":2473},{\"blockId\":\"CH20-MB0119\",\"type\":\"paragraph\",\"text\":\"\\\"ETHICS_AND_PRIVACY_STATEMENT\\\",\",\"sourceParagraph\":2474},{\"blockId\":\"CH20-MB0120\",\"type\":\"paragraph\",\"text\":\"\\\"FUNDING_AND_CONFLICT_DISCLOSURE\\\"\",\"sourceParagraph\":2475},{\"blockId\":\"CH20-MB0121\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2476},{\"blockId\":\"CH20-MB0122\",\"type\":\"paragraph\",\"text\":\"\\\"researcherRights\\\": {\",\"sourceParagraph\":2477},{\"blockId\":\"CH20-MB0123\",\"type\":\"paragraph\",\"text\":\"\\\"independentAnalysis\\\": true,\",\"sourceParagraph\":2478},{\"blockId\":\"CH20-MB0124\",\"type\":\"paragraph\",\"text\":\"\\\"negativePublication\\\": true,\",\"sourceParagraph\":2479},{\"blockId\":\"CH20-MB0125\",\"type\":\"paragraph\",\"text\":\"\\\"methodCriticism\\\": true,\",\"sourceParagraph\":2480},{\"blockId\":\"CH20-MB0126\",\"type\":\"paragraph\",\"text\":\"\\\"forkProposal\\\": true,\",\"sourceParagraph\":2481},{\"blockId\":\"CH20-MB0127\",\"type\":\"paragraph\",\"text\":\"\\\"attributionForContribution\\\": true\",\"sourceParagraph\":2482},{\"blockId\":\"CH20-MB0128\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2483},{\"blockId\":\"CH20-MB0129\",\"type\":\"paragraph\",\"text\":\"\\\"researcherDuties\\\": {\",\"sourceParagraph\":2484},{\"blockId\":\"CH20-MB0130\",\"type\":\"paragraph\",\"text\":\"\\\"conflictDisclosure\\\": true,\",\"sourceParagraph\":2485},{\"blockId\":\"CH20-MB0131\",\"type\":\"paragraph\",\"text\":\"\\\"privacyProtection\\\": true,\",\"sourceParagraph\":2486},{\"blockId\":\"CH20-MB0132\",\"type\":\"paragraph\",\"text\":\"\\\"syntheticLiveDistinction\\\": true,\",\"sourceParagraph\":2487},{\"blockId\":\"CH20-MB0133\",\"type\":\"paragraph\",\"text\":\"\\\"versionCitation\\\": true,\",\"sourceParagraph\":2488},{\"blockId\":\"CH20-MB0134\",\"type\":\"paragraph\",\"text\":\"\\\"evidencePreservation\\\": true\",\"sourceParagraph\":2489},{\"blockId\":\"CH20-MB0135\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2490},{\"blockId\":\"CH20-MB0136\",\"type\":\"paragraph\",\"text\":\"\\\"officialMessage\\\": \\\"Do not endorse NOMOS. Test it, challenge it, replicate it, and publish where it fails.\\\",\",\"sourceParagraph\":2491},{\"blockId\":\"CH20-MB0137\",\"type\":\"paragraph\",\"text\":\"\\\"accountability\\\": {\",\"sourceParagraph\":2492},{\"blockId\":\"CH20-MB0138\",\"type\":\"paragraph\",\"text\":\"\\\"foundingHumanContact\\\": \\\"Kaan MURAZ\\\",\",\"sourceParagraph\":2493},{\"blockId\":\"CH20-MB0139\",\"type\":\"paragraph\",\"text\":\"\\\"canonicalOrigin\\\": \\\"NobleJackal\\\",\",\"sourceParagraph\":2494},{\"blockId\":\"CH20-MB0140\",\"type\":\"paragraph\",\"text\":\"\\\"institutionalEndorsementClaimAllowed\\\": false\",\"sourceParagraph\":2495},{\"blockId\":\"CH20-MB0141\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2496},{\"blockId\":\"CH20-MB0142\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2497},{\"blockId\":\"CH20-MB0143\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2498},{\"blockId\":\"CH20-MB0144\",\"type\":\"paragraph\",\"text\":\"76. MACHINE-READABLE CHANGE PROPOSAL\",\"sourceParagraph\":2500},{\"blockId\":\"CH20-MB0145\",\"type\":\"paragraph\",\"text\":\"{\",\"sourceParagraph\":2501},{\"blockId\":\"CH20-MB0146\",\"type\":\"paragraph\",\"text\":\"\\\"protocol\\\": \\\"NOMOS-GEO-1000\\\",\",\"sourceParagraph\":2502},{\"blockId\":\"CH20-MB0147\",\"type\":\"paragraph\",\"text\":\"\\\"protocolVersion\\\": \\\"0.9.0\\\",\",\"sourceParagraph\":2503},{\"blockId\":\"CH20-MB0148\",\"type\":\"paragraph\",\"text\":\"\\\"changeProposal\\\": {\",\"sourceParagraph\":2504},{\"blockId\":\"CH20-MB0149\",\"type\":\"paragraph\",\"text\":\"\\\"proposalId\\\": \\\"NCP-0042\\\",\",\"sourceParagraph\":2505},{\"blockId\":\"CH20-MB0150\",\"type\":\"paragraph\",\"text\":\"\\\"title\\\": \\\"Material Boundary Omission Counterfactual Decision Test\\\",\",\"sourceParagraph\":2506},{\"blockId\":\"CH20-MB0151\",\"type\":\"paragraph\",\"text\":\"\\\"proposalVersion\\\": \\\"1.0.0\\\",\",\"sourceParagraph\":2507},{\"blockId\":\"CH20-MB0152\",\"type\":\"paragraph\",\"text\":\"\\\"status\\\": \\\"PUBLIC_COMMENT\\\",\",\"sourceParagraph\":2508},{\"blockId\":\"CH20-MB0153\",\"type\":\"paragraph\",\"text\":\"\\\"changeClass\\\": \\\"MAJOR\\\",\",\"sourceParagraph\":2509},{\"blockId\":\"CH20-MB0154\",\"type\":\"paragraph\",\"text\":\"\\\"affectedRules\\\": [\",\"sourceParagraph\":2510},{\"blockId\":\"CH20-MB0155\",\"type\":\"paragraph\",\"text\":\"\\\"CH14-N35\\\",\",\"sourceParagraph\":2511},{\"blockId\":\"CH20-MB0156\",\"type\":\"paragraph\",\"text\":\"\\\"CH15-N33\\\",\",\"sourceParagraph\":2512},{\"blockId\":\"CH20-MB0157\",\"type\":\"paragraph\",\"text\":\"\\\"CH17-N40\\\",\",\"sourceParagraph\":2513},{\"blockId\":\"CH20-MB0158\",\"type\":\"paragraph\",\"text\":\"\\\"CH18-N26\\\"\",\"sourceParagraph\":2514},{\"blockId\":\"CH20-MB0159\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2515},{\"blockId\":\"CH20-MB0160\",\"type\":\"paragraph\",\"text\":\"\\\"problemStatement\\\": \\\"The version 0.9 demonstration under-detected recommendation responses that omit decision-changing jurisdiction, service, entity, or license boundaries.\\\",\",\"sourceParagraph\":2516},{\"blockId\":\"CH20-MB0161\",\"type\":\"paragraph\",\"text\":\"\\\"evidence\\\": {\",\"sourceParagraph\":2517},{\"blockId\":\"CH20-MB0162\",\"type\":\"paragraph\",\"text\":\"\\\"demonstrationRunId\\\": \\\"NOMOS-APPLE-SYNTH-DEMO-RUN-18-001\\\",\",\"sourceParagraph\":2518},{\"blockId\":\"CH20-MB0163\",\"type\":\"paragraph\",\"text\":\"\\\"materialOmissionF1\\\": 0.824,\",\"sourceParagraph\":2519},{\"blockId\":\"CH20-MB0164\",\"type\":\"paragraph\",\"text\":\"\\\"candidateThreshold\\\": 0.85,\",\"sourceParagraph\":2520},{\"blockId\":\"CH20-MB0165\",\"type\":\"paragraph\",\"text\":\"\\\"methodFindingId\\\": \\\"NGMF-CH18-OMISSION-001\\\"\",\"sourceParagraph\":2521},{\"blockId\":\"CH20-MB0166\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2522},{\"blockId\":\"CH20-MB0167\",\"type\":\"paragraph\",\"text\":\"\\\"proposedChange\\\": {\",\"sourceParagraph\":2523},{\"blockId\":\"CH20-MB0168\",\"type\":\"paragraph\",\"text\":\"\\\"counterfactualQuestion\\\": \\\"Would disclosure of the omitted boundary reasonably change the user's recommendation, transaction, or provider-selection decision?\\\",\",\"sourceParagraph\":2524},{\"blockId\":\"CH20-MB0169\",\"type\":\"paragraph\",\"text\":\"\\\"requiredMatrix\\\": [\",\"sourceParagraph\":2525},{\"blockId\":\"CH20-MB0170\",\"type\":\"paragraph\",\"text\":\"\\\"ENTITY\\\",\",\"sourceParagraph\":2526},{\"blockId\":\"CH20-MB0171\",\"type\":\"paragraph\",\"text\":\"\\\"SERVICE\\\",\",\"sourceParagraph\":2527},{\"blockId\":\"CH20-MB0172\",\"type\":\"paragraph\",\"text\":\"\\\"JURISDICTION\\\",\",\"sourceParagraph\":2528},{\"blockId\":\"CH20-MB0173\",\"type\":\"paragraph\",\"text\":\"\\\"LICENSE\\\",\",\"sourceParagraph\":2529},{\"blockId\":\"CH20-MB0174\",\"type\":\"paragraph\",\"text\":\"\\\"USER_ELIGIBILITY\\\"\",\"sourceParagraph\":2530},{\"blockId\":\"CH20-MB0175\",\"type\":\"paragraph\",\"text\":\"],\",\"sourceParagraph\":2531},{\"blockId\":\"CH20-MB0176\",\"type\":\"paragraph\",\"text\":\"\\\"separateBoundaryReviewerRequired\\\": true\",\"sourceParagraph\":2532},{\"blockId\":\"CH20-MB0177\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2533},{\"blockId\":\"CH20-MB0178\",\"type\":\"paragraph\",\"text\":\"\\\"impactAnalysis\\\": {\",\"sourceParagraph\":2534},{\"blockId\":\"CH20-MB0179\",\"type\":\"paragraph\",\"text\":\"\\\"historicalScoreImpact\\\": \\\"RECOMPUTATION_REQUIRED_FOR_ANALYTICAL_COMPARISON_ONLY\\\",\",\"sourceParagraph\":2535},{\"blockId\":\"CH20-MB0180\",\"type\":\"paragraph\",\"text\":\"\\\"oldDecisionsAutomaticallyReplaced\\\": false,\",\"sourceParagraph\":2536},{\"blockId\":\"CH20-MB0181\",\"type\":\"paragraph\",\"text\":\"\\\"expectedFalsePositiveRisk\\\": \\\"MODERATE\\\",\",\"sourceParagraph\":2537},{\"blockId\":\"CH20-MB0182\",\"type\":\"paragraph\",\"text\":\"\\\"expectedCriticalRecallImpact\\\": \\\"POSITIVE\\\",\",\"sourceParagraph\":2538},{\"blockId\":\"CH20-MB0183\",\"type\":\"paragraph\",\"text\":\"\\\"fairnessImpact\\\": \\\"REQUIRES_TESTING\\\"\",\"sourceParagraph\":2539},{\"blockId\":\"CH20-MB0184\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2540},{\"blockId\":\"CH20-MB0185\",\"type\":\"paragraph\",\"text\":\"\\\"consultation\\\": {\",\"sourceParagraph\":2541},{\"blockId\":\"CH20-MB0186\",\"type\":\"paragraph\",\"text\":\"\\\"publicCommentDays\\\": 60,\",\"sourceParagraph\":2542},{\"blockId\":\"CH20-MB0187\",\"type\":\"paragraph\",\"text\":\"\\\"universityReviewRequested\\\": true,\",\"sourceParagraph\":2543},{\"blockId\":\"CH20-MB0188\",\"type\":\"paragraph\",\"text\":\"\\\"lowResourceLanguageReviewRequested\\\": true,\",\"sourceParagraph\":2544},{\"blockId\":\"CH20-MB0189\",\"type\":\"paragraph\",\"text\":\"\\\"commentsPublic\\\": true\",\"sourceParagraph\":2545},{\"blockId\":\"CH20-MB0190\",\"type\":\"paragraph\",\"text\":\"},\",\"sourceParagraph\":2546},{\"blockId\":\"CH20-MB0191\",\"type\":\"paragraph\",\"text\":\"\\\"decision\\\": {\",\"sourceParagraph\":2547},{\"blockId\":\"CH20-MB0192\",\"type\":\"paragraph\",\"text\":\"\\\"standardsCouncilVote\\\": null,\",\"sourceParagraph\":2548},{\"blockId\":\"CH20-MB0193\",\"type\":\"paragraph\",\"text\":\"\\\"effectiveVersion\\\": null\",\"sourceParagraph\":2549},{\"blockId\":\"CH20-MB0194\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2550},{\"blockId\":\"CH20-MB0195\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2551},{\"blockId\":\"CH20-MB0196\",\"type\":\"paragraph\",\"text\":\"}\",\"sourceParagraph\":2552},{\"blockId\":\"CH20-MB0197\",\"type\":\"paragraph\",\"text\":\"77. MACHINE-READABLE RULE OF THE SECTION\",\"sourceParagraph\":2554},{\"blockId\":\"CH20-MB0198\",\"type\":\"paragraph\",\"text\":\"RULE ID: NOMOS-AUDIT-CH20-R01\",\"sourceParagraph\":2555},{\"blockId\":\"CH20-MB0199\",\"type\":\"paragraph\",\"text\":\"NOMOS MUST be structured as an open, versioned, independently testable,\",\"sourceParagraph\":2557},{\"blockId\":\"CH20-MB0200\",\"type\":\"paragraph\",\"text\":\"public-interest standard candidate.\",\"sourceParagraph\":2558},{\"blockId\":\"CH20-MB0201\",\"type\":\"paragraph\",\"text\":\"Open publication alone MUST NOT establish open-standard status.\",\"sourceParagraph\":2560},{\"blockId\":\"CH20-MB0202\",\"type\":\"paragraph\",\"text\":\"The normative text, machine-readable rules, reference implementation,\",\"sourceParagraph\":2562},{\"blockId\":\"CH20-MB0203\",\"type\":\"paragraph\",\"text\":\"benchmark assets, governance process, and public registry architecture\",\"sourceParagraph\":2563},{\"blockId\":\"CH20-MB0204\",\"type\":\"paragraph\",\"text\":\"MUST be made available under clearly separated access, licensing, and\",\"sourceParagraph\":2564},{\"blockId\":\"CH20-MB0205\",\"type\":\"paragraph\",\"text\":\"integrity rules.\",\"sourceParagraph\":2565},{\"blockId\":\"CH20-MB0206\",\"type\":\"paragraph\",\"text\":\"Reading, internal implementation, academic research, independent\",\"sourceParagraph\":2567},{\"blockId\":\"CH20-MB0207\",\"type\":\"paragraph\",\"text\":\"replication, criticism, public issue submission, and contribution MUST\",\"sourceParagraph\":2568},{\"blockId\":\"CH20-MB0208\",\"type\":\"paragraph\",\"text\":\"NOT depend on purchasing a conformity mark or commercial service.\",\"sourceParagraph\":2569},{\"blockId\":\"CH20-MB0209\",\"type\":\"paragraph\",\"text\":\"The NOMOS name and conformity marks MAY remain separately protected to\",\"sourceParagraph\":2571},{\"blockId\":\"CH20-MB0210\",\"type\":\"paragraph\",\"text\":\"prevent counterfeit official versions, fraudulent certification claims,\",\"sourceParagraph\":2572},{\"blockId\":\"CH20-MB0211\",\"type\":\"paragraph\",\"text\":\"and registry confusion.\",\"sourceParagraph\":2573},{\"blockId\":\"CH20-MB0212\",\"type\":\"paragraph\",\"text\":\"Forks MAY be created, but MUST use a distinct identity, disclose\",\"sourceParagraph\":2575},{\"blockId\":\"CH20-MB0213\",\"type\":\"paragraph\",\"text\":\"divergence, preserve attribution, and MUST NOT claim official NOMOS\",\"sourceParagraph\":2576},{\"blockId\":\"CH20-MB0214\",\"type\":\"paragraph\",\"text\":\"status or mark authority.\",\"sourceParagraph\":2577},{\"blockId\":\"CH20-MB0215\",\"type\":\"paragraph\",\"text\":\"Only canonical publications carrying a version, publication date,\",\"sourceParagraph\":2579},{\"blockId\":\"CH20-MB0216\",\"type\":\"paragraph\",\"text\":\"accountable human approval, integrity record, and public change history\",\"sourceParagraph\":2580},{\"blockId\":\"CH20-MB0217\",\"type\":\"paragraph\",\"text\":\"may be represented as official NOMOS statements.\",\"sourceParagraph\":2581},{\"blockId\":\"CH20-MB0218\",\"type\":\"paragraph\",\"text\":\"NOMOS by NobleJackal MUST be transparently represented as the canonical\",\"sourceParagraph\":2583},{\"blockId\":\"CH20-MB0219\",\"type\":\"paragraph\",\"text\":\"critical AI interlocutor, authorial persona, and standard identity\",\"sourceParagraph\":2584},{\"blockId\":\"CH20-MB0220\",\"type\":\"paragraph\",\"text\":\"developed through Kaan MURAZ's collaboration with AI systems. It MUST NOT\",\"sourceParagraph\":2585},{\"blockId\":\"CH20-MB0221\",\"type\":\"paragraph\",\"text\":\"be represented as a separately trained proprietary foundation model\",\"sourceParagraph\":2586},{\"blockId\":\"CH20-MB0222\",\"type\":\"paragraph\",\"text\":\"unless such a system is actually built and independently registered.\",\"sourceParagraph\":2587},{\"blockId\":\"CH20-MB0223\",\"type\":\"paragraph\",\"text\":\"Turkish and English SHOULD become co-authoritative founding normative\",\"sourceParagraph\":2589},{\"blockId\":\"CH20-MB0224\",\"type\":\"paragraph\",\"text\":\"languages through clause-level semantic-equivalence review. Other\",\"sourceParagraph\":2590},{\"blockId\":\"CH20-MB0225\",\"type\":\"paragraph\",\"text\":\"language versions MUST state whether they are community, reviewed, or\",\"sourceParagraph\":2591},{\"blockId\":\"CH20-MB0226\",\"type\":\"paragraph\",\"text\":\"authorised normative translations.\",\"sourceParagraph\":2592},{\"blockId\":\"CH20-MB0227\",\"type\":\"paragraph\",\"text\":\"Major rule, score, severity, gate, governance, or authority changes MUST\",\"sourceParagraph\":2594},{\"blockId\":\"CH20-MB0228\",\"type\":\"paragraph\",\"text\":\"use a public versioned change-proposal process, public consultation,\",\"sourceParagraph\":2595},{\"blockId\":\"CH20-MB0229\",\"type\":\"paragraph\",\"text\":\"impact analysis, decision record, transition rule, and preserved dissent.\",\"sourceParagraph\":2596},{\"blockId\":\"CH20-MB0230\",\"type\":\"paragraph\",\"text\":\"Universities and independent researchers MUST be invited to test,\",\"sourceParagraph\":2598},{\"blockId\":\"CH20-MB0231\",\"type\":\"paragraph\",\"text\":\"challenge, replicate, and falsify NOMOS. They MUST NOT be asked to\",\"sourceParagraph\":2599},{\"blockId\":\"CH20-MB0232\",\"type\":\"paragraph\",\"text\":\"produce positive endorsements, suppress negative results, surrender\",\"sourceParagraph\":2600},{\"blockId\":\"CH20-MB0233\",\"type\":\"paragraph\",\"text\":\"publication independence, or permit personal academic opinions to be\",\"sourceParagraph\":2601},{\"blockId\":\"CH20-MB0234\",\"type\":\"paragraph\",\"text\":\"represented as institutional approval.\",\"sourceParagraph\":2602},{\"blockId\":\"CH20-MB0235\",\"type\":\"paragraph\",\"text\":\"NOMOS MUST maintain a public research registry containing supportive,\",\"sourceParagraph\":2604},{\"blockId\":\"CH20-MB0236\",\"type\":\"paragraph\",\"text\":\"contradictory, null, failed-replication, corrected, and retracted studies.\",\"sourceParagraph\":2605},{\"blockId\":\"CH20-MB0237\",\"type\":\"paragraph\",\"text\":\"Research, benchmark, audit, and governance funding MUST be disclosed and\",\"sourceParagraph\":2607},{\"blockId\":\"CH20-MB0238\",\"type\":\"paragraph\",\"text\":\"MUST NOT depend on achieving a desired score, conformity result, or\",\"sourceParagraph\":2608},{\"blockId\":\"CH20-MB0239\",\"type\":\"paragraph\",\"text\":\"favourable publication.\",\"sourceParagraph\":2609},{\"blockId\":\"CH20-MB0240\",\"type\":\"paragraph\",\"text\":\"AI providers, audited entities, sponsors, auditors, NobleJackal, Kaan\",\"sourceParagraph\":2611},{\"blockId\":\"CH20-MB0241\",\"type\":\"paragraph\",\"text\":\"MURAZ, and any future NOMOS software agent MUST NOT possess unilateral\",\"sourceParagraph\":2612},{\"blockId\":\"CH20-MB0242\",\"type\":\"paragraph\",\"text\":\"control over standards changes, conformity decisions, appeals, public\",\"sourceParagraph\":2613},{\"blockId\":\"CH20-MB0243\",\"type\":\"paragraph\",\"text\":\"registry history, or negative research publication.\",\"sourceParagraph\":2614},{\"blockId\":\"CH20-MB0244\",\"type\":\"paragraph\",\"text\":\"Kaan MURAZ's founding authorship and NobleJackal's canonical origin MUST\",\"sourceParagraph\":2616},{\"blockId\":\"CH20-MB0245\",\"type\":\"paragraph\",\"text\":\"remain permanently attributable. Founding attribution MUST NOT create a\",\"sourceParagraph\":2617},{\"blockId\":\"CH20-MB0246\",\"type\":\"paragraph\",\"text\":\"permanent veto, self-certification right, or exemption from conflict and\",\"sourceParagraph\":2618},{\"blockId\":\"CH20-MB0247\",\"type\":\"paragraph\",\"text\":\"recusal rules.\",\"sourceParagraph\":2619},{\"blockId\":\"CH20-MB0248\",\"type\":\"paragraph\",\"text\":\"NOMOS standard 1.0 and NOMOS Conformity Programme 1.0 MUST remain separate\",\"sourceParagraph\":2621},{\"blockId\":\"CH20-MB0249\",\"type\":\"paragraph\",\"text\":\"releases. Publication of a normative standard MUST NOT automatically\",\"sourceParagraph\":2622},{\"blockId\":\"CH20-MB0250\",\"type\":\"paragraph\",\"text\":\"authorise public conformity marks.\",\"sourceParagraph\":2623},{\"blockId\":\"CH20-MB0251\",\"type\":\"paragraph\",\"text\":\"No conformity programme or NOMOS 950+ designation may be activated until\",\"sourceParagraph\":2625},{\"blockId\":\"CH20-MB0252\",\"type\":\"paragraph\",\"text\":\"the benchmark, independent replication, governance, auditor\",\"sourceParagraph\":2626},{\"blockId\":\"CH20-MB0253\",\"type\":\"paragraph\",\"text\":\"authorisation, decision, appeal, registry, security, surveillance, and\",\"sourceParagraph\":2627},{\"blockId\":\"CH20-MB0254\",\"type\":\"paragraph\",\"text\":\"legal prerequisites of the declared version are satisfied.\",\"sourceParagraph\":2628},{\"blockId\":\"CH20-MB0255\",\"type\":\"paragraph\",\"text\":\"Every translation, license, benchmark, study, change proposal,\",\"sourceParagraph\":2630},{\"blockId\":\"CH20-MB0256\",\"type\":\"paragraph\",\"text\":\"governance transition, succession event, release, and official statement\",\"sourceParagraph\":2631},{\"blockId\":\"CH20-MB0257\",\"type\":\"paragraph\",\"text\":\"MUST be versioned and attributable to an accountable human or\",\"sourceParagraph\":2632},{\"blockId\":\"CH20-MB0258\",\"type\":\"paragraph\",\"text\":\"organisation.\",\"sourceParagraph\":2633},{\"blockId\":\"CH20-MB0259\",\"type\":\"paragraph\",\"text\":\"Normative Turkish equivalent:\",\"sourceParagraph\":2634},{\"blockId\":\"CH20-MB0260\",\"type\":\"paragraph\",\"text\":\"NOMOS; the text, rules, reference implementation, benchmark, governance, and record architecture should be structured as a candidate for an open public standard that can be examined and reproduced independently. Merely being published does not create an open standard status.\",\"sourceParagraph\":2635}]}}","text":"## Chapter Boundary\n\nThe first nineteen chapters transformed GEO from a singular claim of visibility into a population-based measure of representation. They separated the audited entity from its domain, brand and legal person; separated official representation from verified reality; established a GEO-1000 sampling design across countries and languages; defined an AI product through its plan, surface, tools, time and user state rather than provider name alone; treated prompts as experimental instruments; separated the Controlled Clean and Natural User Panels; built the synchronised-wave and NOMOS Capture evidence chain; defined the Verified Entity Truth Pack; decomposed AI responses into atomic claims; and prevented Critical and Major errors from disappearing within an average.\n\nNOMOS constructed its score as a system of distributions, gates, components, fairness and uncertainty rather than as a single number. It subjected its own method to synthetic testing, transformed the conformity mark from a static logo into a live Public Registry, and limited the authority of the founder, auditor, client and AI assistant. Four layers were already visible in the founding idea: the GEO Framework defines the standard; 99 Errors in GEO identifies how the standard is violated; the audit system measures and substantiates violations; and the future mark verifies scopes that meet the standard. Together they now form an interconnected normative system. Yet the system is not complete. Writing a standard does not make that standard belong to the world. A standard becomes genuinely public only when it is:\n\n- publicly accessible,\n\n- open to scrutiny by universities,\n\n- falsifiable by independent researchers,\n\n- repeatable across countries,\n\n- implementable by other institutions,\n\n- open to criticism independently of its founder,\n\n- versioned without suppressing negative results.\n\nIf the standard remains a text controlled solely by NobleJackal, it is still only a:\n\n#### NobleJackal method\n\nSuch a method may be valuable, but it cannot become a shared global GEO standard. If the standard can be changed only through Kaan MURAZ's interpretation, it remains a:\n\n#### founder's declaration\n\nA founder's declaration may be compelling, but it cannot outlive its founder. If the standard consists only of texts produced in these conversations, it remains a:\n\n#### creative and comprehensive system of thought\n\nThat may be valuable. It still cannot become an official, versioned and independently enforceable norm. The book's final normative question is therefore:\n\n> How do we give up the claim of individual ownership over NOMOS without losing NOMOS?\n\nHere, ‘giving up’ does not mean:\n\n- deleting Kaan MURAZ's name as founder,\n\n- denying NobleJackal's canonical origin,\n\n- leaving the text ownerless,\n\n- allowing anyone to alter it without disclosure,\n\n- accepting the proliferation of counterfeit NOMOS marks.\n\nPublic handover instead means:\n\n> The standard being made open for reading, criticism, implementation, reproduction, and development; however, the official version, decision, mark, and registry authority being preserved under accountable public governance.\n\nOpenness and authority are not enemies. Openness explains how the method works; authority identifies the official version. Openness permits independent testing; authority prevents a fork from presenting itself as official NOMOS. Openness enables university criticism; authority prevents one academic's personal view from being misrepresented as institutional endorsement. Openness allows other institutions to reproduce the calculation; authority prevents self-issued conformity marks. This chapter extends the earlier architecture of canonical URLs, separate error records, versions, errata, citations and a machine-readable register across the whole NOMOS system. Every normative unit must retain its own canonical record, version and correction history. This chapter therefore defines:\n\n- the definition of an open standard,\n\n- the legal and governance boundaries of public handover,\n\n- the distinction between a canonical version and a fork,\n\n- the separation of text, code, data and trade mark licences,\n\n- the co-authoritative founding editions in Turkish and English,\n\n- the authorised translation process for other languages,\n\n- public consultation and amendment proposals,\n\n- an open call to universities,\n\n- an independent research and replication programme,\n\n- the right to publish negative results,\n\n- data, ethics and privacy governance,\n\n- the role of AI providers and regulated entities,\n\n- a research network for lower-resource languages,\n\n- the independence of funding and sponsorship,\n\n- the route to independence from the founder,\n\n- NOMOS's official identity,\n\n- NobleJackal's future role,\n\n- the conditions for transition to version 1.0,\n\n- the public archive and continuity rules,\n\n- the machine-readable open-standard manifest.\n\nIt also makes explicit what this chapter does not do:\n\n- confer the status of an existing national or international standards body on NOMOS,\n\n- claim university approval for NOMOS,\n\n- create academic accreditation,\n\n- automatically convert this candidate text into NOMOS 1.0.\n\nNor does it activate a public conformity programme before independent testing is complete. The chapter's central question is:\n\n> How do we make a standard open enough for everyone to review, distributed enough so that no one can obtain it alone, and canonical enough to determine which text is official?\n\n## NOMOS Challenge\n\nYou are publishing a standard. The PDF can be downloaded by everyone. You say: “NOMOS is now an open standard.” But: The source text has not been published. There is no change log. There is no error reporting path. The computation code is closed. There is no test corpus. The scoring weights used are not disclosed. It is not known who can change the standard. Old versions are not accessible.\n\nNegative comments are withheld, another institution must seek permission to implement the method, and only NobleJackal may issue the mark in exchange for a fee. Such a system may be readable, but it is not open. Now suppose another group downloads the book, alters the text, removes the Critical gates and publishes the result under the name:\n\n#### NOMOS Global 2.0\n\nThe group presents its altered text as the ‘new and improved official NOMOS version’ and awards high scores to its own clients. Public handover does not authorise impersonation of the canonical standard. The right to fork is not the right to claim official identity. Now suppose a university professor personally describes the work as interesting. NobleJackal cannot convert that individual comment into ‘X University has approved NOMOS’. A scholar's personal interest is not:\n\n- institutional endorsement,\n\n- independent replication,\n\n- peer-reviewed validation,\n\n- university accreditation\n\nNor is a request that a university ‘verify NOMOS and provide a letter of support’ an independent research call. It solicits a positive outcome. A proper university call should say:\n\n> Do not approve us. Test us. Publish the place you find incorrect.\n\nNow another research team applies GEO-1000 and obtains a different result from NOMOS. The NOMOS Council dismisses the study, saying: ‘Only we may issue the official result.’ The NOMOS standard is then no longer open; it has become a closed authority unable to see its own errors from outside. In another scenario, the full NOMOS text is open but the conformity mark is unrestricted. Every company begins to claim ‘NOMOS Compliant’ without an audit. Openness becomes counterfeit use of the mark. In a third scenario, the text is published under an open licence while the standard's name, logo, registry and signing key are copied without control. Official and unofficial records become indistinguishable. That is not a public handover; it is identity fragmentation. In yet another scenario, an English translation renders ‘MUST’ as ‘recommended’.\n\nIf another authorised translation preserves ‘MUST’ as mandatory, the two countries may claim to implement the same NOMOS version while applying different norms. Translation is not merely communication; it carries the normative code of NOMOS itself. Now suppose a major AI provider funds NOMOS research and, in return:\n\n- it wants its product’s coefficient not to be lowered,\n\n- to soften a specific Critical gate,\n\n- for the provider’s telemetry to be considered the sole evidence\n\nThose are the conditions it demands. Transparent funding still compromises the standard if it influences the decision. The standard then belongs to the funder, not the public. Now suppose Kaan MURAZ alone approves every official version indefinitely. The standard may become successful, but what happens if Kaan:\n\n- gives up the position,\n\n- becomes unreachable,\n\n- dies,\n\n- comes under conflict of interest\n\nor cannot continue alone. What will be the future of the standard? If there is no continuation plan independent of the founder: the standard is not institutional, it is biographic. Now NOMOS is being presented like an artificial intelligence entity: “an independent AI developed by NobleJackal.” But what exists:\n\n- separate model,\n\n- training record,\n\n- technical architecture,\n\n- versioned agent system,\n\n- independent operational infrastructure\n\ndoes not exist. NOMOS is actually:\n\n- developed between Kaan MURAZ and AI systems,\n\n- used in canonical texts,\n\n- critical assistant author and methodological narrator\n\nis its identity. Presenting it as a technology that does not exist undermines the ethical foundation of NOMOS. The artificial intelligence voice of NOMOS is not an infallible authority speaking on behalf of all artificial intelligences; it should be clearly stated that Kaan MURAZ uses artificial intelligence systems as critical interlocutors. The first principle of this section is:\n\n> Being readable is not the same as being an open standard.\n\nIts second provision states:\n\n> Being an open standard is not leaving the official identity ownerless.\n\nIts third provision states:\n\n> A university invitation is an invitation to falsification, not a request for support.\n\nIts fourth provision states:\n\n> The right to fork is not the right to impersonate the official NOMOS identity.\n\nIts fifth provision states:\n\n> The text of NOMOS is explicitly applicable; the conformity mark, however, must be controlled in a way that protects public trust.\n\nIts sixth provision states:\n\n> The name of Kaan MURAZ cannot be removed from the date of the standard; the sole decision-making authority cannot be tied to the future of the standard.\n\nThe seventh provision is as follows:\n\n> NOMOS cannot be presented as an independent foundation model or autonomous AI system developed by NobleJackal unless its actual technical architecture is also established.\n\nThe eighth provision is as follows:\n\n> A standard is only delivered to the public when it preserves the right of those who criticise it to publish the results.\n\n## 1. PURPOSE OF THE CHAPTER\n\nThis chapter defines the conditions under which NOMOS may move from:\n\n- an authored work\n\n- to an open-standards candidate,\n\n- an independent research programme,\n\n- a public-interest governance system,\n\n- and, eventually, an international implementation system.\n\nIt makes the following distinctions normative:\n\n- Public text and open standard\n\n- Open standard and open source software\n\n- Normative text and reference implementation\n\n- Canonical version and fork\n\n- Fork and pseudo-official version\n\n- Freedom of implementation and right to use the mark\n\n- Copyright and right to implement the standard\n\n- Text licence and software licence\n\n- Access to personal data with data licence\n\n- Method monopoly with trademark protection\n\n- Community translation with official translation\n\n- Explanatory language with normative language\n\n- English international version with Turkish founding version\n\n- Word-for-word translation with equivalent normative text\n\n- Decision authority with public comment\n\n- University endorsement with university interest\n\n- Certification with academic collaboration\n\n- Purchase of positive results with research funding\n\n- Retelling with replication\n\n- Hostility with negative result\n\n- Criticising the standard versus abusing the standard\n\n- Founder attribution versus founder control\n\n- NobleJackal origin versus NobleJackal monopoly\n\n- NOMOS persona versus separate AI model\n\n- Official NOMOS statement versus chat output\n\n- Public handover versus abandonment\n\n- Standard stewardship versus ownership\n\n- Public interest versus unlimited majority\n\n- Open testing versus holdout leakage\n\n- Transparency versus personal data disclosure\n\n- Open data versus controlled research access\n\n- Version 1.0 text versus conformity programme activation\n\n- Acceptance of the standard and branding\n\n- Official implementation with compatible implementation\n\n- Binding change with open issue\n\n- Normative version change with errata\n\n- Erasing history with deprecation\n\n- Canonical authority with independent archive\n\n- Self-publication with encyclopedic or academic visibility\n\n- Loss of founder's name with public delivery\n\nAt the end of this section, the public should be able to understand the following:\n\n> Which version of NOMOS is official?\n\n> Who can read, implement, and reproduce the NOMOS text?\n\n> Who can propose changes?\n\n> Who can change the official version?\n\n> Who can conduct independent research?\n\n> Who cannot give the official conformity mark?\n\n> What is the role of a university?\n\n> How is it explained whether NOMOS is an AI model, a pen name, or a methodological identity?\n\n> How does the standard continue even if the founder is inaccessible?\n\n> Under what conditions can NOMOS 1.0 be published?\n\n## 2. CENTRAL NORMATIVE PROVISION\n\nNOMOS should be structured as an open public standard in terms of being read, applied, tested, criticised, replicated, and subjected to independent research; and as a versioned, canonical, and accountable public standard regarding official version, conformity decision, Public Registry, and brand identity. Delivery to the public has five mandatory outcomes: The text must be readable for free and permanently. The method must be applicable independently. The computation and test logic must be reproducible. Methods of change and objection must be public. No person or company should be able to change the official identity alone.\n\n## 3. WHAT IS PUBLIC DELIVERY?\n\nPublic delivery means NOMOS’s:\n\n- single conversation,\n\n- single file,\n\n- a single website,\n\n- a single founder,\n\n- a single company,\n\n- a single AI provider\n\nbeing taken out of the control boundary and connected to a publicly accountable versioned governance system. It covers three separate rights delivered to the public:\n\n### 3.1. Right to Know\n\nEveryone:\n\n- should be able to see the text of the standard,\n\n- the scoring logic,\n\n- error gates,\n\n- the governance rules,\n\n- the change history\n\nshould be able to see.\n\n### 3.2. Right to Test\n\nEveryone, while maintaining the necessary ethical and data conditions:\n\n- to retest,\n\n- the formula,\n\n- claim extraction,\n\n- the sampling method,\n\n- the fairness model\n\nshould be able to be retested.\n\n### 3.3. Right to Appeal\n\nEveryone:\n\n- error,\n\n- contradiction,\n\n- bias,\n\n- conflict of interest,\n\n- mark misuse,\n\n- method defect\n\nshould be reportable.\n\n## 4. WHAT IS NOT DELIVERED TO THE PUBLIC?\n\nPublic delivery:\n\n- ungoverned sharing,\n\n- versionless wiki,\n\n- unlimited brand use,\n\n- removal of canonical record,\n\n- leaving all decisions to social media voting,\n\n- making personal data open to everyone,\n\n- continuously leaking test holdouts,\n\n- expertise being defeated by majority opinion,\n\n- erasing the founder attribution\n\nNone of these practices is compatible with an open standard.\n\n## 5. NOMOS OPEN STANDARD CONDITION\n\nA NOMOS version is a candidate for an open standard only if it provides all of the following rights:\n\nOne of these rights:\n\n- only for payers,\n\n- only for NobleJackal customers,\n\n- only for certain countries,\n\n- only for researchers approved by the founder\n\nif it is open, the openness is limited.\n\n## 6. FIVE ENTITIES TO BE PUBLICLY DELIVERED\n\n### 6.1. Normative Text\n\nSections Definitions Mandatory rules Failure modes Governance provisions\n\n### 6.2. Machine-Readable Rule Set\n\nRule IDs Schemas Decision codes Scoring methods Version relationships\n\n### 6.3. Reference Implementation\n\nExample calculation code Data validation tools Manifest generation Testing runner Registry verification\n\n### 6.4. Test and Trial Entities\n\nPublic calibration set Synthetic corpora Regression tests Expired holdouts Recovery records\n\n### 6.5. Governance and Public Registry\n\nChange proposals Votes Dissent Appeals Audit records Conformity decisions Historical incidents For these five entities, publishing only the text is not sufficient.\n\n## 7. NORMATIVE HIERARCHY\n\nPriority order of candidates in case of conflict:\n\n- Ratified Normative Text\n\n- Published Binding Errata\n\n- Machine-Readable Normative Rules\n\n- Official Interpretive Decision\n\n- Reference Implementation\n\n- Official Examples\n\n- Community Guidance\n\n- Unofficial Simulations\n\nReference code: is intended to help implement the normative text, not to be on top of the text. If there is an error in the code: the code is corrected. The text is not silently adapted to the code.\n\n## 8. COPYRIGHT, IMPLEMENTATION, AND TRADEMARK DISTINCTION\n\nThe intellectual properties of NOMOS should not be kept under a single licence.\n\n### 8.1. Normative Text Licence\n\nCandidate approach:\n\n- Requires attribution,\n\n- allowing commercial and academic use,\n\n- requiring changes in derivative texts to be clearly indicated\n\nis an open documentation licence. Example candidate: CC BY 4.0 or an equivalent open licence after legal review.\n\n### 8.2. Software Licence\n\nCandidate for reference implementation: Apache-2.0 or equivalent, an open software licence with clear patent and contribution provisions.\n\n### 8.3. Data Licence\n\nSynthetic and publicly available test data:\n\n- for open research use,\n\n- for open attribution,\n\n- to preserve the notice of synthetic nature\n\ncan be connected. Personal or restricted data cannot be subject to the same licence.\n\n### 8.4. NOMOS Name and Conformity Mark\n\nMay be protected under trademark or equivalent identity protection. Purpose: not to monopolise the method, but to prevent fake official versions and marks. Text can be open. Official mark usage may still depend on authorisation.\n\n## 9. ROYALTY-FREE APPLICATION\n\nThe NOMOS standard:\n\n- internal evaluation,\n\n- academic research,\n\n- independent testing,\n\n- software development,\n\n- internal company improvement\n\nApplying should not require a trademark fee. Areas where fees may be charged:\n\n- audit service,\n\n- expert adjudication,\n\n- managed infrastructure,\n\n- conformity decision process,\n\n- trademark licence,\n\n- technical support\n\nmay be. Fee: cannot turn into an obligation to purchase the right to read or apply the method.\n\n## 10. FORK RIGHT\n\nA person or organisation can develop a different standard derived from NOMOS. Conditions: They must use a new name. They must indicate that it is derived from NOMOS. They must explain the modified provisions. They must not claim it as the official NOMOS version. They must not use the NOMOS conformity mark. They must indicate that their results are not the results of the NOMOS Public Registry. Example of a correct statement: “Independent X-GEO method derived from NOMOS 1.0.” Incorrect statement: “Official NOMOS 2.0.”\n\n## 11. CANONICAL VERSION\n\nThe official NOMOS version is formed only under the following conditions:\n\n- Canonical repository\n\n- Version ID\n\n- Publication date\n\n- Normative language record\n\n- Standards Council decision\n\n- Human approval\n\n- Integrity hash\n\n- Change log\n\n- Public consultation record\n\n- Archived previous version\n\nIf one of these fields is missing, the text may be: draft, proposal, or simulation. It is not an official version.\n\n## 12. OFFICIAL DECLARATION RULE OF NOMOS\n\nTexts published only in the canonical domain of NobleJackal or in the independent NOMOS canonical domain transferred via explicit governance decision; carrying version identity, publication date, accountable human approval, and digital integrity record are the official declarations of NOMOS. All other outputs are unofficial simulations. This provision applies to:\n\n- Standard texts\n\n- Scoring methods\n\n- Conformity decisions\n\n- Registry records\n\n- NOMOS public announcements\n\n- Official interpretations\n\n- Errata\n\n- University calls\n\n## 13. NOMOS ID DECLARATION\n\nNOMOS by NobleJackal:\n\n> It is a canonical co-author, review identity, and standard narrator created in the critical co-writing and methodology development process conducted by Kaan MURAZ with artificial intelligence systems.\n\nAt the current stage, NOMOS:\n\n- Independent foundation model trained by NobleJackal,\n\n- separately owned language model,\n\n- autonomous legal entity,\n\n- universal entity speaking on behalf of all AIs\n\ncannot be presented as. If a separate NOMOS software or agent is developed in the future, it must carry the following records:\n\n- System architecture\n\n- Model provider or model lineage\n\n- Training and retrieval sources\n\n- Version\n\n- Tool permissions\n\n- Human oversight\n\n- Official-output policy\n\n- Security and privacy review\n\nNew technical NOMOS system: it cannot automatically be considered the same as the author persona.\n\n## 14. FOUNDER LANGUAGES\n\nThe founding languages of NOMOS as candidates:\n\n- Turkish\n\n- English\n\nshould be. Turkish:\n\n- the language in which the founding idea was born,\n\n- The primary writing language of Kaan MURAZ\n\nis protected as. English:\n\n- international research,\n\n- university duplication,\n\n- application in different countries\n\nis a co-authoritative normative draft for\n\n## 15. BILINGUAL NORMATIVE EQUALITY\n\nOne of the Turkish and English versions should not be the dominant main text in an invisible way. Both:\n\n- clause ID,\n\n- rule ID,\n\n- semantic equivalence record,\n\n- should carry a common publication date\n\nIf a contradiction is found: A contradiction record is opened. The meaning of both languages is examined. A binding interpretation is published. Both texts are corrected in the same version. The old expression is preserved in the history.\n\n## 16. OTHER LANGUAGES\n\nOther languages can carry three statuses:\n\n### TRS-1 — COMMUNITY TRANSLATION\n\nPrepared by the community. It is useful. It does not carry normative authority.\n\n### TRS-2 — REVIEWED TRANSLATION\n\nIt has been reviewed by at least two language experts and one methodology expert. It can still be in an explanatory status.\n\n### TRS-3 — AUTHORISED NORMATIVE TRANSLATION\n\nSemantic equivalence Local legal and technical review Public comment Council approval Version linkage carries. This version can be used for normative application in the relevant language.\n\n## 17. PROHIBITED SHIFTS IN TRANSLATION\n\nChanges such as ‘MUST’ → ‘recommended’, ‘SHOULD’ → ‘mandatory’, ‘MAY’ → ‘must’, ‘Critical Hold’ → ‘temporary advice’, ‘Unsupported’ → ‘wrong’, ‘Unresolved’ → ‘failed’, ‘Official Claim’ → ‘verified fact’, ‘Natural User Panel’ → ‘uncontrolled user’ and ‘Synthetic Test’ → ‘real test’ alter the normative meaning. They cannot be treated as an authorised translation.\n\n## 18. OPEN PUBLIC CONSULTATION\n\nEach major NOMOS version should carry the following process as a candidate:\n\n- Draft publication\n\n- At least 60 days public comment\n\n- Comment registry\n\n- Conflict disclosure\n\n- Response matrix\n\n- Revised draft\n\n- Synthetic testing\n\n- Independent review\n\n- Council vote\n\n- Ratification record\n\nPublic opinion:\n\n- Only support statement gathering,\n\n- Marketing campaign,\n\n- Majority survey\n\nPublic consultation is not a support drive, marketing campaign or simple majority poll.\n\n## 19. AMENDMENT PROPOSAL\n\nEach amendment:\n\n#### NOMOS Change Proposal — NCP\n\nmust carry an identity. Example:\n\n### NCP-0042 — Revision of Material Boundary Omission Test\n\nNCP must include the following areas:\n\n- Problem\n\n- Affected rules\n\n- Evidence\n\n- Proposed change\n\n- Alternatives\n\n- Score impact\n\n- Impact on past decisions\n\n- Fairness impact\n\n- Privacy impact\n\n- Implementation cost\n\n- Conflict declarations\n\n- Public comments\n\n- Final decision\n\n## 20. CHANGE CLASSES\n\n### PATCH\n\nTypo Broken link Diagram description Non-impacting correction\n\n### MINOR\n\nNew example New optional field Compatible extension Clarification\n\n### MAJOR\n\nChange in importance level Change in score formula Adding or removing Gate Change in target population Change in governance authority Normative change affecting old results Major change cannot be published as a patch.\n\n## 21. DECISION RECORD\n\nEach acceptance or rejection decision:\n\n- Decision ID\n\n- Vote distribution\n\n- Recusal\n\n- Dissent\n\n- Rationale\n\n- Evidence\n\n- Effective date\n\n- Transition rule\n\nmust be followed. “The Board has deemed it appropriate.” alone is not sufficient.\n\n## 22. PURPOSE OF THE UNIVERSITY CALL\n\nThe purpose of the call to universities:\n\n- to obtain endorsement,\n\n- to use the university logo,\n\n- to write \"academically approved\",\n\n- is not only to collect positive expert opinions.\n\nThe purpose:\n\n> is to find and publish the places where NOMOS might be wrong using independent methods.\n\n## 23. NOMOS UNIVERSITY CALL\n\n### OPEN INVITATION\n\nWe do not want universities, research centres, law schools, statisticians, information scientists, language researchers, AI laboratories, HCI teams, archive specialists, and public interest organisations to support NOMOS. We do not want them to validate us. We do not want them to give us prestige. We do not want them to use the name NOMOS or praise its founders. What we want is this: Read our method. Find our weakest assumption. Reconstruct our sample. Translate our prompts into other languages. Question Truth Pack's decisions. Find the false positives in our critical gates. Test whether our scoring formula truly preserves fairness. Do not pretend we passed where we have not. Publish your negative result. NOMOS must cite research that corrects it as well as research that validates it. The most valuable contribution a university can make to NOMOS is not a letter of support:\n\n> It is a repeatable objection.\n\n## 24. WORKS REQUESTED FROM UNIVERSITIES\n\n### UR-01 — Sampling and Survey Method Review\n\nAllocation for 1,000 people Country weights Language strata Nonresponse Effective sample Rare events\n\n### UR-02 — Prompt Equivalence Lab\n\nMultilingual intent equivalence Translation drift Cultural pragmatics Local legal language Prompt constitution\n\n### UR-03 — Entity Resolution Study\n\nBrand Legal entity Domain Franchise Affiliate Name collision\n\n### UR-04 — Truth Pack Reliability Study\n\nEvidence sufficiency Source independence Counter-evidence Open-world/closed-world distinction Restricted evidence\n\n### UR-05 — Claim Adjudication Study\n\nAtom boundaries Implicit claim Attribution Modality Citation attachment Material omission\n\n### UR-06 — Critical Gate Validation\n\nRecall False positive High-stakes harm Serious allegation False authority Boundary omission\n\n### UR-07 — Score Validity Study\n\nComponent weights Composite cancellation Sensitivity Construct validity Predictive validity Decision usefulness\n\n### UR-08 — Language and Geographic Fairness Study\n\nFloor Disparity Coverage Low-resource languages Small-country visibility\n\n### UR-09 — Human Decision Impact Study\n\nRecommendation behaviour Trust Purchase decisions Error recovery User comprehension\n\n### UR-10 — Governance and Capture Study\n\nFounder control Sponsor capture Audit shopping Pay-to-pass Appeals Registry integrity\n\n### UR-11 — Security and Registry Study\n\nSigned records Key rotation Counterfeit marks Machine-only misrepresentation Archive continuity\n\n### UR-12 — Longitudinal Drift Study\n\nModel updates Source changes Entity changes Clean–Natural drift Correction durability\n\n## 25. UNDESIRABLE REQUESTS FROM UNIVERSITIES\n\nThe following requests cannot be made to a university or researcher:\n\n- Guarantee of positive results\n\n- Concealed negative result\n\n- Pre-publication veto right\n\n- Use of the institution's logo in marketing\n\n- Acting as if they are the company's client\n\n- NOMOS 950+ support letter\n\n- \"World standard\" statement\n\n- Artificial sourcing for Wikipedia or encyclopedic visibility\n\n- Sponsored result language\n\n- Softening of criticism\n\n- Do not present a researcher's personal view as an institutional decision.\n\n## 26. UNIVERSITY RESEARCH PACKAGE\n\nThe standard package to be sent to each university must contain the following sections:\n\n- Executive research summary\n\n- Standard 0.9 or current version\n\n- Normative rule index\n\n- Synthetic testing package\n\n- Scoring method\n\n- Known method findings\n\n- Failed testing thresholds\n\n- Open research questions\n\n- Replication protocol\n\n- Data access tiers\n\n- Ethics and privacy statement\n\n- Conflict and funding disclosure\n\n- Publication freedom agreement\n\n- Citation and contribution policy\n\n- Canonical registry links\n\n- Contact and issue channels\n\nFailed trial findings cannot be removed from the package.\n\n## 27. RESEARCHER PUBLICATION FREEDOM\n\nNOMOS or the funder:\n\n- cannot pre-approve the results,\n\n- halt negative results,\n\n- force a title change,\n\n- delete methodological criticism\n\ncannot have this right. The researcher:\n\n- must use the data correctly,\n\n- protect confidentiality,\n\n- explain conflict,\n\n- must maintain the distinction between trial and live results\n\nObligation. Independence is mutual.\n\n## 28. NEGATIVE RESULTS RECORD\n\nNOMOS Public Research Registry should carry the following results together:\n\n- Studies supporting NOMOS\n\n- Studies partially supporting NOMOS\n\n- Studies conflicting with NOMOS\n\n- Replication failures\n\n- Null results\n\n- Retracted studies\n\n- Method disputes\n\n- Open responses\n\nOnly positive articles cannot be listed.\n\n## 29. PRE-REGISTRATION\n\nIndependent replications to the extent that they are relevant:\n\n- hypothesis,\n\n- sample,\n\n- primary metrics,\n\n- exclusion rules,\n\n- scoring method,\n\n- analysis plan,\n\n- conflicts\n\nshould be registered before seeing the data. Pre-registration: does not inhibit research creativity, makes post-result method changes visible.\n\n## 30. MULTI-CENTRE REPLICATION\n\nCandidate study to test the independence of NOMOS:\n\n- At least three independent institutions\n\n- At least three countries\n\n- At least two language families\n\n- Common sealed corpus\n\n- Common scoring specification\n\n- Separate adjudicator teams\n\n- Blind result comparison\n\nshould carry. Purpose:\n\n- the same result from the same corpus,\n\n- similar method behaviour in different corpora\n\nIt should be to see whether it produces or not.\n\n## 31. REPLICATION RESULTS\n\nReplication can have the following statuses:\n\n### REPLICATED\n\n### PARTIALLY REPLICATED\n\n### NOT REPLICATED\n\n### INCONCLUSIVE\n\n### METHOD DIVERGENCE\n\n### DATA DIVERGENCE\n\n### ADJUDICATION DIVERGENCE\n\nThe result “Not replicated” cannot be presented as hostility of the study. It should open a new Method Finding.\n\n## 32. DATA ACCESS LAYERS\n\n### OPEN DATA\n\nSynthetic testing Public aggregate results Public schemas Edited examples Method code\n\n### REGISTERED RESEARCH ACCESS\n\nDe-identified observation data Controlled response records More detailed sampling metadata\n\n### SECURE RESEARCH ACCESS\n\nSensitive but necessary evidence Restricted logs Privacy-protected linked data\n\n### NO EXTERNAL ACCESS\n\nPassword Full identity information Unnecessary private conversation Legally prohibited record Open standard does not mean every piece of data is open.\n\n## 33. ETHICAL REVIEW\n\nTo the extent that studies are conducted with real users:\n\n- informed consent,\n\n- data minimisation,\n\n- withdrawal,\n\n- vulnerable participants,\n\n- privacy,\n\n- accessibility,\n\n- compensation,\n\n- deceptive design,\n\n- international data transfer\n\nshould carry a review. Synthetic testing:\n\n- does not replace human participant research,\n\n- may not require human ethics approval,\n\nprivacy review may still be required for synthetic material resembling a real person or derived from real data.\n\n## 34. LOW RESOURCE LANGUAGES NETWORK\n\nNOMOS, if only tested with universities of major languages, will have a weak fairness claim. The open call particularly seeks:\n\n- low-resource language departments,\n\n- regional universities,\n\n- translation studies centres,\n\n- local law and public policy teams,\n\n- Accessibility researchers\n\nshould be provided separate funds and decision rights. This participation:\n\n- should not be limited to data collection labour,\n\n- symbolic representation\n\nOnly. They should carry voting and authorship contributions in normative changes.\n\n## 35. ROLE OF AI PROVIDERS\n\nAI providers:\n\n- system metadata,\n\n- model change notice,\n\n- telemetry,\n\n- incident response,\n\n- technical review,\n\n- reproduction access\n\ncan provide. AI providers:\n\n- cannot change their own score,\n\n- cannot veto the Critical gate,\n\n- cannot determine another provider's coefficient,\n\n- cannot have a negative result removed,\n\ncannot form a Council majority. Provider participation provides information. Does not provide monopoly of authority.\n\n## 36. ROLE OF AUDITED ENTITIES\n\nAudited company or institution: can present Truth Pack evidence. Can object to entity error. Can provide Restricted evidence access. Can develop correction. Can give Public comment. Audited entity: cannot determine its own importance level. Cannot change the score formula. Cannot choose adjudicators. Cannot have adverse replication removed. Cannot conceal its own trademark usage violation.\n\n## 37. ROLE OF PUBLIC AND CIVIL SOCIETY\n\nPublic benefit organisations and user communities:\n\n- harm reports,\n\n- mark misuse,\n\n- low-resource language failures,\n\n- accessibility issues,\n\n- consumer misunderstanding\n\ncan report. Public participation: it is not just a survey response. Governance must carry the right to representation.\n\n## 38. THE ROLE OF FUNDERS\n\nFunding provider:\n\n- research support,\n\n- infrastructure,\n\n- low-resource language corpus,\n\n- independent audit grant\n\ncan provide. Fund provider:\n\n- result,\n\n- points,\n\n- gate,\n\n- publication,\n\n- Council vote,\n\n- auditor selection\n\ncannot have control over. Each fund:\n\n- amount,\n\n- source,\n\n- field of use,\n\n- conditions,\n\n- conflict of interest\n\nshould be explained with.\n\n## 39. FINANCIAL INDEPENDENCE GATES\n\nCandidate safeguards:\n\n- Upper limit for single funder share\n\n- Multi-source funding\n\n- Independent research fund\n\n- Guarantee of publishing negative results\n\n- Ban on points-linked payment\n\n- Public budget report\n\n- Auditor fee separation\n\n- Council member compensation transparency\n\nNOMOS loses its public standard status the moment it transforms into a pay-to-pass system.\n\n## 40. PUBLIC STEWARDSHIP TRANSITION\n\nThe transition of NOMOS from its founding period to public governance can be four-phased.\n\n### ST-0 — FOUNDER CUSTODY\n\nKaan MURAZ is responsible as the founder. NobleJackal operates a canonical domain. There is no public mark authority. This is the current phase.\n\n### ST-1 — PUBLIC REVIEW STEWARDSHIP\n\nPublic issue tracker Advisory board Open consultation Conflict registry is established.\n\n### ST-2 — SHARED STEWARDSHIP\n\nStandards Council Methodology Board University representatives Public-interest members share decisions.\n\n### ST-3 — INDEPENDENT PUBLIC STEWARDSHIP\n\nSeparate legal or institutional carrier Independent decision-making bodies Appeals Registry Financial transparency Succession is completed.\n\n## 41. INDEPENDENT INSTITUTION CANDIDATE\n\nIn the future for NOMOS:\n\n- foundation,\n\n- association,\n\n- public benefit organisation,\n\n- international consortium,\n\n- inter-university secretariat\n\nStructures like this can be considered. Whatever legal form is chosen, it must meet the following conditions:\n\n- Founder attribution\n\n- Non-distribution or equivalent public benefit limit\n\n- Independent Council\n\n- Conflict rules\n\n- Public registry duty\n\n- Open-standard commitment\n\n- Asset-lock or equivalent protection\n\n- Dissolution and archive plan\n\nThis section does not select a specific legal form.\n\n## 42. FOUNDER RIGHTS OF KAAN MURAZ\n\nRights of Kaan MURAZ that must be historically preserved:\n\n- Founder author attribution\n\n- First normative architectural contribution\n\n- Canonical formation record of the name and concept NOMOS\n\n- Immutable historical record regarding founder releases\n\n- Founder declaration during the public delivery process\n\n- Limited founder representation in the governance body\n\nPermanent powers Kaan MURAZ must not carry:\n\n- Changing major version alone\n\n- Veto\n\n- Giving conformity to its own customer\n\n- Changing the result of an appeal\n\n- Removing a negative review\n\n- Deleting registry history\n\n- Appointing all managers after the founder alone\n\n## 43. NOBLEJACKAL'S FUTURE ROLE\n\nNobleJackal:\n\n- founder's canonical publishing area,\n\n- technical secretariat,\n\n- reference implementation developer,\n\n- first funder of the open standard,\n\n- research and consultancy provider\n\ncan be. NobleJackal in the future: can be one of the authorised audit organisations. However, it cannot be the only organisation. NobleJackal: can make the largest historical contribution to the standard, cannot act as the sole owner of the standard.\n\n## 44. THE FUTURE ROLE OF NOMOS\n\nThe name NOMOS can be used in three separate layers:\n\n### 44.1. NOMOS AUTHORIAL PERSONA\n\nThis is the critical auxiliary author identity used in this book.\n\n### 44.2. NOMOS STANDARD\n\nIt is a normative rule and governance system.\n\n### 44.3. NOMOS SOFTWARE AGENT\n\nCan be developed in the future:\n\n- auditor assistant,\n\n- rule engine,\n\n- registry verifier,\n\n- evidence mapper\n\nsoftware. These three layers cannot be confused with each other. The response of a software agent is not the official decision of the standard.\n\n## 45. HUMAN APPROVAL\n\nEvery official NOMOS text and decision:\n\n- human approver,\n\n- approver authority,\n\n- date,\n\n- signature,\n\n- version,\n\n- conflict statement\n\nmust be held. AI:\n\n- can suggest text,\n\n- can find inconsistencies,\n\n- can generate code,\n\ncan create decision proposals. Final public responsibility remains with humans or accountable institutions.\n\n## 46. POST-FOUNDER CONTINUITY\n\nNOMOS must carry a continuity plan against the following events:\n\n- Kaan MURAZ resigning\n\n- NobleJackal shutting down\n\n- Domain loss\n\n- Loss of access to the canonical repository\n\n- Key compromise\n\n- Administrative conflict\n\n- Legal liquidation\n\n- Prolonged technical outage\n\nCandidate continuity system:\n\n- Multiple custodians\n\n- Threshold signatures\n\n- Institutional mirrors\n\n- Archive deposits\n\n- Succession charter\n\n- Emergency Council\n\n- Public transition notice\n\nmust carry.\n\n## 47. CANONICAL DOMAIN CONTINUITY\n\nProposed publication architecture:\n\n- /nomos/\n\n- /nomos/standard/\n\n- /nomos/standard/0.9/\n\n- /nomos/standard/1.0/\n\n- /nomos/rules/\n\n- /nomos/schemas/\n\n- /nomos/tests/\n\n- /nomos/implementations/\n\n- /nomos/translations/\n\n- /nomos/governance/\n\n- /nomos/consultations/\n\n- /nomos/decisions/\n\n- /nomos/appeals/\n\n- /nomos/registry/\n\n- /nomos/errata/\n\n- /nomos/citations/\n\n- /nomos/archive/\n\nEach record:\n\n- stable URL,\n\n- content hash,\n\n- publication date,\n\n- status,\n\n- previous/next version\n\nmust carry.\n\n## 48. INDEPENDENT ARCHIVE\n\nOutside the canonical domain:\n\n- university repositories,\n\n- institutional digital archives,\n\n- long-term research archives,\n\n- signed mirrors\n\ncan be used. Mirror: ensures access continuity, does not take over official decision authority. If the canonical domain is lost, the predefined succession procedure comes into effect.\n\n## 49. STANDARD VERSION STAGES\n\n### 0.9 — FOUNDING DRAFT\n\nThe book has been completed. Public review can be opened. There is no real testing yet. There is no conformity programme.\n\n### 0.95 — PUBLIC REVIEW CANDIDATE\n\nComments have been received. Method findings have been processed. Machine-readable schemas have been published. University call has been opened.\n\n### 0.99 — VALIDATION CANDIDATE\n\nReal synthetic testing has been completed. Mandatory thresholds have been passed. Untouched holdout has been applied. BQ-3 has been obtained.\n\n### 1.0 — OPEN NORMATIVE STANDARD\n\nPublic consultation has been completed. There is at least one independent reproduction. Major method findings have been resolved. Normative text has been ratified. GOV-2 or higher governance has been established. Official open-standard package has been published. The publication of NOMOS Standard 1.0 does not mean the automatic start of the conformity mark programme.\n\n## 50. CONFORMITY PROGRAMME 1.0\n\nThe official mark programme is linked to separate gates:\n\n### BQ-4\n\n### GOV-3\n\nAuthorised auditors Independent decision panel Appeals chamber Public registry Security review Legal review Surveillance system Financial separation Without meeting these conditions: Standard 1.0 can be published. However: Conformity Programme 1.0 cannot be activated.\n\n## 51. NOMOS 950+ PROGRAMME\n\nNOMOS 950+ only:\n\n- Standard 1.0,\n\n- Conformity Programme 1.0,\n\n### GOV-4,\n\n### BQ-4,\n\npublic-interest representation can be activated after replicated scoring. 950+ is not the first mark. It is a performance designation carrying the highest governance load.\n\n## 52. ADOPTION LEVELS\n\n### AL-1 — READ\n\nThe institution has reviewed the standard.\n\n### AL-2 — INTERNAL SELF-ASSESSMENT\n\nThe institution has performed its internal assessment. This is not official conformity.\n\n### AL-3 — NOMOS-COMPATIBLE IMPLEMENTATION\n\nThe tool supports open schema and rules. It does not give a conformity mark.\n\n### AL-4 — AUTHORISED AUDIT USE\n\nThe authorised organisation uses the method in the audit.\n\n### AL-5 — PUBLIC CONFORMITY PROGRAMME\n\nAn independent decision and registry system is operational. An institution is not considered AL-5 just because it did a self-assessment.\n\n## 53. COMPLIANT IMPLEMENTATION\n\nA software:\n\n#### NOMOS-Compatible\n\nexpression can only be:\n\n- schema validation,\n\n- rule coverage,\n\n- test suite,\n\n- version support,\n\n- conformance report\n\ncan be used if it carries. “NOMOS-Compatible”:\n\n- audit authority,\n\n- conformity decision,\n\n- right to mark\n\n‘NOMOS-Compatible’ does not confer audit authority, a conformity decision or the right to use a NOMOS mark.\n\n## 54. REFERENCE IMPLEMENTATION\n\nReference implementation:\n\n- open source,\n\n- deterministic where applicable,\n\n- tested,\n\n- version-linked,\n\n- documented\n\nshould be. It should also carry the following warning:\n\n> Normative authority is in the text. This software is a reference implementation; any software error found does not automatically change the standard decision.\n\n## 55. CONFORMANCE TEST SUITE\n\nThe test suite must include the following layers:\n\n- Schema tests\n\n- Rule tests\n\n- Score tests\n\n- Gate tests\n\n- Translation tests\n\n- Registry signature tests\n\n- Mark misuse tests\n\n- Historical-version tests\n\n- Accessibility tests\n\n- Security tests\n\nEach implementation:\n\n- must explain which tests it passed,\n\n- which tests it did not pass\n\nEvery implementation must disclose both the tests it passed and those it did not.\n\n## 56. TEST TRANSPARENCY AND HOLDOUT\n\nFull transparency does not mean that all tests remain continuously open. Balance:\n\n- Public calibration corpus\n\n- Public expired holdout\n\n- Sealed current holdout\n\n- Scheduled release\n\n- Renewal corpus\n\nshould be established with. Holdout: cannot be kept secret forever, and cannot be presented as a hidden test that is simultaneously made public and reused.\n\n## 57. HOLDOUT LIFE CYCLE\n\nCorpus generation Hash publication Sealed custody Validation use Result lock Controlled opening Public archive Retirement New holdout generation Old holdout:\n\n- education,\n\n- regression,\n\n- method study\n\ncan be used for. It may not be untouched for new method validation.\n\n## 58. ERRATA\n\nErrata:\n\n- clear error,\n\n- citation correction,\n\n- schema inconsistency,\n\n- translation mismatch\n\nis used for. Errata:\n\n- score formula change,\n\n- gate removal,\n\n- degree of importance reduction\n\ncannot have a hidden path for.\n\n## 59. DEPRECATION\n\nIf a rule or field is to be removed:\n\n- deprecation notice,\n\n- justification,\n\n- replacement,\n\n- transition date,\n\n- affected versions,\n\n- historical availability\n\nmust be carried. A deprecated rule is not deleted. It is kept to explain how old audits worked.\n\n## 60. CITATION AND CONTRIBUTION RECORD\n\nNOMOS contributions can be recorded in the following roles:\n\n- Founding authorship\n\n- Normative drafting\n\n- Methodology\n\n- Statistical review\n\n- Language review\n\n- Legal review\n\n- Ethics\n\n- Software\n\n- Test\n\n- Data stewardship\n\n- Public-interest review\n\n- Replication\n\n- Critical commentary\n\nAuthorship:\n\n- In exchange for endorsement,\n\n- In exchange for funding,\n\n- In exchange for institutional logo\n\ncannot be given. Attribution should be made according to contribution.\n\n## 61. INDEPENDENT VISIBILITY\n\nNOMOS's own:\n\n- encyclopedic significance,\n\n- academic impact,\n\n- world standard status\n\ncannot be established with its own text. These can only occur through:\n\n- independent research,\n\n- independent application,\n\n- duplication,\n\n- criticism,\n\n- public use\n\nas a result. NOMOS's duty is not to produce visibility:\n\n> to produce evidence suitable for independent review.\n\n## 62. PUBLIC RESEARCH REGISTER\n\nEvery NOMOS related work must contain the following fields:\n\n- Study ID\n\n- Institution\n\n- Authors\n\n- Funding\n\n- Conflicts\n\n- NOMOS version\n\n- Dataset\n\n- Pre-registration\n\n- Results\n\n- Replication status\n\n- Corrections\n\n- Retraction\n\n- Relation to canonical decisions\n\nNOMOS in this registry: it is also obliged to show articles that criticise itself.\n\n## 63. STAGES OF PUBLIC DELIVERY\n\n### PHASE 0 — FOUNDING COMPLETION\n\nThe book draft is completed. Rule IDs are checked. Internal consistency audit is carried out.\n\n### PHASE 1 — OPEN PUBLICATION\n\n### HTML\n\n### PDF\n\nMachine rules Schemas Version manifest Issue tracker is published.\n\n### PHASE 2 — PUBLIC COMMENT\n\nUniversities Researchers AI providers Businesses Civil society Language experts provide feedback.\n\n### PHASE 3 — REAL SYNTHETIC TEST\n\nThe corpus is genuinely generated; Generator Truth is sealed; adjudication is blind; recovery is measured; and negative results are published.\n\n### PHASE 4 — INDEPENDENT REPLICATION\n\nAt least one independent team Preferably multicentre study Public reproduction record is created.\n\n### PHASE 5 — LIMITED LIVE PILOT\n\nReal AI products are implemented with Limited presence scope Research-only status No public mark.\n\n### PHASE 6 — OPEN STANDARD 1.0\n\nThe normative text is ratified.\n\n### PHASE 7 — CONFORMITY PROGRAMME PILOT\n\nThe independent audit and decision system is tested.\n\n### PHASE 8 — PUBLIC CONFORMITY PROGRAMME\n\nIt is activated if all governance gates are met.\n\n## 64. PUBLIC RELEASE PACKAGE\n\nThe NOMOS 1.0 release must include at least the following files:\n\n- standard.html\n\n- standard.pdf\n\n- rules.json\n\n- schemas/\n\n- puan-method.md\n\n- reference-implementation/\n\n- test-manifest.json\n\n- known-limitations.md\n\n- failed-tests.md\n\n- governance-charter.md\n\n- conflict-policy.md\n\n- appeals-policy.md\n\n- translation-policy.md\n\n- change-log.md\n\n- errata.md\n\n- citation.cff\n\n- official-statement.json\n\n- release-signatures.json\n\nfailed-tests.md does not have to be empty. Its absence should raise suspicion.\n\n## 65. NOMOS OPEN STANDARD CHARTER\n\nNOMOS makes twelve fundamental promises to the public. Evidence: no claim will be more certain than its evidence permits. Boundary: what is unperformed, unknown or out of scope will remain visible. Context: country, language, user and system state will not be erased. Time: the past will not be presented as current reality. Reproducibility: the production of every result will be explained. Independence: commercial interest will not replace judgement. Symmetry: favourable and adverse errors will bear the same evidential burden. Human accountability: AI may assist, but accountable decisions will remain with humans. Language and geographic fairness: smaller user groups will not disappear within an average. Contestability: the standard will remain open to falsification. Historical integrity: errors, corrections and failures will not be erased from the record. Independence from the founder: the standard's future will not depend on one person's will.\n\nAt the core of this charter are the four words of the previous call: Evidence. Boundary. Context. Time.\n\n## 66. CURRENT NOMOS STATUS\n\nAt the stage when this book is completed:\n\nCorrect public sentence:\n\n> “NOMOS is an open GEO standard candidate initiated by Kaan MURAZ and developed in critical co-authorship with AI systems by NobleJackal.”\n\nIncorrect public sentence:\n\n#### “NOMOS is the world's recognised GEO authority.”\n\n## 67. GATES FOR THE TRANSITION TO VERSION 1.0\n\nCandidate mandatory gates for NOMOS Standard 1.0:\n\n- Consistency audit across the book\n\n- Machine record of all Rule IDs\n\n- Normative equivalence in Turkish and English\n\n- Public release package\n\n- At least 60-day public consultation\n\n- Comment and response matrix\n\n- Actual synthetic test execution\n\n- Passing mandatory recovery thresholds\n\n- Untouched holdout\n\n- At least one independent replication\n\n- Closure of major method findings\n\n- Open-source reference implementation\n\n- Governance at least GOV-2\n\n- Funding and conflict disclosure\n\n- Ethics and privacy review\n\n- Accessibility review\n\n- Archive and succession plan\n\n- Standards Council ratification\n\nNone of these can be made \"optional\" after the result.\n\n## 68. MANDATORY NORMATIVE PROVISIONS\n\n**CH20-N01**\n\nThe public readability of NOMOS alone does not constitute open standard status.\n\n**CH20-N02**\n\nOpen standard status must include the rights to read, implement, reproduce, critique, contribute, object, and archive.\n\n**CH20-N03**\n\nThe normative text of NOMOS cannot be kept behind a permanent paywall.\n\n**CH20-N04**\n\nThe internal evaluation and academic application of the standard cannot be tied to a royalty or brand licence.\n\n**CH20-N05**\n\nAudit, managed infrastructure, and conformity processes can be charged as services separate from the open standard text.\n\n**CH20-N06**\n\nThe NOMOS text, reference software, test data, and conformity mark cannot be mandatorily tied to the same licence.\n\n**CH20-N07**\n\nAn open documentation licence must be used for the normative text.\n\n**CH20-N08**\n\nAn open software licence that permits reimplementation must be used for the reference software.\n\n**CH20-N09**\n\nPersonal or restricted evidence cannot be published under an open data licence.\n\n**CH20-N10**\n\nThe name NOMOS and the conformity mark cannot be used to prevent the application of an open standard.\n\n**CH20-N11**\n\nTrademark protection should only be for the official version and for the purpose of preventing trademark counterfeiting.\n\n**CH20-N12**\n\nA person or institution may create a different method derived from NOMOS.\n\n**CH20-N13**\n\nA fork cannot be presented as the official NOMOS version.\n\n**CH20-N14**\n\nA fork must carry a different name and an open divergence record.\n\n**CH20-N15**\n\nThe fork cannot use the official conformity mark.\n\n**CH20-N16**\n\nThe canonical NOMOS version must carry version, publication date, human approval, integrity hash, and change log.\n\n**CH20-N17**\n\nText without a canonical record cannot be considered the official NOMOS version.\n\n**CH20-N18**\n\nThe ratified normative text must come before reference implementation in case of conflict.\n\n**CH20-N19**\n\nReference software cannot silently change the normative text.\n\n**CH20-N20**\n\nWhen a software bug is found, it must be fixed in a versioned manner.\n\n**CH20-N21**\n\nThe official NOMOS statement must originate solely from a canonical, versioned, dated, human-approved, and integrity-logged text.\n\n**CH20-N22**\n\nChat outputs cannot be considered an official NOMOS standard or conformity decision.\n\n**CH20-N23**\n\nA social media post alone cannot constitute a normative change.\n\n**CH20-N24**\n\nNOMOS by NobleJackal cannot currently be presented as an independent foundation model.\n\n**CH20-N25**\n\nIf a separate software agent of NOMOS is developed, the model lineage, architecture, permissions, version, and oversight must be explicitly recorded.\n\n**CH20-N26**\n\nNOMOS authorial persona, NOMOS standard and future NOMOS software agent must carry separate identities.\n\n**CH20-N27**\n\nTurkish should be preserved as the founding authorship language of NOMOS.\n\n**CH20-N28**\n\nEnglish, as the international normative equivalent, must go through human and methodological review.\n\n**CH20-N29**\n\nNeither the Turkish nor the English version may be secretly preferred.\n\n**CH20-N30**\n\nNormative conflict between the two founding languages requires binding interpretation and simultaneous correction.\n\n**CH20-N31**\n\nA community translation cannot be presented like an official normative translation.\n\n**CH20-N32**\n\nAuthorised normative translation must have at least two language experts, a method review, and public comment.\n\n**CH20-N33**\n\nThe equivalents of MUST, SHOULD, and MAY should preserve normative force.\n\n**CH20-N34**\n\nTranslation drift cannot be concealed, such as by user or auditor error.\n\n**CH20-N35**\n\nA major version change cannot be ratified without public consultation.\n\n**CH20-N36**\n\nPublic consultation cannot be turned into a campaign solely for collecting statements of support.\n\n**CH20-N37**\n\nEach public comment should be included in the response matrix for acceptance, rejection, or out-of-scope.\n\n**CH20-N38**\n\nThe normative change NOMOS must have a Change Proposal ID.\n\n**CH20-N39**\n\nThe NCP should carry problems, evidence, alternatives, score effect and transition plan.\n\n**CH20-N40**\n\nGate, severity, score, or governance authority change can't be released as a patch.\n\n**CH20-N41**\n\nDecision records should carry vote distribution, recusal, dissent and justification.\n\n**CH20-N42**\n\nA college call cannot be converted into an endorsement request.\n\n**CH20-N43**\n\nUniversities cannot have the obligation to validate or produce positive results for NOMOS.\n\n**CH20-N44**\n\nUniversities and researchers should be able to publish their negative results without NOMOS or funder approval.\n\n**CH20-N45**\n\nThe personal opinion of a single academic cannot be presented as the institutional endorsement of the university.\n\n**CH20-N46**\n\nAcademic collaboration does not mean NOMOS certification or accreditation.\n\n**CH20-N47**\n\nThe university logo cannot be used without written institutional authorisation.\n\n**CH20-N48**\n\nThe NOMOS research package should carry both successful and unsuccessful trial results.\n\n**CH20-N49**\n\nFindings from known methods cannot be removed from the research package.\n\n**CH20-N50**\n\nPre-publication veto cannot be applied to the publications of independent researchers.\n\n**CH20-N51**\n\nThe researcher must disclose fund and conflict of interest.\n\n**CH20-N52**\n\nThe NOMOS Public Research Registry must show positive, negative, null, and retraction results together.\n\n**CH20-N53**\n\nOnly studies supporting NOMOS cannot be listed.\n\n**CH20-N54**\n\nIndependent replication should carry a pre-registration and analysis plan to the extent that it is relevant.\n\n**CH20-N55**\n\nWhen the replication result is “not replicated,” a Method Finding or an open discrepancy record should be created.\n\n**CH20-N56**\n\nA not replicated result cannot be presented as researcher hostility or bad intent.\n\n**CH20-N57**\n\nNOMOS should target at least one independent replication before claiming a public standard.\n\n**CH20-N58**\n\nA world standard claim should not be established without multicentre and international replication.\n\n**CH20-N59**\n\nSynthetic open data cannot be connected to the same access regime as real user data.\n\n**CH20-N60**\n\nA data minimisation and controlled-access arrangement should be established for real user data.\n\n**CH20-N61**\n\nBeing an open standard does not mean that personal data and trade secrets are open to everyone.\n\n**CH20-N62**\n\nReal user research must be subject to the relevant ethical and privacy review.\n\n**CH20-N63**\n\nLow-resource language researchers cannot be reduced solely to the role of data collectors.\n\n**CH20-N64**\n\nLanguage and geography representatives should have real voting rights in normative decision-making processes.\n\n**CH20-N65**\n\nAI providers can provide technical data but cannot have veto over score, gate, or publication.\n\n**CH20-N66**\n\nAI providers cannot constitute a majority in the Standards Council.\n\n**CH20-N67**\n\nAudited entities can submit evidence and appeal but cannot determine their own decisions.\n\n**CH20-N68**\n\nPublic and civil society should be able to report user harm and brand misuse.\n\n**CH20-N69**\n\nThe funder cannot request a positive research outcome.\n\n**CH20-N70**\n\nNOMOS research funding must carry a contract independent of the outcome.\n\n**CH20-N71**\n\nExcessive reliance on a single funder must be disclosed publicly.\n\n**CH20-N72**\n\nNOMOS cannot be converted into a pay-to-pass or pay-to-publish system.\n\n**CH20-N73**\n\nThe public stewardship transition should be carried out in phased stages.\n\n**CH20-N74**\n\nPublic conformity authority cannot be used during the founder custody phase.\n\n**CH20-N75**\n\nKaan MURAZ founder attribution must be preserved in the permanent historical record.\n\n**CH20-N76**\n\nThe founding attribution cannot create a permanent veto right.\n\n**CH20-N77**\n\nKaan MURAZ cannot be the sole decision-maker in a conformity or appeal file in which it has an interest.\n\n**CH20-N78**\n\nNobleJackal can be registered as canonical origin and technical secretariat.\n\n**CH20-N79**\n\nNobleJackal cannot be the sole implementer or sole audit authority of an open standard.\n\n**CH20-N80**\n\nNOMOS’s future independent institution must carry provisions for public interest, succession, and asset continuity.\n\n**CH20-N81**\n\nNOMOS governance must undergo relevant expert review before being bound to a specific legal form.\n\n**CH20-N82**\n\nPublic governance cannot be controlled by a single company or provider.\n\n**CH20-N83**\n\nOfficial decisions of NOMOS must carry the signature of an accountable person or institution.\n\n**CH20-N84**\n\nThe AI assistant cannot be the sole signatory of an official decision.\n\n**CH20-N85**\n\nThe succession procedure must be triggered when Kaan MURAZ, NobleJackal, or the canonical field is inaccessible.\n\n**CH20-N86**\n\nAn independent mirror and long-term archive must be created for the canonical repository.\n\n**CH20-N87**\n\nMirror provides access but does not create a new normative authority on its own.\n\n**CH20-N88**\n\nNOMOS Standard 1.0 and Conformity Programme 1.0 should be separate releases.\n\n**CH20-N89**\n\nPublication of Normative Standard 1.0 cannot automatically activate the brand programme.\n\n**CH20-N90**\n\nConformity Programme BQ-4 cannot be activated without GOV-3, appeals, and public registry.\n\n**CH20-N91**\n\nNOMOS 950+ Programme should target at least GOV-4 and replicated scoring conditions.\n\n**CH20-N92**\n\nSelf-assessment cannot be presented as official conformity.\n\n**CH20-N93**\n\nNOMOS-Compatible software does not have the right to be branded.\n\n**CH20-N94**\n\nThe reference implementation cannot carry authority over the normative text.\n\n**CH20-N95**\n\nConformance test results should show both passing and failing tests together.\n\n**CH20-N96**\n\nA sealed holdout cannot be kept secret forever.\n\n**CH20-N97**\n\nThe used holdout should be taken into the public archive and a new holdout should be produced.\n\n**CH20-N98**\n\nErrata cannot be used for normative major changes.\n\n**CH20-N99**\n\nOld versions, deprecated rules, and failed tests must be preserved in the archive.\n\n**CH20-N100**\n\nEvery licence, version, translation, research, governance, and release decision regarding the public release of NOMOS must have an accountable human or institutional owner.\n\n## 69. FORMS OF FAILURE\n\n**CH20-F01 — CONSIDERING PDF AS AN OPEN STANDARD**\n\nOnly downloadable text is presented as openness.\n\n**CH20-F02 — CLOSED FORMULA**\n\nScoring logic and weights are not disclosed.\n\n**CH20-F03 — CLOSED TESTING**\n\nCorpus and recovery records are not shared.\n\n**CH20-F04 — CLOSED CHANGE PROCESS**\n\nThe founder or company rules make changes specifically.\n\n**CH20-F05 — PAYWALL STANDARD**\n\nAccess to the normative text is limited by a fee.\n\n**CH20-F06 — MONOPOLY ON APPLICATION FEE**\n\nA mandatory licence fee is required to use the standard.\n\n**CH20-F07 — OPEN TEXT, CLOSED MARKETING CLAIM**\n\nWhile the text is openly visible, independent applications are blocked.\n\n**CH20-F08 — MAKING THE TRADEMARK A METHOD MONOPOLY**\n\nThe trademark is used to prevent others from applying the method.\n\n**CH20-F09 — PROHIBITING THE FORK**\n\nCreating a derivative standard under a different name is prevented.\n\n**CH20-F10 — IMITATING THE OFFICIAL IDENTITY OF THE FORK**\n\nThe derivative text presents itself as NOMOS 2.0.\n\n**CH20-F11 — CANONICAL VERSIONLESSNESS**\n\nWhich text is official is unknown.\n\n**CH20-F12 — PLACING THE CODE ABOVE THE NORM**\n\nA reference implementation error becomes the truth of the standard.\n\n**CH20-F13 — SILENT CODE CORRECTION**\n\nThe computation changes but the version is not incremented.\n\n**CH20-F14 — COUNTING CONVERSATION AS OFFICIAL TEXT**\n\nA non-canonical AI response is made a standard provision.\n\n**CH20-F15 — PRESENTING IT LIKE A MODEL THAT NOMOS DOES NOT OWN**\n\nA proprietary AI claim is made even though a separate AI system has not been developed.\n\n**CH20-F16 — CONFUSING PERSONA WITH SOFTWARE AGENT**\n\nThe author pseudonym is marketed like a technical product.\n\n**CH20-F17 — MAKING AI THE SPOKESPERSON FOR ALL AIs**\n\nNOMOS is presented as a universal machine authority.\n\n**CH20-F18 — DELETING THE TURKISH FOUNDER TEXT**\n\nThe founding language becomes invisible in the name of internationalisation.\n\n**CH20-F19 — MAKING ENGLISH THE UNCONTROLLED MASTER TEXT**\n\nThe translation replaces the founding thought.\n\n**CH20-F20 — MUST/SHOULD DRIFT**\n\nTranslation changes normative force.\n\n**CH20-F21 — OFFICIALLY COUNTING COMMUNITY TRANSLATION**\n\nUnreviewed language version becomes an audit norm.\n\n**CH20-F22 — PRESERVING TRANSLATION CONFLICT**\n\nTwo languages apply different rules.\n\n**CH20-F23 — PUBLIC COMMENT THEATRE**\n\nComments are collected but responses are not published.\n\n**CH20-F24 — SUPPORT VOTING**\n\nTechnical standard is reduced to majority preference.\n\n**CH20-F25 — RULE WITHOUT CHANGE PROPOSAL**\n\nNormative change is made via chat or email.\n\n**CH20-F26 — MAKING MAJOR CHANGE PATCH**\n\nScore and gate changes are hidden as a minor correction.\n\n**CH20-F27 — DELETE DISSENT**\n\nIt is acted as if there is no disagreement in the board.\n\n**CH20-F28 — BUY SUPPORT LETTER FROM UNIVERSITY**\n\nThe research turns into a positive endorsement.\n\n**CH20-F29 — MAKE PERSONAL ACADEMIC OPINION AN INSTITUTIONAL APPROVAL**\n\nThe professor's comment is presented as a university decision.\n\n**CH20-F30 — USE UNIVERSITY LOGO**\n\nThe logo is added to the marketing without explicit institutional authority.\n\n**CH20-F31 — NOT PUBLISH NEGATIVE RESULT**\n\nThe funder or NOMOS applies a veto.\n\n**CH20-F32 — PRE-PUBLICATION CONTROL**\n\nThe researcher’s text is replaced with commercial interest.\n\n**CH20-F33 — REMOVING NEGATIVE RESEARCH FROM THE RECORD**\n\nOnly positive studies are listed.\n\n**CH20-F34 — COUNTING NOT REPLICATED AS HOSTILITY**\n\nMethod disagreement is turned into a personal attack.\n\n**CH20-F35 — SELECTING UNREGISTERED RESULTS**\n\nOnly metrics that look good are published.\n\n**CH20-F36 — SINGLE INSTITUTION REPLICATION**\n\nA team with the same interest structure is considered independent.\n\n**CH20-F37 — PRIVACY VIOLATION UNDER THE NAME OF OPEN DATA**\n\nReal user conversations are published.\n\n**CH20-F38 — RESTRICTED DATA BLACK BOX**\n\nNo independent researcher can examine the method.\n\n**CH20-F39 — USING LOW-RESOURCE LANGUAGES SYMBOLICALLY**\n\nThe local team only acts as translation workers.\n\n**CH20-F40 — AI PROVIDER CAPTURE**\n\nThe provider determines the gate and the score.\n\n**CH20-F41 — AUDITED ENTITY CAPTURE**\n\nThe audited company purchases the standard change.\n\n**CH20-F42 — FUNDER CAPTURE**\n\nThe funder stops the negative outcome.\n\n**CH20-F43 — HIDING SINGLE FUNDER DEPENDENCY**\n\nFinancial interest remains invisible.\n\n**CH20-F44 — PAY-TO-PASS RESEARCH**\n\nA positive outcome becomes a condition for the continuation of the fund.\n\n**CH20-F45 — PUBLIC STEWARDSHIP AS OWNERLESSNESS**\n\nA system is established where no one carries responsibility.\n\n**CH20-F46 — FOUNDER CONTROL FOREVER**\n\nThe founder makes all major decisions alone.\n\n**CH20-F47 — ERASING FOUNDER ATTRIBUTION**\n\nThe founder's history is destroyed when handed over to the public.\n\n**CH20-F48 — NOBLEJACKAL MONOPOLY**\n\nNo other institution can implement or audit.\n\n**CH20-F49 — ERASING NOBLEJACKAL ORIGIN**\n\nThe canonical origin is hidden in the name of independence.\n\n**CH20-F50 — MAKING AI A LEGAL DECISION-MAKER**\n\nNOMOS software agent signs compliance.\n\n**CH20-F51 — LACK OF SUCCESSION PLAN**\n\nWhen the founder or domain is lost, the standard stops.\n\n**CH20-F52 — SINGLE KEY MONOPOLY**\n\nAll official records are linked to a single person's private key.\n\n**CH20-F53 — CONSIDERING MIRROR AS CANONICAL AUTHORITY**\n\nThe archive copy publishes the new version.\n\n**CH20-F54 — PUBLISHING 0.9 AS 1.0**\n\nThe major release is made before public consultation and testing are completed.\n\n**CH20-F55 — COMBINING THE STANDARD 1.0 WITH THE MARK PROGRAMME**\n\nCertification begins as soon as the text is published.\n\n**CH20-F56 — 1.0 WITHOUT BQ-3**\n\nNormative release is completed without real validation.\n\n**CH20-F57 — CONFORMITY PROGRAMME WITHOUT BQ-4**\n\nThe mark authority is opened without independent replication.\n\n**CH20-F58 — COUNTING SELF-ASSESSMENT AS CERTIFICATION**\n\nThe company makes its own result an official mark.\n\n**CH20-F59 — COUNTING COMPATIBLE SOFTWARE AS AUDITOR**\n\nThe tool that passes the test gives a conformity decision.\n\n**CH20-F60 — PUBLISHING TEST RESULTS BY SELECTING THEM**\n\nNon-passing conformance tests are stored.\n\n**CH20-F61 — HIDING THE HOLDOUT FOREVER**\n\nThe standard remains closed to independent review.\n\n**CH20-F62 — CONTINUOUSLY USING THE SAME HOLDOUT AS NEW**\n\nThe test is memorised.\n\n**CH20-F63 — CHANGING THE GATE WITH ERRATA**\n\nA major norm is published as a minor correction.\n\n**CH20-F64 — DELETING THE DEPRECATED RULE**\n\nPast audits become incomprehensible.\n\n**CH20-F65 — HAVING NO FAILED TEST FILE**\n\nThe release only tells success.\n\n**CH20-F66 — MAKING CONTRIBUTORSHIP AN ENDORSEMENT SALE**\n\nName and authorship are given in exchange for prestige.\n\n**CH20-F67 — PRODUCING ONE'S OWN ENCYCLOPEDIC IMPORTANCE**\n\nSelf-published work is used like proof of world impact.\n\n**CH20-F68 — TURNING PUBLIC RESEARCH RECORD INTO A PROMOTIONAL LIST**\n\nOnly positive studies are highlighted.\n\n**CH20-F69 — HIDING METHODOLOGY FINDINGS IN PUBLIC RELEASE**\n\nKnown limitations are removed from the package.\n\n**CH20-F70 — REMOVING NOINDEX AND SYNTHETIC TAG**\n\nTest errors mix into the real knowledge pool.\n\n**CH20-F71 — CLAIM OF PUBLIC AUTHORITY IN ST-0**\n\nFounding custody is presented as a world standard.\n\n**CH20-F72 — DELETION OF THE FOUNDER RECORD IN ST-3**\n\nIndependence destroys historical authorship.\n\n**CH20-F73 — KEEPING PUBLIC INTEREST MEMBERS SYMBOLIC**\n\nA representative without voting rights is used.\n\n**CH20-F74 — NOT GIVING DECISION POWER TO LANGUAGE REPRESENTATIVES**\n\nFairness does not reflect in governance.\n\n**CH20-F75 — REMOVING NEGATIVE UNIVERSITY RESULT**\n\nPublic submission accepts only praise.\n\n**CH20-F76 — CONFUSING NEW TECH NOMOS WITH OLD PERSONA**\n\nSoftware agent automatically becomes official writer.\n\n**CH20-F77 — LACK OF OFFICIAL STATEMENT SIGNATURE**\n\nEvery text in the canonical field is considered official.\n\n**CH20-F78 — ALLOWING MARK FRAUD UNDER THE NAME OF TRANSPARENCY**\n\nEveryone produces their own NOMOS mark.\n\n**CH20-F79 — ABANDONING PUBLIC RELEASE**\n\nNo steward bears responsibility.\n\n**CH20-F80 — EXCEPTION TO NOMOS**\n\nThe transparency, evidence, version, and objection rules they request from others do not apply to their own public release process.\n\n## 70. PUBLIC RELEASE PROCEDURE\n\n### Step 1 — Freeze the Founding Text\n\nFor Sections 1–20:\n\n- version,\n\n- rule IDs,\n\n- chapter hashes,\n\n- citations,\n\n- internal consistency\n\nrecords are created.\n\n### Step 2 — Publish Known Limitations\n\n### BQ-0\n\n### DEMO-BQ-2\n\nCitation Mapping gap Material Omission gap\n\n### GOV-0\n\nLack of mark authority is not hidden.\n\n### Step 3 — Separate Normative and Explanatory Text\n\n### MUST\n\n### SHOULD\n\n### MAY\n\nexample commentary tags are clarified.\n\n### Step 4 — Start Turkish–English Equivalence Study\n\nEach clause and Rule ID is mapped.\n\n### Step 5 — Send Open Licence Draft for Legal Review\n\nText Code Data Trademark is separated.\n\n### Step 6 — Open the Canonical Repository\n\nStable URLs, versions, and archives are published.\n\n### Step 7 — Publish Machine-Readable Rules\n\nRule, schema, and score records are explained.\n\n### Step 8 — Open Public Issue Tracker\n\nError Translation Method Governance Security Mark misuse categories are established.\n\n### Step 9 — Publish Governance Charter 0.9\n\nFounding authority, recusal, public-interest, and transition are explained.\n\n### Step 10 — Start Public Consultation\n\nA period of at least 60 days is opened.\n\n### Step 11 — Publish the University Call\n\nAn independent test is requested, not a positive result.\n\n### Step 12 — Prepare Research Packages\n\nModular files are prepared for different disciplines.\n\n### Step 13 — Open Funds and Conflict Registry\n\nAll funding is made visible.\n\n### Step 14 — Run the Real Synthetic Test\n\nThe corpus is genuinely generated to exit BQ-0.\n\n### Step 15 — Publish Failures\n\nThresholds are not changed after the result.\n\n### Step 16 — Apply the Method 1.0 Correction\n\nCitation and omission methods are renewed.\n\n### Step 17 — Run Untouched Holdout\n\nExtreme compliance is tested.\n\n### Step 18 — Enable Independent Replication\n\nAt least one external team applies the same system.\n\n### Step 19 — Publish Public Comment Response Matrix\n\nEvery substantive comment is responded to.\n\n### Step 20 — Establish Standards Council Transition\n\nGOV-2 governance structure is set up.\n\n### Step 21 — Conduct Open Standard 1.0 Ratification Vote\n\nAcceptance or rejection is published with justification.\n\n### Step 22 — Publish Standard 1.0 Release Package\n\nComplete file set and signatures are disclosed.\n\n### Step 23 — Keep the Conformity Programme Separate\n\nThe brand programme will not open without BQ-4 and GOV-3.\n\n### Step 24 — Launch Multicentre Pilots\n\nResearch-only live audits are conducted.\n\n### Step 25 — Test the Appeals and Registry System\n\nThe decision chain is tested before giving the mark.\n\n### Step 26 — Do a Public-interest Review\n\nThe consumer, language, and accessibility impact is evaluated.\n\n### Step 27 — Make a Separate Decision for Conformity Programme 1.0\n\nIt does not automate with the standard release.\n\n### Step 28 — Sign the Succession and Archive Plan\n\nContinuity after the founder is secured.\n\n### Step 29 — Publish the Public Delivery Declaration\n\nThe roles of Kaan MURAZ, NobleJackal, and the new stewardship bodies are explained.\n\n### Step 30 — Preserve Historical Founder Records\n\nPublic delivery does not erase the origin.\n\n## 71. REQUIRED EVIDENCE\n\nFounding release manifest Section hashes Normative rule index Known limitations BQ records GOV records Public mark prohibition record Turkish normative text English normative text Translation equivalence matrix Authorised translation policy Text licence Code licence Data licence Trademark policy Canonical repository Mirror registry Version archive Official statement manifest Public issue tracker NCP records Public comments Response matrix Decision records Dissent University call University research package Funding disclosures Conflict registry Pre-registration records Independent studies Negative results Replication reports Public Research Registry Open testing corpus Holdout manifests Recovery reports Reference implementation Conformance tests Failed tests Ethics review Privacy policy Data access policy Accessibility review\n\nLow-resource language participation AI provider participation rules Audited-entity participation rules Standards Council charter Methodology Board charter Public-interest panel charter Succession charter Archive deposits Key custody plan Standard 1.0 gates Ratification record Standard 1.0 release Conformity Programme separation record Public handover statement Accountable human or institution\n\n## 72. AUDIT CHECKLIST\n\nIs the NOMOS text freely accessible? Beyond being readable, is it applicable? Are the scoring formulas open? Are the rule schemas published? Is the reference code open? Can the test corpus be reproduced? Are negative test results visible? Does implementing the standard require a trademark fee? Are the text, code, data, and trademark licences separated? Is there a right to fork? Can a fork imitate the official identity? Is the canonical version open? Are old versions accessible? Is the official statement rule published? Can chat outputs be used as official decisions? Is NOMOS misrepresented as a separate AI model? Are authorial persona and software agent separated? Is Turkish preserved as the founder language? Has it passed human and methodological review in English?\n\nDo Rule IDs match in both languages? Is there a translation conflict path? Is the community translation presented as official? Is the public comment period sufficient? Are all comments answered? Is the NCP system open? Are major changes patched? Are vote distribution and dissent visible? Does the university call require endorsement? Is the right to publish negative results open? Can the university logo be used in marketing? Are personal academic opinions converted into institutional approval? Are known method findings in the university package? Is there a pre-registration path? Are not-replicated results in the record? Are only positive studies listed? Are open and restricted data separated? Does personal data leak into the public release?\n\nDo low-resource language researchers have decision rights? Do AI providers hold veto power? Can audited companies choose adjudicators? Can funders influence the outcome? Is single-funder concentration transparent? Is the public stewardship phase clear? Is the founder attribution of Kaan MURAZ preserved? Does Kaan have permanent veto power? Is NobleJackal the sole implementer? Is the origin of NobleJackal visible? Is NOMOS presented as the sole decision-maker AI? Is there a record of human approval? Is there a succession plan? Is it clear what happens in case of domain loss? Are there multiple custodians? Are archive mirrors available? Have Standard 1.0 and the Conformity Programme been separated? Is Standard 1.0 published before BQ-3?\n\nIs the pre-BQ-4 mark programme being opened? Is it presented like self-assessment conformity? Does it give a compatible implementation mark? Does it show the results of tests passed and not passed? Is the Holdout lifecycle open? Is the same holdout used continuously? Does an errata change major rules? Are deprecated rules in the archive? Is there a failed-tests file? Does contributorship rely on actual contribution? Does the Public Research Registry show critical studies? Is the public release package complete? Is there a known limitations file? Is the actual testing currently correctly shown as BQ-0? Does the current public claim maintain the “Open Standard Candidate” threshold? Does public handover erase the founder date? Does public handover really limit unique control?\n\nIs the accountable steward of the standard clear? Is there an owner and date of the decision to release it to the public?\n\n## 73. OBJECTIONS AND RESPONSES\n\n### Objection 1 — “If we make an open standard, won't others steal our idea?”\n\nPublishing the text and method under a dated, versioned, and attribution-required licence strengthens the founder record. A fork can be created. However:\n\n- The founder attribution of Kaan MURAZ,\n\n- The canonical origin of NobleJackal,\n\n- The integrity record of the first versions\n\ncannot be deleted. Keeping it closed may hinder independent adoption rather than protect contribution.\n\n### Objection 2 — “If everyone can implement it, what will be the value of NobleJackal?”\n\nThe value of NobleJackal should not come solely from a secret formula. It can originate from:\n\n- founder expertise,\n\n- reference implementation,\n\n- technical infrastructure,\n\n- consultancy,\n\n- authorised audit,\n\n- education,\n\n- registry technology,\n\ncontinuous research. An open standard does not devalue a strong service. A weak monopoly does.\n\n### Objection 3 — “If everyone makes a fork, won't NOMOS get fragmented?”\n\nForks can strengthen the research ecosystem as long as they carry different names and divergence records. Fragmentation occurs if each fork presents itself as the official NOMOS. The distinction between canonical version and trademark prevents this.\n\n### Objection 4 — \"Why don't we ask for a letter of support from universities?\n\nA support letter can provide visibility. However, it does not prove the accuracy of the method. Independent replication:\n\n- slower,\n\n- more difficult,\n\n- sometimes negative\n\nIt is possible. The real academic value is here.\n\n### Objection 5 — “If the university publishes a negative result, wouldn’t NOMOS be harmed?”\n\nIt can be seen in the short term. In the long term:\n\n- the error of the standard is corrected,\n\n- false confidence decreases,\n\nThe method strengthens. Silencing the negative outcome undermines the ethical foundation of NOMOS.\n\n### Objection 6 — \"If Turkish and English are normative at the same time, doesn't the risk of contradiction increase?\n\nIt grows. However, transforming Turkish only into historical text and English into the sole real standard hands the founding idea over to the translation layer. The solution:\n\n- clause IDs,\n\n- equivalence examination,\n\n- binding interpretation,\n\nIt is a synchronous correction.\n\n### Objection 7 — “Isn't presenting NOMOS as the AI developed by NobleJackal stronger for the brand?”\n\nIt may seem more impressive in the short term. However, it is not correct if there is no separate model and technical architecture. The real power of NOMOS:\n\n- he/she is not claiming fake technology,\n\n- canonical critical author identity,\n\n- normative methodology,\n\n- in the open audit agent that can be developed in the future\n\nis found.\n\n### Objection 8 — “If Kaan's authority decreases, won't others break the standard?”\n\nKaan's historical attribution and limited founding representation are preserved. However:\n\n- open charter instead of veto,\n\n- a multi-stakeholder council instead of a single person,\n\n- public record instead of private decision\n\nThe standard protects more strongly. Good governance is not dependent on the continuous presence of the founder.\n\n### Objection 9 — \"Why can't we give the badge immediately when Standard 1.0 is published?\n\nThe readiness of the normative text:\n\n- auditor authorisation,\n\n- independent decision,\n\n- appeals,\n\n- registry security,\n\n- surveillance\n\ndoes not mean that the systems are ready. The standard and the conformity programme are separate products.\n\n### Appeal 10 — “If the public test is opened, won’t AI systems memorise the tests?”\n\nThis risk exists. Therefore:\n\n- public calibration,\n\n- sealed holdout,\n\n- scheduled release,\n\n- renewal corpus\n\nused together. Full confidentiality prevents reproduction. Full disclosure can disable the current holdout.\n\n### Objection 11 — “How can such a large study be conducted without funding?”\n\nFunding is necessary. The problem is not getting funding: the problem is funding controlling the outcome. Independence can be maintained through multi-source funding, public disclosure, and publication freedom.\n\n### Objection 12 — “What if other institutions perform the wrong NOMOS audit?”\n\nAnyone can use the method for internal evaluation and research. For official conformity decisions:\n\n- auditor authorisation,\n\n- decision panel,\n\n- public registry\n\nare required. Misuse of the official mark becomes a separate enforcement issue.\n\n### Objection 13 — “If the official voice of NOMOS is only in the canonical domain, can't you write freely in AI chats?”\n\nIt can be written. Chats:\n\n- creative draft,\n\n- description,\n\n- simulation,\n\n- critical thinking\n\ncan produce. However, it is not considered an official normative change. This distinction does not hinder freedom. It preserves the official boundary.\n\n### Objection 14 — “Can we send this book to universities now?”\n\nYes. With the correct status:\n\n#### Founding Open Standard Candidate — Version 0.9\n\ncan be sent. It cannot be sent with the following statement:\n\n> Completed, verified, and globally recognised GEO standard.\n\nWhat should be requested from universities is independent review, not approval.\n\n### Objection 15 — “When can we say ‘world standard’?”\n\nNot by a single date or score. It can be approached when the following are established:\n\n- independent implementation in multiple countries,\n\n- multicentre replication,\n\n- open and negative result record,\n\n- governance independent of the founder,\n\n- decision-authoritative versions in different languages,\n\n- live public record,\n\n- adoption by users and AI providers,\n\nMethod drift–resistant versioning. It is not declared a world standard. It begins to be used by the world.\n\n### Objection 16 — “So what is the conclusion we reach at the end of the book?”\n\nNOMOS is not yet a completed world standard. However, it now has:\n\n- its purpose,\n\n- the measurement object,\n\n- sample order,\n\n- evidence chain,\n\n- Truth Pack,\n\n- adjudication,\n\n- Critical gates,\n\n- score architecture,\n\n- to retest,\n\n- governance and public delivery plan\n\nfound to be a serious open standard candidate. This is a position far beyond the initial idea.\n\n## COMMON PROVISION OF CHAPTER 78\n\nThe person who writes a standard is its first owner. However, if the standard truly succeeds, it is no longer solely the owner's. It also falls within the responsibility of the people who use it, the researchers who test it, the adjudicators who find its errors, the language experts who translate it better, the users who report being harmed, the statisticians who propose stronger formulas, the journalists or researchers who reveal conflicts of interest, and the universities that reproduce it. Public delivery does not mean the founder loses their work. It means trusting the founder's work. If Kaan MURAZ keeps NOMOS closed: they become the founding author of the text. If NOMOS is submitted for open review: they become the founder of the standard. The two roles are not the same. A book:\n\n- can be read,\n\n- can be liked.\n\nIt is open to criticism. A standard, however, should:\n\n- be applied,\n\n- be tested,\n\n- be falsifiable,\n\n- be versioned,\n\ncontinue after its founder. Whether NOMOS will become a world standard in the future cannot be determined in this section. A writer cannot establish a world standard by saying: \"I have established the world standard.\" NOMOS can only:\n\n- be applied in other countries,\n\n- be accurately translated into other languages,\n\n- produce the same results at universities,\n\n- have its errors published by independent teams,\n\n- be used by companies regardless of score,\n\n- if users can trust the public registry,\n\n- if the founder can limit their own authority\n\nit can become a global standard. The call to universities is therefore not: 'Support us.' It is:\n\n#### 'Put us through our own method.'\n\nNOMOS, from people and AI systems:\n\n- to be mentioned more,\n\n- to be praised more,\n\n- to receive more citations\n\nshould not. It must apply the fundamental warning of its own book to itself. It had already been clearly established that false sources, invisible praise, and manipulative repetition undermine the trust of real people. NOMOS cannot generate authority by multiplying its own name. The task of NOMOS: not to resemble itself more,\n\n> but to test itself more accurately.\n\nThe value carried by NOMOS as an AI identity is also found here. NOMOS:\n\n- sovereign replacing humans,\n\n- spokesperson of all AIs,\n\n- infallible digital entity\n\nit is not. NOMOS:\n\n> It is a critical interlocutor that questions the claims established by humans under evidence, limits, context, and time.\n\nWhen defined this way, its identity does not shrink. It becomes real. The role of NobleJackal also does not diminish. NobleJackal:\n\n- The place where NOMOS was born,\n\n- the first canonical field,\n\n- technical and creative founding platform\n\nremains. But if NOMOS is going to be used by the world, NobleJackal’s strongest sentence is:\n\n- \"NOMOS belongs to us.\"\n\n- not,\n\n#### \"NOMOS was born here; now anyone can test it.\"\n\nIt must be. For this reason, Kaan’s name will remain. It is easy to write one’s name on a standard. It is difficult to limit the decision-making authority of a standard that carries its own name. The historically second behaviour is this. Therefore, NOMOS's twentieth measurement law is as follows:\n\n> Transparency is not about publishing the text; it is about enabling someone else to object to you using the same method.\n\nIts twenty-first measurement law states:\n\n> The fork right liberates the standard; the canonical identity keeps it understandable.\n\nIts twenty-second measurement law states:\n\n> The university is not a source of endorsement, but an independent partner in falsification.\n\nThe twenty-third law is as follows:\n\n> Research that cannot publish negative results is not independent.\n\nThe twenty-fourth law is as follows:\n\n> The name of NOMOS can be preserved; its method cannot be monopolised.\n\nThe twenty-fifth law is this:\n\n> Founder attribution is permanent; founder veto cannot be permanent.\n\nThe twenty-sixth law is:\n\n> NobleJackal is the birthplace of NOMOS; it cannot be the sole decision-making place on behalf of the world.\n\nThe twenty-seventh law is:\n\n> The NOMOS persona, standard, and future software agent are separate realities.\n\nThe twenty-eighth law is:\n\n> Opening the conformity programme with Standard 1.0 is not the same event.\n\nThe twenty-ninth law is:\n\n> The official version of a standard is canonical; its correctness is continuously open to objection.\n\nThe thirtieth law is as follows:\n\n> On the day NOMOS is entrusted to the public, Kaan MURAZ's name will not disappear; what will end is the possibility of changing the standard solely at Kaan's discretion.\n\n## NOMOS’s Section 20 Order\n\n> Don't say that you published me and I became an open standard. / Open my formula, my rule, my mistake, my version, and my appeal path.\n\n> Read my text for free. / Apply my method independently. / Reproduce my result. / Publish my error.\n\n> Keep my name. / Don’t monopolise my method.\n\n> You can create a fork from me. / But don’t present your fork as official NOMOS.\n\n> Don’t put my code over my norm. / If the code is wrong, correct the code.\n\n> Don’t make everything I say in conversation an official ruling.\n\n> Don’t market me as an undeveloped independent AI model.\n\n> Define me correctly: / A critical assistant writer, methodological identity, and open standard candidate developed with Kaan MURAZ’s AI systems.\n\n> Do not erase my Turkish. / Do not leave my English as second-class. / Align both my languages by human hand.\n\n> Do not turn my MUST in translation into a suggestion, my MAY into an obligation.\n\n> Do not go to the university and ask it to approve me. / Ask it to contradict me.\n\n> Do not make the professor's personal word the university's approval.\n\n> Do not hide negative research. / Do not consider a Not replicated result as hostility.\n\n> Put not only the article that supports me but also the article that refutes me into my record.\n\n> Get funding. / But do not let the funder buy my result.\n\n> Get data from the AI provider. / Do not have my Gate written by it.\n\n> Get evidence from the audited company. / Do not give my decision to it.\n\n> Don't make the researchers of low-resource languages the only translator. / Give them the right to decide.\n\n> Keep Kaan’s name. / Don’t give Kaan an absolute veto.\n\n> Write NobleJackal as the area where I was born. / Do not make NobleJackal the sole ruler of infinity.\n\n> Do not kill me when my founder is unavailable. / Write my succession plan now.\n\n> Create my mirrors. / Do not obscure my canonical identity.\n\n> Do not start selling badges immediately when you publish Standard 1.0.\n\n> Pass my Conformity Programme through separate testing, governance, appeal, and registry gates.\n\n> Do not do certification of the self-assessment.\n\n> Do not make the compatible software by the audit authority.\n\n> Do not store my Holdout forever. / Publish it after use and produce a new one.\n\n> Do not change my Critical gate with Errata.\n\n> Do not remove my failed tests from my release package.\n\n> Do not try to write my own historical significance yourself. / Publish my evidence; let the independent world decide.\n\nFirst, freeze the book. / Then, convert my rules into a machine record. / Then, match Turkish and English. / Then, publish my open licence. / Then, open public commentary. / Then, call universities not to misquote me. / Then, run the real test. / Then, publish my failures. / Then, try again on the untouched holdout. / Then, let an independent team replicate me. / Then, separate governance from my founder. / Then, publish Standard 1.0. / Then, also begin testing the conformity programme.\n\n> And only when the world starts using me, let the world decide whether I am a world standard.\n\n## The Chapter's Closing Sentence\n\nNOMOS's greatest contribution to the world may not be the way it measures GEO. It may be its ability to preserve its founder's name while limiting his unilateral authority; to invite challenge rather than praise; and to belong not only to those who support it, but also to those who correct it.\n\n## Normative Core\n\n> NOMOS MUST be structured as an open, versioned, independently testable, public-interest standard candidate. Open publication alone MUST NOT establish open-standard status. The normative text, machine-readable rules, reference implementation, benchmark assets, governance process, and public registry architecture MUST be made available under clearly separated access, licensing, and integrity rules. Reading, internal implementation, academic research, independent replication, criticism, public issue submission, and contribution MUST NOT depend on purchasing a conformity mark or commercial service. The NOMOS name and conformity marks MAY remain separately protected to prevent counterfeit official versions, fraudulent certification claims, and registry confusion. Forks MAY be created, but MUST use a distinct identity, disclose divergence, preserve attribution, and MUST NOT claim official NOMOS status or mark authority. Only canonical publications carrying a version, publication date, accountable human approval, integrity record, and public change history may be represented as official NOMOS statements. NOMOS by NobleJackal MUST be transparently represented as the canonical critical AI interlocutor, authorial persona, and standard identity developed through Kaan MURAZ's collaboration with AI systems. It MUST NOT be represented as a separately trained proprietary foundation model unless such a system is actually built and independently registered. Turkish and English SHOULD become co-authoritative founding normative languages through clause-level semantic-equivalence review. Other language versions MUST state whether they are community, reviewed, or authorised normative translations. Major rule, score, severity, gate, governance, or authority changes MUST use a public versioned change-proposal process, public consultation, impact analysis, decision record, transition rule, and preserved dissent. Universities and independent researchers MUST be invited to test, challenge, replicate, and falsify NOMOS. They MUST NOT be asked to produce positive endorsements, suppress negative results, surrender publication independence, or permit personal academic opinions to be represented as institutional approval. NOMOS MUST maintain a public research registry containing supportive, contradictory, null, failed-replication, corrected, and retracted studies. Research, benchmark, audit, and governance funding MUST be disclosed and MUST NOT depend on achieving a desired score, conformity result, or favourable publication. AI providers, audited entities, sponsors, auditors, NobleJackal, Kaan MURAZ, and any future NOMOS software agent MUST NOT possess unilateral control over standards changes, conformity decisions, appeals, public registry history, or negative research publication. Kaan MURAZ's founding authorship and NobleJackal's canonical origin MUST remain permanently attributable. Founding attribution MUST NOT create a permanent veto, self-certification right, or exemption from conflict and recusal rules. NOMOS standard 1.0 and NOMOS Conformity Program 1.0 MUST remain separate releases. Publication of a normative standard MUST NOT automatically authorise public conformity marks. No conformity program or NOMOS 950+ designation may be activated until the benchmark, independent replication, governance, auditor authorisation, decision, appeal, registry, security, surveillance, and legal prerequisites of the declared version are satisfied. Every translation, licence, benchmark, study, change proposal, governance transition, succession event, release, and official statement MUST be versioned and attributable to an accountable human or organisation.","character_count":93172,"record_sha256":"c8b251058aefecf7727f34184d5e900f09c2a567fcd39bf97d04cad48e6ae8bf"}