150
followers ·
2 following None yet
posted an update 1 day ago ✅ Article highlight: *Adversaries, Data Poisoning, and Incentive Governance for Training Worlds* (art-60-171, v0.1)
TL;DR:
This article argues that training worlds become adversarial markets.
If gameplay data trains agents, players, UGC authors, operators, and supply-chain actors will try to shape the data. If labels and rewards shape what gets learned, then labels and rewards are governance surfaces too. 171 turns data poisoning and incentive gaming into receipted lifecycles.
Read:
https://huggingface.co/datasets/kanaria007/agi-structural-intelligence-protocols/blob/main/article/60-supplements/art-60-171-adversaries-data-poisoning-and-incentive-governance-for-training-worlds.md
Why it matters:
• makes “training set T is admissible for run R” a governed claim
• treats poisoning as a caseable process, not a vague abuse report
• fails closed when monitoring is unhealthy or detector drift is detected
• treats labels, rewards, collusion, and sybil pressure as governance problems
• connects data integrity to courts, appeals, and bounded publication
What’s inside:
• training substrate governance contracts
• adversary taxonomy for players, UGC, operators, and supply-chain actors
• quarantine → adjudication → inclusion / exclusion pipeline
• monitoring SLOs, monitor health receipts, and detector drift incidents
• label economy contracts and reward distribution receipts
• anti-sybil and collusion monitoring
• admissibility verdict receipts for deciding what may train the next run
Key idea:
Do not say:
*“we filtered poisoned data.”*
Say:
*“this substrate was admitted under this governance contract, adversary taxonomy, monitoring SLO, quarantine/adjudication trail, label economy, reward policy, and admissibility verdict.”*
Data and rewards are governance with receipts.
posted an update 3 days ago ✅ Article highlight: *Performance Governance for World-Scale Autonomy* (art-60-166, v0.1)
TL;DR:
This article argues that performance is not just an engineering concern. It is a governance surface.
World-scale autonomy fails when NPC cognition saturates compute, latency spikes, queues grow, and operators quietly change rules to keep the world alive. 166 turns “playable under load” into a contract: pinned SLOs, budget enforcement, staged degradation, safe-mode regimes, and receipts.
Read:
https://huggingface.co/datasets/kanaria007/agi-structural-intelligence-protocols/blob/main/article/60-supplements/art-60-166-performance-governance-for-world-scale-autonomy.md
Why it matters:
• connects NPC resource budgets to real SLOs and runtime enforcement
• treats high-end NPC cognition as burstable, not always-on
• makes degradation a governed decision instead of panic ops
• keeps safe-mode NPC and safe-mode economy playable without rewriting history
• prevents “performance fix” from becoming an unpublished reality change
What’s inside:
• a *performance governance contract* for staying playable under load
• SLO observability for tick lag, commit latency, receipt backlog, and crash-free rate
• runtime budget manager profiles and budget enforcement receipts
• a degradation ladder: GREEN → YELLOW → ORANGE → RED
• safe-mode policies for NPCs and economy
• playability invariants that must survive even under RED conditions
Key idea:
Do not say:
*“the world still runs under load.”*
Say:
*“this world operated under this performance contract, this SLO profile, this budget manager, this degradation policy, and these receipts proving what changed and what remained invariant.”*
Performance is governance with receipts.
posted an update 5 days ago ✅ Article highlight: *NPC Society Safety: Bounded Autonomy at Scale* (art-60-162, v0.1)
TL;DR:
This article argues that high-autonomy NPC societies need governance, not trust.
Once NPCs can trade, persuade, coordinate, punish, restrict access, or move resources, they stop being “scripts” and become institution-shaped actors. Their actions must be bounded by explicit capability envelopes, resource budgets, policy envelopes, monitoring, safe-mode, and appeal paths.
Read:
https://huggingface.co/datasets/kanaria007/agi-structural-intelligence-protocols/blob/main/article/60-supplements/art-60-162-npc-society-safety.md
Why it matters:
• makes NPC autonomy legible before it affects players, markets, or world state
• prevents runaway coordination, resource extraction, soft coercion, and authority confusion
• treats persuasion and communication as governed actions, not harmless flavor text
• links NPC-caused harm to incident handling and appealable due process
What’s inside:
• *NPC agency profiles* that define allowed and denied capabilities
• *resource budget profiles* with hard ceilings and spend receipts
• *NPC policy envelopes* for rules, disclosure, influence limits, fairness, and escalation
• materiality profiles for deciding which NPC acts require stronger governance
• 148-anchored monitoring for runaway spend, disparate treatment, inducement, and extraction loops
• safe-mode policies that narrow autonomy when monitoring is inconclusive
• 160-compatible incident and recourse paths when NPC actions cause harm or disputes
Key idea:
Do not say:
*“the NPC society is autonomous.”*
Say:
*“these NPCs may act only within these agency profiles, budgets, policy envelopes, monitoring receipts, safe-mode rules, and appeal paths.”*
NPC autonomy doesn’t scale by trust.
It scales by bounds, budgets, and receipts.
View all activity None yet
view post ✅ Article highlight: *Adversaries, Data Poisoning, and Incentive Governance for Training Worlds* (art-60-171, v0.1) TL;DR: This article argues that training worlds become adversarial markets. If gameplay data trains agents, players, UGC authors, operators, and supply-chain actors will try to shape the data. If labels and rewards shape what gets learned, then labels and rewards are governance surfaces too. 171 turns data poisoning and incentive gaming into receipted lifecycles. Read: kanaria007/agi-structural-intelligence-protocols Why it matters: • makes “training set T is admissible for run R” a governed claim • treats poisoning as a caseable process, not a vague abuse report • fails closed when monitoring is unhealthy or detector drift is detected • treats labels, rewards, collusion, and sybil pressure as governance problems • connects data integrity to courts, appeals, and bounded publication What’s inside: • training substrate governance contracts • adversary taxonomy for players, UGC, operators, and supply-chain actors • quarantine → adjudication → inclusion / exclusion pipeline • monitoring SLOs, monitor health receipts, and detector drift incidents • label economy contracts and reward distribution receipts • anti-sybil and collusion monitoring • admissibility verdict receipts for deciding what may train the next run Key idea: Do not say: *“we filtered poisoned data.”* Say: *“this substrate was admitted under this governance contract, adversary taxonomy, monitoring SLO, quarantine/adjudication trail, label economy, reward policy, and admissibility verdict.”* Data and rewards are governance with receipts. See translation
CityOS Under SI-Core: A Worked Example Across All Invariants
