# ARXIV_METADATA.yaml # Metadata for arXiv-style submission of the SZL Ouroboros Thesis v18. # Compatible with Hugging Face Papers upload schema. # Generated: 2026-05-28 # Doctrine: v6 — all counts verifiable against cited sources. title: > The Ouroboros Substrate: a Governance-Mathematical Foundation for Verifiable Agentic AI authors: - name: "Lutar, Stephen P" orcid: "0009-0001-0110-4173" affiliation: "SZL Holdings" email: "stephen@szlholdings.com" affiliations: - id: szl name: "SZL Holdings" url: "https://github.com/szl-holdings" date: "2026-05-28" # --- Abstract (250-300 words) --- abstract: | We present the Ouroboros Substrate (SZL Holdings, versions 14 through 18), a formally verified execution environment for agentic AI in which every governance invariant is machine-checked by the Lean 4 kernel against Mathlib v4.13.0. The central mathematical object is the Lambda-axis score, a nine-dimensional product-of-receipts governance scalar defined by the Lutar Calculus and bounded by the axiom system A1-A15. Axioms A1-A4 (monotonicity, positive homogeneity, Egyptian-fraction normalisation, max bound) characterise Lambda uniquely. Axiom A15 is the SHA-256 collision-resistance assumption, explicitly flagged as a cryptographic open problem under NIST FIPS 180-4 and not derivable from Mathlib. The v18.0 thesis (206 pages, DOI 10.5281/zenodo.20434276) contains 72 named theorem environments and 4 corollaries across eight chapters. Of the 31 theorems in Chapter 2 and 16 in Chapter 6, the Lean Czar audit (Chapter 7) certifies: 17 lake-verified with no sorry and no new axiom; 22 skeleton-pending with honest sorry markers; 1 open-problem- axiomatic (SBOM Lambda-chain total order, conditional on A15); and 1 pure metatheorem (Lean kernel soundness). A separate catalogue of 16 TH-V18 stubs (TH-V18-01 through TH-V18-16) builds clean under lake build with 13 axiom declarations and 8 in-code sorry tactics across the production tree, verified by grep at the chapter-bind tag 2026-05-28. The 32-module Ouroboros runtime (DOI 10.5281/zenodo.20434308) executes 934+ inline assertions (exit code 0) and emits SCITT-compatible, SHA-256-chained governance receipts at every agent action. An OpenTelemetry SEMCONV extension surfaces Lambda scores as first-class telemetry. Domain grafts span seven observability vendors, four cybersecurity platforms, graph neural networks, sparse self-attention, agentic IDE governance, and sovereign-AI provenance. All code, proofs, and data publicly archived via Zenodo DOIs [10.5281/zenodo.19944926, 10.5281/zenodo.20424992, 10.5281/zenodo.20424995, 10.5281/zenodo.20424996, 10.5281/zenodo.20431181, 10.5281/zenodo.20434276, 10.5281/zenodo.20434308]. # --- Classification --- msc_primary: "68Q60" # Specification and verification (program logics, model checking) msc_secondary: - "03B70" # Logic in computer science - "68T05" # Learning and adaptive systems (AI/ML) - "94A60" # Cryptography arxiv_categories: primary: "cs.LO" # Logic in Computer Science cross_list: - "cs.LG" # Machine Learning - "cs.SE" # Software Engineering keywords: - "agentic AI governance" - "Lambda-axis calculus" - "Lean 4 formal verification" - "Mathlib v4.13.0" - "dual-witness receipts" - "Schur concavity" - "PAC-Bayes bounds" - "Doctrine v6" - "SCITT receipt chain" - "sovereign AI" - "NIST FIPS 180-4" - "OpenTelemetry" # --- DOI provenance (7 Zenodo records, all HTTP 200 as of 2026-05-28) --- dois: - id: "10.5281/zenodo.19944926" label: "Concept DOI (rolling latest -> v18.0)" url: "https://doi.org/10.5281/zenodo.19944926" - id: "10.5281/zenodo.20424992" label: "Ouroboros Thesis v14 — Lutar Calculus / HUKLLA / DPI" url: "https://doi.org/10.5281/zenodo.20424992" - id: "10.5281/zenodo.20424995" label: "Ouroboros Thesis v15 — Knot Calculus / Catoni PAC-Bayes / Khipu DAG" url: "https://doi.org/10.5281/zenodo.20424995" - id: "10.5281/zenodo.20424996" label: "Ouroboros Thesis v16 — Feynman Path-Integral / Gates Hamming [8,4,4]" url: "https://doi.org/10.5281/zenodo.20424996" - id: "10.5281/zenodo.20431181" label: "Ouroboros Thesis v17 — Wheeler / Shannon / QEC / Matched-Filter" url: "https://doi.org/10.5281/zenodo.20431181" - id: "10.5281/zenodo.20434276" label: "Ouroboros Thesis v18.0 — Multi-track Substrate Expansion (this paper)" url: "https://doi.org/10.5281/zenodo.20434276" - id: "10.5281/zenodo.20434308" label: "Lutar v18.0.0 — Formal Lean 4 Governance Kernel (software)" url: "https://doi.org/10.5281/zenodo.20434308" # --- GitHub repositories --- github_repos: - url: "https://github.com/szl-holdings/lutar-lean" description: "Lean 4 proof library — lake build, Mathlib v4.13.0, 13 axiom declarations" - url: "https://github.com/szl-holdings/ouroboros" description: "Python runtime — 32-module registry, 934+ inline assertions, Apache-2.0" - url: "https://github.com/szl-holdings/szl-trust" description: "Public trust portal — CPS receipts, mocked:false, deterministic replay" - url: "https://github.com/szl-holdings/szl-cookbook" description: "Engineering skills for Anthropic MCP pattern — 9 documented skills" - url: "https://github.com/szl-holdings" description: "SZL Holdings organisation root" # --- License --- license: "CC-BY-4.0" # --- Honest caveats (Doctrine v6 required) --- honest_caveats: - "Axiom A15 (sha256_collision_resistant) is a cryptographic open problem; SBOM Lambda-chain total-order theorem (Ch. 2, Thm 7) is conditional on it." - "16 of 47 primary-chapter theorems (all Chapter 6 and 6 Chapter 2 frontier theorems) are skeleton-pending with honest sorry markers." - "8 in-code sorry tactics remain in the production tree at the chapter-bind tag; discharge plans are named in Chapter 7 for each." - "Five frontier axioms (A12 SelfRefactoring, A13 Resonance, A14 GradientLambda, A16 SAEBounded, A17 ParetoConvergence) are not yet declared as axiom symbols in Lutar/; their status is narrative-reserved." - "The 934+ test count covers 30 in-scope modules (v14-v18.24); the live 32-module registry adds 2 later grafts (v18.25 and v19.0)."