Citation integrity
Policy Window's core claim is that every classification cites a primary source. This page makes that claim measurable and checkable rather than asking you to take it on trust. Every number below is computed deterministically from the published catalog and recomputed each deploy.
Last recomputed 2026-07-06 · machine-readable: /wiki/verification.json
What this attestation certifies — and what it doesn't
✓ Attested
- Every published coverage cell carries a non-trivial primary-source citation pointer.
- Every published instrument carries an https source URL.
- Where a provision excerpt is present, it is published for you to check against the source — and a verbatim quote is distinguished from a disclosed paraphrase, never conflated.
- Each verified Wikidata QID dereferences to the correct entity on wikidata.org.
✗ Not attested here
- That a cited provision's text supports the classification — that is human editorial review (Coverage Games), disclosed in the methodology.
- Real-time source-URL liveness — checked weekly by CI, not at page load.
- Excerpt coverage is honestly low (20.9%); backfilling verbatim excerpts is the standing editorial priority.
Reproduce this catalog
The catalog's coverage claims hash to a stable fingerprint you can re-derive yourself from /wiki/catalog/json and compare — “deterministic by construction” made checkable. A matching hash proves you hold byte-identical catalog claims.
Recipe: SHA-256 (hex) of the published coverage cells rendered as one line each, `<instrument>:<topic>\t<type>\t<citation>\t<confidence>` (confidence empty if unset), the lines sorted ascending and joined with '\n'. Re-derive from /wiki/catalog/json: for each entry in `.coverage` emit the same line, sort, hash. Full machine attestation: /wiki/verification.json.
Tamper-evident citations
Every date-pinned snapshot you cite is content-addressed. Fetch /wiki/<slug>/snapshot.json?asOf=YYYY-MM-DD and record its contentHash (sha256 of the captured catalog row). Later, re-fetch and recompute sha256 over the returned serializedRow: if it still equals your recorded hash, the snapshot you cited is provably unaltered — so a regulator, court, or paper can prove a Policy Window ?asOf=citation hasn't changed under it.
Per-instrument integrity
| Instrument | Juris. | Cells | Cited | Excerpts | Source | QID |
|---|---|---|---|---|---|---|
| Anthropic Responsible Scaling Policy (RSP) v2 | US | 12 | 12 | 0 | ✓ | — |
| ASEAN Guide on AI Governance and Ethics | ASEAN | 12 | 12 | 0 | ✓ | — |
| African Union Continental AI Strategy | African_Union | 12 | 12 | 0 | ✓ | — |
| Bletchley Declaration on AI Safety | global | 12 | 12 | 0 | ✓ | — |
| Brazil AI Bill (PL 2338/2023) | BR | 13 | 13 | 0 | ✓ | — |
| California SB-1047: Safe and Secure Innovation for Frontier AI Models Act | US | 12 | 12 | 0 | ✓ | — |
| California SB 243: Companion Chatbots | US | 3 | 3 | 3 | ✓ | — |
| California SB-53: Transparency in Frontier Artificial Intelligence Act (TFAIA) | US | 7 | 7 | 3 | ✓ | — |
| California SB 942: AI Transparency Act | US | 5 | 5 | 3 | ✓ | — |
| Provisions on the Administration of Deep Synthesis of Internet Information Services | CN | 7 | 7 | 7 | ✓ | — |
| Interim Measures for Generative AI Service Management | CN | 22 | 22 | 0 | ✓ | — |
| Council of Europe Framework Convention on AI | council_of_europe | 22 | 22 | 0 | ✓ | — |
| Google DeepMind Frontier Safety Framework | US | 11 | 11 | 0 | ✓ | — |
| DFARS Subpart 252.204 (Safeguarding Covered Defense Information and Cyber Incident Reporting) | US | 5 | 5 | 2 | ✓ | — |
| DoD Responsible AI Strategy and Implementation Pathway | US | 6 | 6 | 4 | ✓ | — |
| EU AI Act | EU | 22 | 22 | 18 | ✓ | ✓ |
| General Data Protection Regulation (GDPR) | EU | 4 | 4 | 0 | ✓ | ✓ |
| EU General-Purpose AI Code of Practice | EU | 5 | 5 | 0 | ✓ | — |
| Revised Product Liability Directive (Directive (EU) 2024/2853) | EU | 3 | 3 | 1 | ✓ | — |
| Directive (EU) 2024/2831 on improving working conditions in platform work | EU | 5 | 5 | 5 | ✓ | — |
| FedRAMP AI Cloud Procurement Guidance | US | 6 | 6 | 1 | ✓ | — |
| G7 Hiroshima AI Process Code of Conduct | G7 | 22 | 22 | 0 | ✓ | — |
| GSA Generative AI and Specialized Computing Infrastructure Acquisition Resource Guide | US | 6 | 6 | 3 | ✓ | — |
| India Digital Personal Data Protection Act + AI Advisory (MEITY) | IN | 13 | 13 | 0 | ✓ | — |
| Italy Law No. 132/2025 on Artificial Intelligence (Legge 23 settembre 2025, n. 132) | IT | 15 | 15 | 15 | ✓ | — |
| Japan AI Promotion Act (Act on the Promotion of Research, Development and Utilization of AI-Related Technologies) | JP | 10 | 10 | 10 | ✓ | — |
| Japan METI AI Guidelines for Business | JP | 12 | 12 | 0 | ✓ | — |
| Meta Frontier AI Framework | US | 12 | 12 | 0 | ✓ | — |
| NIST AI Risk Management Framework | US | 22 | 22 | 7 | ✓ | — |
| NIST AI RMF Generative AI Profile | US | 13 | 13 | 0 | ✓ | — |
| New York RAISE Act: Responsible AI Safety and Education Act | US | 5 | 5 | 3 | ✓ | — |
| OECD AI Principles (Recommendation) | OECD | 22 | 22 | 0 | ✓ | — |
| OMB Memorandum M-24-10 (Advancing Governance, Innovation, and Risk Management for Agency Use of AI) | US | 7 | 7 | 4 | ✓ | — |
| OpenAI Preparedness Framework | US | 12 | 12 | 0 | ✓ | — |
| Seoul Declaration on Safe, Innovative and Inclusive AI | global | 12 | 12 | 0 | ✓ | — |
| Singapore Model AI Governance Framework for Generative AI | SG | 13 | 13 | 0 | ✓ | — |
| UK-US AI Safety Institute Memorandum of Understanding | global | 11 | 11 | 0 | ✓ | — |
| UK Pro-Innovation Approach to AI Regulation (White Paper) | UK | 22 | 22 | 0 | ✓ | — |
| UN Global Digital Compact | UN | 11 | 11 | 11 | ✓ | — |
| UN GA Resolution on Safe, Secure, Trustworthy AI | UN | 22 | 22 | 0 | ✓ | — |
| UNESCO Recommendation on the Ethics of Artificial Intelligence | UNESCO | 12 | 12 | 9 | ✓ | — |
| Executive Order 14110 on Safe, Secure, Trustworthy AI | US | 22 | 22 | 0 | ✓ | — |
| Executive Order 14179 — Removing Barriers to American Leadership in AI | US | 8 | 8 | 0 | ✓ | — |
| TAKE IT DOWN Act (Tools to Address Known Exploitation by Immobilizing Technological Deepfakes on Websites and Networks Act) | US | 2 | 2 | 1 | ✓ | — |
| White House Voluntary AI Commitments | US | 13 | 13 | 0 | ✓ | — |
“Cited” = cells carrying a primary-source citation pointer. “Excerpts” = cells additionally carrying a provision excerpt (a verbatim quote or a disclosed paraphrase, distinguished on the cell). “QID” = a Wikidata identifier verified to resolve to the instrument.
Article prose citation integrity
The coverage matrix is not the only cited surface. Every published article also carries developed analytical body sections, and the “every claim cites a primary source” standard extends to them. The figures below are parsed deterministically from the prose and are machine-readable at /wiki/prose-citations.json (CC0 citation data, per I4OC) so a researcher or AI agent can ingest the analysis and its citation network, not just the matrix.
Honest scope: this attests that each prose section carries inline citations and that most bibliographic anchors resolve to a catalogued source; it does not auto-confirm that each cited work supports the exact sentence — that is the named-editor / adversarial-review layer disclosed under charter §7.9/§7.10.