Gap Scan — 2026-05-22

Live run. Auto-fixes and thin-anchor expansions were committed to the wiki. The 7 new pages were drafted and then accepted live the same day (see "New pages created" below). One foundational source is queued at Wiki/queue/INGEST-law-following-ai-okeefe-2025.md, awaiting ingest.

Scanned

Recent window (last 48h): 6 New Developments Log/ files (2026-05-21-2327 through 2026-05-23-0511); wiki pages edited 2026-05-21 to -22. Rotation slice: 2 of 14 — concepts/ G–O. Broken-link analysis: 18,202 wiki links scanned across 1,150 pages → 485 unique broken targets, 751 instances. After alias resolution and doc-artifact filtering, ~60 are genuine missing-page gaps. ~22 were actionable; the 10 highest-scored substantive gaps were actioned (link-alias fixes are cheap maintenance and not counted against the cap).

Gaps actioned

Auto-filled (live) — committed to the wiki

Link aliases fixed — 12 instances across 7 pages, rewritten to existing targets:

• [[concepts/ai-x-risk]] → [[concepts/ai-existential-risk]] — 5 instances across anticipatory-ai-ethics, the Lazar source page, entities/seth-lazar.
• [[entities/aisi]] → [[entities/aisi-network]] — 7 instances across 2026-05-17-brief, sociotechnical-risk-governance, the Mulligan-Marda-Wang source page, entities/deirdre-mulligan. (All instances refer to AI Safety Institutes as a category; aisi-network is the collective page.)

Thin anchors expanded — heavily-linked entity pages strengthened from supporting web sources:

• Yoshua Bengio — 235 → ~520 words (in-degree 16). Added a scientific-contributions section and his June 2025 launch of LawZero / the "Scientist AI" non-agentic-safety program. (Sources: yoshuabengio.org, lawzero.org, time.com.)
• Timnit Gebru — 218 → ~480 words (in-degree 16, was sources_count: 0). Added Gender Shades, "Stochastic Parrots," and the founding and premise of the DAIR Institute. (Sources: dair-institute.org, macfound.org.)

Reclassified mid-run

• ai-as-cultural-technology — flagged in the dry run as a link-alias for ai-as-social-technology. The alias-resolution check found the ai-as-social-technology page explicitly distinguishes itself from it (Gopnik's "cultural technology" vs. Farrell-Shalizi's "social technology"). It is a distinct concept — moved from "alias fix" to a proposed new page.

Queued — foundational source (verified, awaiting ingest)

• "Law-Following AI: Designing AI Agents to Obey Human Laws" — O'Keefe, Ramakrishnan, Tay & Winter, 94 Fordham L. Rev. 57 (2025). Anchors 5 broken [[concepts/law-following-ai]] links and is referenced from Normative Competence; the paper was not in Raw Sources/. Authenticity verified — PASS (title/authorship/citation corroborated across 5 independent references; canonical host identified as the Fordham Law Review PDF). Queued: Wiki/queue/INGEST-law-following-ai-okeefe-2025.md. Verification trail: Wiki/queue/gap-scan/proposed-sources/law-following-ai-okeefe-2025.md.

New pages created

Accepted 2026-05-22. All 7 pages below were reviewed and accepted into their real folders the same day — they are now live wiki pages, and the formerly-broken inbound links to them resolve. Going forward the gap-identifier creates new pages live by default (CLAUDE.md v4.1).

• concepts/responsible-scaling-policy — 13 broken inbound links, the most-referenced missing page in the wiki. Drafted from the already-ingested Anthropic's Responsible Scaling Policy (Version 2.2) / Anthropic's Responsible Scaling Policy (Version 3.1).
• concepts/safety-cases — 6 broken inbound links. Drafted from Safety Cases for Frontier AI (in-degree 27, previously a leaf node).
• concepts/law-following-ai — 5 broken inbound links. Drafted from secondary framing; flagged to be upgraded once the queued primary source is ingested.
• concepts/ai-ethics-frameworks — planned in Crystallize Session: Wiki Gaps & User Knowledge Map (2026-04-13) but never created; the other three pages planned that session now exist.
• concepts/ai-as-cultural-technology — 4 broken inbound links (see "Reclassified" above).
• entities/alison-gopnik — 6 broken inbound links; co-architect of the cultural/social-technology framings the wiki tracks.
• legislation/defense-production-act — 5 broken inbound links; the statutory basis EO 14110 used to compel frontier-model reporting.

Authenticity-verification notes

One source verified (Law-Following AI) — PASS. No sources rejected. Tooling note: firecrawl's JSON extraction on the SSRN page returned hallucinated placeholder metadata because the page is Cloudflare-gated (HTTP 403); that output was discarded and the paper verified via independent references instead — the protocol working as intended.

Deferred backlog (over the daily cap — re-surfaces next run)

• legislation/california-ab-2013 — 6 broken links; only the source summary AB 2013 — Training Data Documentation (California) exists, no legislation/ page (schema requires one).
• legislation/iowa-sf-2417 — 4 broken links — missing state AI law page.
• concepts/agi — 4 broken links — borderline; may be a link-alias for AGI Timelines. Needs link-context inspection.
• companies/sapient-intelligence + models/hrm-text — from the May 22–23 dev-logs; deferred pending check that the nightly ingest did not already create them.
• legislation/illinois-sb-317 — Consumer AI Notice Act, passed the IL Senate 2026-05-21.
• government/nsf (X-Labs $1.5B initiative) and a WISeR page (AI Medicare prior-authorization model).
• OpenAI unit-distance / Erdős proof — primary proof artifact cited in the May 23 dev-log; candidate foundational pull.
• Missing entity pages (3–6 broken links each): david-mcdonald, amy-zhang, rakshit-trivedi, nik-marda, victor-zhenyi-wang, herbert-simon, brian-armstrong, jack-dorsey.
• Missing source pages flagged by broken links: anthropic-pwc-expanded-partnership (7), topol-marcus-llm-patient-outcomes (6), trustworthy-agents-in-practice (5).

One-line summary

First live run: 12 broken-link instances auto-fixed, 2 thin entity anchors expanded, 1 foundational source verified and queued, and 7 new pages created — led by responsible-scaling-policy, which had 13 dangling links and now has a concept page connecting two already-ingested sources.