Gap Scan — 2026-05-23 (Run 2, evening)

Second run today. An earlier gap scan ran ~15:32 (Gap Scan — 2026-05-23, 8 gaps actioned). This is the evening scheduled run; it scans content that appeared after the morning run — the 2026-05-23 Ingest-Reflect cycle and the 20:57 dev-log — plus the lint backlog and the morning run's deferred items, which "re-surface next run." Written to a -run2 file so the morning report is preserved. The 5 new pages below were committed live to their real Wiki/ folders (status: active, confidence: medium). One foundational source is verified and queued at Wiki/queue/INGEST-secret-loyalties-kwon-2026.md.

Scanned

Recent window (last 48h): the 2026-05-23 Ingest-Reflect cycle (3 foundational ingests + 7 dev-log files folded into ~25 pages; 13 new pages created — all post-dating the morning gap scan) and the one dev-log not yet processed by the nightly cycle, 2026-05-23-2057-ai-developments.md. Rotation slice: 2 of 14 — concepts/ G–O (same slice as the morning run; the slice index is date-derived). Broken-link analysis: 23,769 wikilinks across 1,226 pages → 534 unique broken targets / 789 instances. After alias resolution, doc-artifact filtering, and dropping the morning run's actioned/deferred items and the Ingest-Reflect cycle's 13 new pages, ~14 genuine candidates remained. The candidate list this run drew most heavily from the Lint Report's own gap list ("Unfixable gaps" + "Suggested next sources") and the morning run's deferred backlog. The 6 highest-scored were actioned.

Gaps actioned (6 of ~14 found)

New pages created (live)

• United States v. Google (Search) — United States v. Google LLC (1:20-cv-03010-APM, D.D.C., Judge Mehta). Flagged explicitly in the Lint Report "Unfixable gaps" ("No US v. Google search-antitrust litigation page"); the May 22 Google appeal was folded into AI Antitrust as prose with no case-tracking page. Live thread + wiki core area (AI competition). Built from the Winston & Strawn remedies analysis + NYT/Reuters appeal coverage; the Dec 2025 final judgment's explicit GenAI scope is the AI hook.
• Artificial General Intelligence (AGI) — Artificial General Intelligence as a definitional/discourse concept. 6 broken [[concepts/agi]] inbound links across embodied-ai-vs-agi (marked (planned)), ai-as-social-technology, sapient-intelligence, entities/henry-farrell, and a source page. Alias-resolution confirmed it is distinct from AGI Timelines (forecasting), Transformative AI (TAI) (impact-defined), and General-Purpose AI (GPAI) (EU regulatory term) — those pages explicitly cross-reference [[concepts/agi]] as a separate node. Built as a hub on the definitional contest.
• Iowa SF 2417 — Conversational AI Services Act — Iowa Conversational AI Services Act (new Iowa Code Ch. 554J). 4 broken inbound links from the chatbot-law-wave clusters in policy-briefs/state-chatbot-laws-tracker and concepts/state-level-ai-regulation. Real, signed law (Gov. Reynolds, 2026-05-06) — verified against the Iowa Legislature BillBook and corroborating trade press.
• AI Whistleblower Protection Act (S.1792) — Grassley's S.1792 (119th Congress). Named in the Lint Report "Suggested next sources" and already referenced on AI Whistleblowing and Radical Optionality: Governing Transformative AI Under Uncertainty after this cycle's contradiction fix. Built from congress.gov + the Senate Judiciary press release. (The bill text is cited inline to congress.gov, the wiki's standard practice for legislation primary texts — no separate source ingest queued.)
• Civitai — the largest open generative-AI image-model hub; the canonical case study in generative-AI content moderation and payment-processor-as-regulator dynamics. 4 broken inbound links, mis-prefixed entities/civitai; Civitai is a for-profit media-gen company → companies/.

Pages expanded (live)

None this run. The rotation-slice (concepts/ G–O) anchor pages and the recent-window pages scanned were already adequately built — the Ingest-Reflect cycle had just refreshed ~25 of them.

Link aliases fixed

• [[entities/civitai]] → [[companies/civitai]] — 4 instances across 3 pages (concepts/synthetic-media-deepfakes, sources/tech-policy-press-cuevas-take-it-down-ecosystem-2026-05-13 ×2, legislation/take-it-down-act). Bundled with the new companies/civitai page. A now-stale "(if exists, otherwise create…)" parenthetical on the Cuevas source page's depends-on line was also trimmed.

Queued — foundational sources

• "AIs with Secret Loyalties are a Serious but Addressable Threat" — Kwon et al., preprint, ~May 2026. Named in the Lint Report "Suggested next sources" as "a named research agenda on covert model loyalties/backdoors; not yet folded for lack of a clear home concept page." Authenticity verified — PASS: full PDF parsed directly from the canonical host (formationresearch.com, the lead author's research org — an acceptable host for a preprint); title, the 12-author "Kwon et al." block, and the abstract read from the fetched document; corroborated independently by the Forethought "ForeWord" newsletter. Queued: Wiki/queue/INGEST-secret-loyalties-kwon-2026.md. Verification trail: Wiki/queue/gap-scan/proposed-sources/secret-loyalties-kwon-2026.md. The INGEST task instructs the ingest to also create a concepts/secret-loyalties home page so the source is not a leaf node.

Authenticity-verification notes

One source verified — PASS (Kwon et al.). No sources rejected; nothing fabricated was saved. Per the gap-identifier large-PDF rule, the Kwon paper (~98K characters of markdown) was not saved to Raw Sources/ — the INGEST task carries the verified canonical URL for a fresh fetch at ingest time. One open item flagged for the ingest: no arXiv ID or DOI was located (the document is labelled only "Preprint"); the ingest should re-check, as one may be posted after this scan.

Corrections to the lint report

• The Lint Report "Unfixable gaps" entry "No Claude Opus 4.7 model page" is incorrect — Claude Opus 4.7 already exists. The broken references the lint cycle saw are plain-text mentions of "claude-opus-4-7"/"Opus 4.7", not broken [[wikilinks]]. No action needed; noted so the next lint pass can drop the stale entry. (The Claude Opus 4.7 system card as a foundational source is still a valid future QUEUE candidate — deferred below.)

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

• CMS/Medicare AI-deployment page — lint "Unfixable gaps": the WISeR prior-authorization model has no home page. Needs a government/cms page or a concepts/ai-prior-authorization concept. Score 3.
• companies/waymo — lint "Unfixable gaps"; the May 21 robotaxi flood-pause could not be folded. Genuine scope decision (autonomous-vehicle AI in/out of wiki scope) — defer to a human call. Score 2.
• Claude Opus 4.7 system card — foundational source (lint "Suggested next sources"); would anchor the evaluation-awareness thread. Score 3 — QUEUE candidate next run.
• OpenAI Erdős / unit-distance proof writeup — foundational candidate for AI for Science (lint); carried from both prior runs. Score 2.
• trustworthy-agents-in-practice — 9 inbound references in inconsistent forms ([[sources/...]], bare [[...]], [[trustworthy-agents-anthropic]]); Wiki/log.md records the source page as both deleted and created. This is a lint cleanup of references to a deleted page, not a clean gap-identifier pull — flagged for lint. Score 2.
• anthropic-pwc-expanded-partnership (7 broken links) — likely a supporting source that belongs folded into Anthropic, not a sources/ page. Lint cleanup. Score 2.
• government/nsf — National Science Foundation; NAIRR's lead agency + the X-Labs initiative. Scope-ambiguous (a research funder, not a clear deployer or regulator) — deliberately deferred. Score 2.
• 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, entities/anu (Australian National University), entities/cac (Cyberspace Administration of China).
• concepts/long-industrial-revolution — 4 broken links; AI-and-economic-history concept. Score 1.
• From the 20:57 dev-log (not gaps — noted for the nightly run): the May 18 Oakland jury verdict for OpenAI in Musk v. OpenAI folds into the existing Musk v. Altman (and OpenAI / Microsoft / Brockman) page (not a missing page). SpaceX's AI-reframed IPO S-1 and the John Doerr WSJ interview are borderline-scope items — left for the nightly developments-log run.

One-line summary

Evening run, 6 gaps actioned: five new pages created live — the US v. Google search-antitrust case and a CMS-flagged-style cluster of lint-backlog pages (concepts/agi, Iowa SF 2417, the Grassley AI Whistleblower Protection Act, Civitai) — plus the Kwon et al. "secret loyalties" preprint verified and queued; one stale lint claim (the non-existent "missing" Claude Opus 4.7 page) corrected for the record.