AI Policy Wiki
Dashboard
⚲

Gap Scan — 2026-08-18

Daily gap hunt — what the wiki is missing and how each gap was triaged.

Scanned

Recent window: 3 New Developments Log/ files inside 48 hours (2026-08-17-2205, 2026-08-17-0808, 2026-08-16-2205); 40+ wiki pages carrying last_updated: 2026-08-17 or 2026-08-18, concentrated in companies/, concepts/, entities/ and legislation/.

Rotation slice: 5 — entities/ H–N (128 pages).

Broken-link analysis: 257 distinct broken targets across 283 references, after correcting for the escaped-pipe scanner artifact (see below). Maximum in-degree on any broken target is 2.

Queue state at start of run: 0 open INGEST- tasks; 140 records in gap-scan/proposed-sources/; 22 notes in gap-scan/needs-review/; 0 unprocessed dev-log files.

Run-order note: the nightly developments-log cycle had already folded the 2026-08-17 digests — the American Bankers Association submission was in industries/financial.md and the Brockman essay in concepts/defensive-ai-paradox.md before this scan began. Live-thread gaps were therefore assessed as missing anchors for already-folded facts, not as unfolded news.

Lock note: bin/vault-lock.sh acquire exits 1 in this sandbox because there are no git credentials, so the remote cannot be fetched and the lock cannot be arbitrated — a different failure than the "held locally, published by the sync loop" path the scheduled-task instructions anticipate. The lock was taken with VAULT_LOCK_OFFLINE=1 after confirming that no .vault/lock.json existed in the working tree, that the last audit-log operation had completed, and that the run fell outside the primary machine's 09:20–11:35 routine window. That forfeits arbitration; it is recorded here because it is a deviation from the documented protocol.

Gaps actioned (6 of 14 found)

New pages created (live)

  • American Bankers Association — The association's August 14, 2026 submission to House Financial Services Committee Democrats was folded into industries/financial.md with no entity page behind it, leaving a recurring federal-preemption advocate with no anchor. Built from the association's own AI topic page, its banking-journal report on the submission, and the Inside AI Policy lede already cited on the financial page. Covers the preemption position, the technology-neutrality and regulatory-perimeter argument anchored on Gramm-Leach-Bliley, the comment-letter record (FSB July 22, 2026; NIST AI-agent security March 9, 2026; NIST AI RMF joint trades February 12, 2026), the April 2026 interagency model-risk-management revision and its generative/agentic carve-out, and member-facing activity. confidence: medium, sources_count: 3. 1,006 words, 9 inline citations, zero broken links.

Pages expanded (live)

  • Mark Warner — 154 words, confidence: low, in-degree 14 — the thinnest high-in-degree page in the wiki's coverage of Senate AI policy, describing a six-bill package by naming one bill. Expanded to 1,078 words from Warner's own Senate press releases and floor remarks, which had not been used: the AI AGENT Act and its June 29, 2026 discussion draft, the SAFE AI Act, the Secure AI Development Act, the National Workforce Transition Fund, the PHD Talent Act, and the related Kennedy, Rounds and Investing in American Workers measures. All three original citations retained. 5 new (Source: URL) cites. confidence low → medium; sources_count 3 → 6.
  • Knight First Amendment Institute (Knight Columbia) — 286 words, in-degree 16 (highest in slice 5). The page described the institute purely as an essay publisher and omitted the litigation and advocacy half of its work, despite the institute describing itself as working through "strategic litigation, research, policy advocacy, and public education." Expanded to 887 words: the July 22, 2026 Ninth Circuit filing against xAI's challenge to California's training-data disclosure law, the Amazon v. Perplexity opinion analysis, the March 2025 Texas school-surveillance records action, the October 2023 congressional letter, the "Artificial Intelligence and Democratic Freedoms" project and its 2024–25 symposium history, Seth Lazar's August 2024 appointment as Senior AI Advisor, and eleven further AI essays in the series not previously listed. All six original publishes: wikilinks and every original fact retained. 3 new (Source: URL) cites. sources_count 6 → 8; confidence held at high.

None. Alias resolution was run against all 257 broken targets and produced zero unambiguous fixes — the fourth consecutive run with this result. The residue is genuinely missing pages, not misdirected links: author surnames from sources/ Relationships blocks (entities/adam-smith, entities/kevin-klyman, entities/sacha-altay, entities/joseph-bernstein, entities/lee-anne-fennell), institution shorthand (entities/sais, entities/uw, entities/carnegie-mellon, entities/ecb, entities/bank-of-england, entities/federal-reserve), and forward references explicitly marked "(planned)" or "(candidate)" in their source pages (concepts/talent-flow-china-us, concepts/agi-paths, concepts/algorithmic-decisionmaking). Two near-misses were checked and rejected: concepts/algorithmic-decisionmaking → concepts/algorithmic-accountability and concepts/enterprise-ai → concepts/enterprise-ai-deployment-gap are related but not the same subject, and silently redirecting them would assert an equivalence the source pages do not.

Queued — foundational sources

  • Greg Brockman, "The Defender's Window" (2026-08-16) — cited by name in a full paragraph of concepts/defensive-ai-paradox.md and carried in the 2026-08-17 22:05 digest, with no sources/ page and no Raw Sources/ file. Verified: blog.gregbrockman.com (author-controlled Svbtle domain; twitter:creator "@gdb"), live fetch, full text, distinctive passages confirmed. Saved: Raw Sources/Brockman - The Defender's Window (blog.gregbrockman.com, 2026-08-16).md. Queued: INGEST-brockman-defenders-window.md.
  • Sayash Kapoor & Arvind Narayanan, "Do AI Risks Require Extraordinary Government Intervention?" (2026-05-21) — a direct follow-on to AI as Normal Technology, which the wiki already carries at AI as Normal Technology and AI as Normal Technology. Searching Wiki/ and Raw Sources/ for the title returned zero matches. Verified: knightcolumbia.org, the original publisher; author bios, date and a distinctive passage all confirmed in the fetched full text. Queued URL-only: no raw file saved, because the essay is a stable, unpaywalled page on a canonical host best fetched fresh at ingest — the same disposition used for Colorado HB 26-1263 on 2026-08-13. Verification trail: gap-scan/proposed-sources/kapoor-narayanan-extraordinary-intervention-2026-05.md. Queued: INGEST-kapoor-narayanan-extraordinary-intervention.md.

Deferred to review (quality gate)

  • companies/spacex — "SpaceX" is named on 56 content pages with no page of its own, which reads as a large missing-company gap. It is more likely a naming problem: companies/xai.md already carries the merged entity's Snapshot rows, S-1, compute leases and IPO history, and creating a second page would violate the no-duplication rule rather than close a gap. The page slug and title still name the division retired in July 2026. Three dispositions and their costs are set out in gap-scan/needs-review/2026-08-18-spacex-page-vs-xai-page-placement.md; the choice is a supersede-class decision affecting inbound links across dozens of pages, and this scan did not make it.

Authenticity-verification failures

None. Both pulled sources came from author- or publisher-controlled canonical hosts, both returned full text rather than an excerpt, and both had a distinctive passage confirmed present in the fetched document. Neither carries a formal identifier, and neither should — a personal-blog essay and a nonprofit institute's Deep Dive piece have no DOI or arXiv namespace.

Scanner note — the escaped-pipe artifact, quantified

A naive \[\[([^\]\|#]+) scan reports 350 distinct broken targets across 416 references. 93 of those distinct targets (133 references) are not broken. They are aliased links whose pipe is backslash-escaped — [[companies/anthropic\|Anthropic]] — so the capture group swallows the trailing backslash and no slug matches. There are 200 escaped pipes in the wiki. Stripping the trailing backslash before comparison yields the true figure of 257 distinct / 283 references. This artifact has been noted in four prior reports; recording the count here so a future reader can tell at a glance whether a raw scan has been corrected. The bin/lint-scan.py figure of 244 distinct / 270 references, generated the same morning, is close but not identical — the two scanners differ slightly in what they exclude, and reconciling them is a small lint task rather than a gap.

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

  • entities/jam-kraprayoon — 229 words, in-degree 10, confidence: medium, sources_count: 2. IAPS frontier-security researcher; expandable from the IAPS research index and the Foreign Affairs co-authorship already cited. Score 4.
  • entities/jack-goldsmith — 135 words, the shortest page in the slice with in-degree ≥6 (8). Lawfare co-founder; the page acknowledges an AI intersection without documenting one. Score 3.
  • entities/john-thune — 228 words, in-degree 8. The Washington Post brief indicates the Thune–Klobuchar duty-of-care framework rests on developer self-testing, a mechanism the page does not record; R Street has published commentary on the proposal. Score 3.
  • entities/maria-cantwell — 257 words, in-degree 9. The competing pre-deployment-vetting position in the same Senate Commerce dispute. Score 3.
  • entities/imf — 346 words, sources_count: 1, in-degree 8, and the source of three of the wiki's broken institution-shorthand links. Score 2.
  • entities/horizon-institute — 585 words but sources_count: 0, in-degree 8. Score 2.
  • World Liberty Financial / WorldClaw — named in the 2026-08-17 08:08 digest (43 of 90 models on WorldClaw's site from Chinese firms including entity-listed Z.ai; Trump family owns 38% of World Liberty). No page for either; export-controls and conflict-of-interest angle. Score 3.
  • Derek Thompson — the interlocutor in the queued Kapoor/Narayanan essay; his underlying essay may itself be a foundational gap. Score 2.
  • Author-surname and institution-shorthand broken targets (score 1–2 each), unchanged from prior runs.

One-line summary

One entity page created and two thin high-in-degree anchors rebuilt from primary sources, two foundational essays verified and queued for the user's ingest review, and one apparent 56-reference company gap deferred to review as an entity-boundary question rather than a missing page; the alias-fix yield was zero for the fourth run running, and the lock was taken offline because this sandbox cannot reach the git remote.