AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-08-29

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

Contents

Scanned

Recent window: 4 New Developments Log/ files (2026-08-27 08:10 and 22:05; 2026-08-28 08:10 and 22:10) and 73 wiki pages edited in the last 48 hours. Rotation slice: 2 — concepts/ G–O (44 pages). Mechanical inputs: bin/lint-scan.py (214 distinct broken targets, 0 referenced 2+ times) and bin/primary-doc-scan.py --days 3 (6 digests; 15 cited primary-document URLs, 4 missing from Raw Sources/, 1 "possible match under another address"; 5 keyword-lane hits).

Gaps actioned (6 of 21 found)

New pages created (live)

Pages expanded (live)

  • Federal AI Compliance Landscape — score 5 (live thread +3, core +2). 1,006 → 1,367 words, sources_count 6 → 8, last_updated 2026-06-06 → 2026-08-29. The page catalogued the M-25-21 framework but carried nothing on how agencies apply its "principal basis" test in practice, which is the question the August 2026 OPM memorandum turns on. Added a thematic ## Agency interpretation of the high-impact standard section covering the presumed-high-impact list, OPM's reading of independent review as supplying a decision's justification, the two failure conditions the memorandum names (ratification without review of the underlying record; AI as sole scoring instrument without a reviewable basis), and the CAIO documentation-and-waiver path. 2 new (Source: URL) cites. Confidence held at high; not raised. Additive only — no section rewritten, no fact or citation removed.

None. Sixth consecutive run reporting zero, and the reason is unchanged and settled: bin/lint-scan.py puts 0 of 214 broken targets in the referenced-2+ lane, and the single-reference long tail is author surnames, institution shorthand and deliberate forward references, which the skill directs not to re-audit. Broken-link count unchanged across the run at 214 distinct / 214 refs, as expected — none of the six actioned gaps was a broken target.

Queued — foundational sources

  • OPM, Use of Artificial Intelligence in the Federal Hiring Process (memorandum, 2026-08-27) — score 8 (live thread +3, uncaptured primary document +3, US regulation/procurement core +2). Named in the 2026-08-28 22:10 digest's ## Primary documents block as "behind source 3, not retrieved." Verified: opm.gov/chcoc/latest-memos/, HTTP 200, 14-page PDF, signed by Director Scott Kupor. Not saved to Raw Sources/ — queued URL-only with the reason recorded, see the capture note below. Queued: INGEST-opm-ai-federal-hiring-memo-2026.md. Verification record: queue/gap-scan/proposed-sources/opm-ai-federal-hiring-memo-2026.md.
  • H.R. 10146, AI Advertising Disclosure Act (introduced 2026-08-24) — score 8 (live thread +3, uncaptured primary document +3, core +2). Verified: congress.gov, HTTP 200, full single version. Saved: Raw Sources/H.R. 10146 - AI Advertising Disclosure Act (119th Congress, 2026-08-24).md. Queued: INGEST-hr-10146-ai-advertising-disclosure-act.md.
  • H.R. 10152, Open-Source AI Leadership Act (introduced 2026-08-27) — score 8, same basis. Verified: congress.gov, HTTP 200, full single version. Saved: Raw Sources/H.R. 10152 - Open-Source AI Leadership Act (119th Congress, 2026-08-27).md. Queued: INGEST-hr-10152-open-source-ai-leadership-act.md.

Authenticity-verification failures

None. Four documents pulled (the OPM memorandum, both bills, and OPM's USA Class release), four verified against their canonical hosts, zero rejected.

One capture was deliberately withheld from Raw Sources/ rather than saved, which is a different outcome from a verification failure and is recorded as such. The OPM memorandum is authentic and its body text extracts cleanly, but Appendix A's Table 1 — the four-column framework applying the high-impact test to seven hiring use cases, and the operative part of the document — comes back with cell contents scrambled across columns and rows, so that clauses such as "However, OPM would" and "the AI use serves as a principal" sit detached from their sentences. Raw Sources/ is the evidentiary record the source-fidelity guarantees rest on; a capture whose central table is silently reordered is worse than none, because the damage is invisible downstream. The queue task and verification record both carry the instruction to fetch fresh with a structure-preserving parser or transcribe Appendix A page by page against the rendered PDF.

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

  • Louisiana Attorney General's certified petitions against Snap and Discord (score 3), two lane-1 hits. Dropped from scoring, not deferred: INGEST-louisiana-ag-snap-discord-petitions-2026.md is open in the queue and records the blocker — both PDFs are image-only scans returning roughly one character per page, so no text-layer capture is possible without a docket copy or a page-verified OCR pass. Nothing this run changes that.
  • The draft executive order creating a self-regulatory organization for frontier developers (score 6 by scoring, unactionable in fact). The 2026-08-28 22:10 digest's lead item, sourced to two people who have viewed a draft, with the article body paywalled and only the abstract retrievable. There is no published instrument to capture and no confident page to write: the reporting states the drafting has stalled. "Self-regulatory organization" is already named on four content pages including Demis Hassabis (the FINRA-modelled standards body proposal the draft is traced to) and AI Governance (umbrella). Declined rather than deferred as a document gap; re-enters as a concept gap if a text or a second independent account appears.
  • legislation/eu-cyber-resilience-act (score 3). The CRA's September 11, 2026 reporting layer over NIS2, GDPR and MDR vigilance was the 2026-08-28 22:10 digest's international item, and the CRA is named on only 2 content pages (Regulating Under Uncertainty: Governance Options for Generative AI (G'sell, 2024), Inside My AI Law & Policy Class 21: Understanding How the EU Regulates AI (Farahany, November 2025)). EU cyber regulation sits adjacent to the wiki's core areas rather than inside them; below the cap on that basis, not on merit.
  • entities/adam-thierer (score 3). Named on Federal Trade Commission (FTC) and Radical Optionality: Governing Transformative AI Under Uncertainty, and the named voice in the digest's Paxton data-center-liability item. Two mentions is thin grounds; carried.
  • Two slice-2 thin anchors (scores 2 and 1) — the whole of what the rotation slice produced. The Long Industrial Revolution (in-degree 10, 482 words, confidence: medium, sources_count: 1, last_updated: 2026-06-06 — 84 days stale) and Metagaming (in-degree 9, 611 words, sources_count: 1, 34 days stale). Both scored below the three live-thread candidates. Worth recording that slice 2 is in good shape: 2 thin anchors at in-degree ≥6 across 44 pages, against 4 in slice 1 yesterday.
  • Live-thread items owned by the nightly fold, not by this skill (score 3 each). The 2026-08-28 22:10 digest is unprocessed (unprocessed_devlog=1), so developments-log still owns the Nvidia–Hugging Face acquisition at $12.9bn, the Anthropic IPO figures, the EU AI Act enforcement letters, the Tencent Hy4 release, the a16z Machine Age fund, and the Kioxia–Sandisk Japan investment. Two of these were checked against the page set and would be genuine gaps if the fold does not create them: companies/cognition and companies/workday both exist, but a16z resolves correctly to the existing Andreessen Horowitz (a16z) — an alias check that would otherwise have read as a 28-mention missing-company gap.
  • Twenty-one source-robustness-check candidates across slices 11–13 — sixth consecutive report, structural. Unchanged, led by California SB 53 — Transparency in Frontier AI Act (in-degree 254) and TAKE IT DOWN Act — Source Summary (97). Out of this skill's lane; needs a source-robustness-check run.
  • The 28 date-stamped headers in lint-report.md (score 1) — thirteenth cycle, structural. Unchanged in count and in substance. The check does not distinguish a litigation procedural history from a timeline dump, so the number will not reach zero and re-listing it each run adds nothing.
  • Carried unchanged. models/deepseek-v3 (in-degree 49 on 608 words, routed to model-robustness-check on 2026-08-22, still unaddressed); the CISA AA26-231A advisory and the Exponential View AI-economy report, both of which have open queue tasks or existing Raw Sources/ files and were dropped from scoring rather than deferred.

Dedupe drops (all correct)

Five candidates were dropped as done or in flight rather than scored. The Louisiana AG petitions, the DeepMind double-blind evaluations report and the CISA Siemens PLC advisory each carry an open INGEST- task. Bill Gates's August 26 memo and the 128-signatory collective cyber-defense letter each already have a Raw Sources/ file. The 2026-08-27 evening digest's OpenAI Hugging Face technical report is both captured and ingested.

Checks run before closing

  • No broken outbound links introduced. Every wikilink on the three new pages and the one expanded page was resolved against the full page set after writing: zero unresolved on all four.
  • Style scan clean, first pass. All four touched pages checked programmatically against the banned-header, hyperbole, self-reference, standing-Predictions and date-led-header lists: zero hits on each.
  • Revision fidelity. Federal AI Compliance Landscape is the only existing page touched and the edit was purely additive — one new ## section inserted between existing sections, one page added to an existing **related:** line, and two frontmatter fields updated. No section was rewritten; all nine original sections, all tables and all citations are present unchanged.
  • A date error was caught and corrected before close, recorded rather than fixed silently: the OPM queue task first gave the Building the AI Workforce of the Future memorandum as 2026-12-15 against the document's own December 15, 2025. The OPM — Office of Personnel Management (AI Deployer) page carried the correct year throughout.
  • A source-fidelity discrepancy is preserved rather than smoothed. The 2026-08-28 digest attributes an 87-day average federal time-to-hire to "Adam Starr, OPM's chief information officer." The memorandum is signed by Director Scott Kupor and gives no such figure — it cites the Merit Hiring Plan target of 80 days. The 87-day number reaches the record through a separate Starr blog post by way of Nextgov. Both attributions are kept distinct on the page and flagged in the queue task.
  • Confidence set honestly. All three new pages are confidence: medium; none met the three-independent-source bar. Federal AI Compliance Landscape holds at high and was not raised.
  • Cap respected. Six gaps actioned against a 6–10 cap; 21 candidates found. Four scrapes and one search, no crawls, no failures.
  • Dashboard rebuilt. bin/build-dashboard.py wrote 2,058 article pages.

One-line summary

Three new pages (OPM — Office of Personnel Management (AI Deployer), AI Advertising Disclosure Act (H.R. 10146, 119th Congress), Open-Source AI Leadership Act (H.R. 10152, 119th Congress)) and one expansion (Federal AI Compliance Landscape), with three federal primary documents verified and queued — two bills captured in full, and the OPM hiring memorandum deliberately left uncaptured because its operative appendix table extracts scrambled, which is the item needing the curator's attention.

Categories: Briefings | briefing | gap-scan