AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-08-17

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

Contents

Scanned

Recent window: 3 New Developments Log/ files in the last 48 hours (2026-08-16 22:05, 2026-08-16 08:08, 2026-08-15 22:08). Rotation slice: 4 — entities/ A–G, 184 pages.

Corrected broken-link scan (trailing-backslash de-escaped): 250 distinct broken targets across 279 instances before this run's edits; 245 across 272 after. A first-pass naive scan reported 341 across 408 — the 91-target difference is the escaped-pipe artifact recorded on 2026-08-13, -14, -15 and -16, confirmed again and not acted on.

Gaps actioned (6 of 27 candidates)

New pages created (live)

Pages expanded (live)

  • Gillian K. Hadfield — the slice's strongest thin anchor: in-degree 24, 288 words, medium/sources_count: 1, untouched since 2026-06-06. Added a full career section (Queen's, Stanford JD/PhD under Milgrom, Wald clerkship, Berkeley→Toronto→NYU→USC→Toronto→Johns Hopkins), the OpenAI advisory term with its 2018–2023 bounds, the Vector Institute CIFAR chair and AI2050 fellowship, a regulatory-markets section built from the primary arXiv record, and a publications section covering the Hadfield–Weingast rule-of-law work and the AI-alignment papers. 288 → 944 words; sources_count 1 → 4; confidence held at medium. Two corrections against the prior text: the page described her as "CEO of the Schwartz Reisman Institute," where the sources show she was its inaugural director at Toronto and has since moved to Johns Hopkins; and the JHU appointment is dated to the university's June 2025 announcement rather than left undated. 3 new (Source: URL) cites.
  • Benjamin Wittes — in-degree 10 on 130 words, the shortest page in the slice above the in-degree bar. Added Lawfare's founding date and institutional structure (Lawfare Institute, Brookings cooperation, Roger Parloff as co-editor), the co-founders' prior government roles, the 2023 multimedia-magazine characterisation in Wittes's own words, the 2017 travel-ban and Comey episodes, and attributed reception from both David Ignatius and Glenn Greenwald so the assessment is presented from both sides. 130 → 397 words; sources_count 2 → 4. 2 new cites.
  • Three broken targets removed from Gillian K. Hadfield during the rewrite — [[entities/johns-hopkins]], [[entities/schwartz-reisman-institute]], [[entities/barry-weingast]]. No page exists under any alias for the three, and all three facts survive as named prose rather than links, so the rewrite resolved them without creating three institution stubs. Net wiki-wide effect of this run: 250 → 245 distinct broken targets.

Queued — foundational sources

  • Regulatory Markets: The Future of AI Governance (Hadfield & Clark) — gap type 1. A grep for "regulatory market" across the content folders returns no substantive mainspace treatment, despite Gillian K. Hadfield carrying 24 inbound links and co-author Jack Clark being an Anthropic co-founder. Verified: arxiv.org/abs/2304.04914, DOI 10.48550/arXiv.2304.04914, published Jurimetrics 65:195–240 (2026); authors, title and the defining passage corroborated independently. Raw file not pre-saved — ~45-page article on the large-PDF route. Verification record: queue/gap-scan/proposed-sources/regulatory-markets-hadfield-clark-2023.md. Queued: INGEST-regulatory-markets-hadfield-clark.md. Not ingested.

Authenticity-verification failures

None. One source pulled, one verified.

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

  • Deirdre K. Mulligan — attempted and deferred (score 5). In-degree 15 on 284 words, medium/3, stale since 2026-06-06; the second-strongest thin anchor in the slice. The Berkeley iSchool faculty page exceeded the fetch size limit and no second primary profile was tried within the cap. Expanding her from a weaker secondary source would have failed the quality gate, so the page was left untouched and the backup taken for it was not used. First candidate next run.
  • Slice-4 thin anchors not reached (score 3–4 each). entities/guidelight-ai-standards (in-degree 14, 229 words, confidence: low — the only low page above the in-degree bar anywhere in the slice, and the strongest remaining), entities/constellation-energy (11, 322 words), entities/chris-lehane (10, 467 words, medium/1 — OpenAI's chief global affairs officer, core to the policy thread), entities/center-for-democracy-and-technology (10, 343 words), entities/ai-now (9, 740 words, medium/1), entities/anthropic-model-welfare (9, 244 words), entities/dpc (8), entities/gabe-pereyra (8), entities/alison-gopnik (8), entities/amy-zhang (8), entities/david-mcdonald (8), entities/ed-markey (8, 226 words on 6 sources), entities/brianna-rosen (8, 186 words), entities/brian-armstrong (7), entities/divya-siddarth (7, 226 words on confidence: high — a confidence-versus-sourcing mismatch of the kind flagged in slice 2 on 2026-08-15). Fourteen pages; the slice is materially under-built and one run cannot clear it.
  • concepts/regulatory-markets — deliberately not created (score 4). The natural downstream anchor for the queued source, and the wiki has precedent for concept pages carrying a named framework. Held because the concept is the paper's argument: writing it from an abstract before the primary text is ingested would produce exactly the thin, second-hand page the quality gate exists to prevent. Create it in the same cycle that processes INGEST-regulatory-markets-hadfield-clark.md.
  • Author and institution pages implied by sources/ pages — backlog partially broken, seventh recording. Two of the twelve standing names were built this run (Eric Goldman, Orin S. Kerr). Still open: entities/kevin-klyman, entities/alondra-nelson, entities/lee-anne-fennell, entities/michael-j-d-vermeer, entities/sacha-altay, entities/joseph-bernstein, entities/clayton-christensen, entities/samuel-weinbach, entities/ilhan-scheer, entities/dawn-song, plus entities/ryan-calo, entities/michelle-mello and entities/daniel-ho newly surfaced. Ten remain from the original list.
  • Institution shorthand targets (score 1–2), unchanged. entities/carnegie-mellon, entities/santa-fe-institute, entities/sais, entities/uw, entities/federal-reserve, entities/ecb, entities/bank-of-england, entities/fsb, entities/doj, entities/cma, entities/ec. Recorded for a fifth consecutive run. sais, uw, ec and cma are abbreviations better rewritten to full names in the citing prose than created as pages; that rewrite is a lint task, and recording it a fifth time without routing it there is itself the finding.
  • Concept targets named but pageless (score 2 each), unchanged. concepts/inverse-cooking-problem and concepts/inverse-trust-problem (held under needs-review/2026-07-10-inverse-cooking-trust-coined-terms.md as coined terms), concepts/a-vision-of-democratic-ai, concepts/data-broker-regulation, concepts/talent-flow-china-us, concepts/economic-possibilities-for-artificial-intelligence, concepts/ai-and-language-models, concepts/algorithmic-decisionmaking (sixth consecutive run), concepts/dead-internet-theory.
  • The escaped-pipe scanner artifact. Eighth consecutive recording. A naive \[\[([^\]\|#]+) capture reads the escaped pipe in [[folder/slug\|Alias]] table links as a trailing backslash. This run: 341 first-pass targets, 250 corrected. The fix is one regex in bin/lint-scan.py; it is lint's lane, not gap-identifier's, and it has now been reported eight times without being applied.
  • index.md operational-path backlog (score 1). Twelve index.md pointers to paths that moved under _meta/ in the v4.3 reorganisation. Index maintenance, not a content gap; carried across six runs.

Self-audit of this run

  • Broken-link count moved in the right direction. 250 → 245 distinct targets, 279 → 272 instances. Three resolved by page creation (concepts/section-230, entities/eric-goldman, entities/orin-kerr) and three by the Hadfield rewrite dropping unresolvable institution links; one placeholder ([[sources/...]]) was introduced in a provenance note during drafting, caught by the post-write scan, and rewritten as plain text before the run closed.
  • Style scan clean. All five created or revised pages checked against the banned-header, hyperbole, self-reference and standing-Predictions lists; zero hits.
  • Revision fidelity. Both expanded pages were backed up to _meta/_revision-backups/entities/ before rewriting. Every pre-existing fact and citation on Gillian K. Hadfield and Benjamin Wittes survives; the only substantive change to prior content is the Schwartz Reisman role correction, which replaces an unsourced claim with a sourced one rather than dropping a fact. The unused deirdre-mulligan backup could not be removed (permission denied on the synced folder) and is harmless.
  • Cap respected. Six gaps actioned against a 6–10 cap; four web pulls, no crawls.

One-line summary

Six gaps actioned — three new pages (Section 230 (47 U.S.C. § 230), Eric Goldman, Orin S. Kerr), two thin anchors expanded (Gillian K. Hadfield 288→944 words, Benjamin Wittes 130→397), and the Hadfield–Clark regulatory-markets paper verified and queued for review; the run also surfaced six genuinely missing US-law anchor pages (CFAA, COPPA, Authors Guild v. OpenAI, Moody v. NetChoice, the foreign direct product rule, p(doom)) that no prior scan had recorded, and left slice 4 with fourteen thin entities/ anchors still unreached.

Categories: Briefings | briefing | gap-scan