Gap Scan — 2026-08-09

Scanned

Recent window: 4 New Developments Log/ files (2026-08-07 08:05 and 22:05; 2026-08-08 08:10 and 22:11) and 56 wiki pages with last_updated in the last 48 hours. Rotation slice: 10 — analysis/ (all), 26 pages.

Broken-link pass over 1,784 content pages found 2,581 distinct [[wikilink]] targets and 270 unresolved after alias normalisation, of which only 8 carry an inbound count of 2 or more. The open INGEST- queue was empty at the start of the run; gap-scan/proposed-sources/ held 20 prior verification records and was checked for overlap.

Gaps actioned (7 of 19 found)

New pages created (live)

• India IT Rules Amendment 2026 (Synthetically Generated Information) — score 5. India had zero legislation/ pages before this run, against four China pages plus Brazil, Japan, South Korea, the EU and the UK, while being named across more than twenty mainspace pages. The 2026-08-08 22:11 file carried the synthetically-generated-information amendments in detail. 1,380 words; confidence: medium. Depends on queued source: IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026.
• California AB 412 (AI Copyright Transparency Act) — score 5. The AI Copyright Transparency Act faces a Senate Appropriations suspense-file vote on 13 August 2026 and had zero mainspace mentions, despite AI Copyright Litigation — Analysis carrying an in-degree of 77 and California AB 2013 — Generative AI Training Data Transparency already covering the statute AB 412 would build on. 909 words; confidence: medium.
• Agent Plugins 1.0.0 — score 4. Agent Plugins 1.0.0, published 6 August 2026 by a technical steering committee drawn from AWS, Cursor, Microsoft, OpenAI and Vercel, had zero mainspace mentions while Model Context Protocol (MCP) — one of the two component formats it packages — is referenced across twelve pages. 631 words; confidence: medium. Depends on queued source: Agent Plugins Specification v1.0.0.
• Clément Delangue — score 4. Named in prose across nine mainspace pages with no page behind him; deferred on 08-06 and again on 08-08. Quoted in the 2026-08-08 window on accountability for agent-driven intrusions. 520 words; confidence: medium.
• Frontier Security Institute — score 4. Quoted twice in the window as the outside voice on the NIST evaluation draft, cited on three existing pages including NIST AI 200-2 — The TEVV-Athlon Framework for Evaluating AI Systems and National Institute of Standards and Technology (NIST); deferred on 08-08. 311 words; confidence: low, with an explicit sourcing-and-confidence section — the institute is newly launched and no organisational publications, funding disclosures or board composition were locatable.

Pages expanded (live)

• Federal AI Adoption — Patterns and Tensions — score 4, and the slice-10 finding. In-degree 19 on 758 words, the highest-reliance underbuilt page in analysis/, untouched since 2026-07-25. Expanded to 1,630 words with three additions and one repair: a new Measured adoption across agencies section carrying the governmentwide use-case counts (710 in 2023 to more than 3,600 in 2025), the large/midsize/small concentration figures, the per-agency mission composition, and the OMB M-25-21 high-impact reclassification analysis; a build-out of the procurement architecture with the OneGov launch date, the USAi launch announcement and named GSA officials, FedRAMP 20x, and the unquantified return-on-investment position attributed to GSA's chief AI officer; and a new tension subsection on the absence of a common federal evaluation standard, anchored to NIST AI 200-2. sources_count 10 → 15. Confidence held at high (10 prior sources plus 5 new, no contradictions). 5 new (Source: URL) citations. Depends on queued source: Brookings, "Assessing the state of AI adoption across the federal government".

Link aliases fixed

• Federal AI Adoption — Patterns and Tensions ## Relationships — the line planned [[government/dhs]], planned [[government/dod]], planned [[government/fda]] described three pages that now exist. Replaced with a typed deploys-in: line covering all eight government deployers the page's body actually discusses (DHS — Department of Homeland Security (AI Deployer), DOD — Department of Defense (AI Deployer), FDA — Food and Drug Administration (AI Deployer), ICE — Immigration and Customs Enforcement (AI Deployer), CBP — Customs and Border Protection (AI Deployer), DOT — Department of Transportation (AI Deployer), DOGE — Department of Government Efficiency (AI Deployer), GSA — General Services Administration (AI Deployer)), which the page had been describing in prose without linking. 8 links added, 3 stale "planned" markers removed.

Queued — foundational sources

• NIST AI 200-2 ipd, "The TEVV-Athlon Framework for Evaluating AI Systems" — standards/nist-ai-200-2 exists as a page but had no Raw Sources/ file; it was built entirely from developments-log coverage, and is now load-bearing on three pages. Verified: nist.gov landing page and nvlpubs.nist.gov PDF, DOI 10.6028/NIST.AI.200-2.ipd. Queued: INGEST-nist-ai-200-2-tevv-athlon.md.
• IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 — the primary text behind the new India page. Verified: MeitY consolidated rules PDF (meity.gov.in, "updated as on 10.02.2026") plus the PIB press release. Queued: INGEST-india-it-rules-amendment-2026.md.
• Agent Plugins Specification v1.0.0 — the primary text behind the new standards page. Verified: agent-plugins.org/specification, corroborated by Vercel, Google's developer blog, The Decoder and a schema URL under the same canonical host. Queued: INGEST-agent-plugins-specification-1-0-0.md.
• Brookings, "Assessing the state of AI adoption across the federal government" — the source of every figure in the expanded page's new quantitative section. Verified: brookings.edu/articles/. Queued: INGEST-brookings-federal-ai-adoption-inventories.md.

Verification records for the first three are in Wiki/_meta/queue/gap-scan/proposed-sources/; the Brookings verification is carried inline in its INGEST- file. No raw files were downloaded this run — all four are large PDFs or living multi-page documents better parsed fresh at ingest time, per the skill's standing exception. None ingested.

Authenticity-verification failures

None. All four pulled sources resolved to canonical hosts. Three weak links are recorded rather than passed over:

• India — missing G.S.R. number. The Gazette notification number was not captured in any retrieved source. The ingest run must recover it before the raw file is treated as fully identified.
• Agent Plugins — no external identifier. A community specification carries no DOI, docket or public-law citation; the identifier trail is the canonical domain, the versioned JSON Schema URL and the public repository. This is why Agent Plugins 1.0.0 is set to medium rather than higher.
• Brookings — no publication date captured. The retrieved text carries no visible dateline, which bounds the freshness of every figure taken from it. Flagged in the INGEST- task as the first thing to recover.

Source-fidelity findings

Two items from the recent window, both recorded under the checks in .claude/skills/source-fidelity/SKILL.md:

1. Send-date-as-event-date, 2026-08-08 22:11 file, India item. The digest presents the IT Rules amendments in the present tense off a Lok Sabha written reply dated 5 August 2026 and gives no notification or commencement date. The instrument was notified 10 February 2026 and in force from 20 February 2026 — roughly six months earlier. The August 5 reply restated existing rules; it did not make them. The page created this run carries the correct dates; the digest item is not corrected in place, and the discrepancy is recorded in the verification record.
2. Staging date versus announcement date, NIST AI 200-2. The nist.gov page states an announcement date of 7 August 2026 while its own metadata carries dcterms.created: 2026-08-04. The August 4 stamp is a staging artifact. Recorded in the INGEST- task so the ingest run does not adopt it as an event date.

Run-order note

This scan ran before the nightly developments-log fold had processed 2026-08-08-2211-ai-developments.md, contrary to the skill's stated run order; bin/lint-scan.py reports unprocessed_devlog=1 for that file. The two gaps drawn from it — India's IT Rules amendments and Agent Plugins 1.0.0 — had no pages and no Raw Sources/ files under any alias, so neither was at risk of duplicating the pending fold; both now exist for that fold to link into. The remaining items in that file (Alibaba's Qwen revenue share, Apple's Qwen guide, the ByteDance pre-training run, the NIST TEVV-Athlon draft, the Google DeepMind reshuffle) all have pages updated within the window and were treated as behind the fold, not as gaps.

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

• The remaining California suspense-file bills, all with zero mainspace mentions and all facing votes on 13 August 2026: AB 1988 (the PAUSE Act, chatbot safety), SB 867 (companion chatbots in toys — the closest fit to Companion Chatbot Harms — Cross-Cutting Analysis, in-degree 22, and to Children's Artificial Intelligence Toy Safety Act of 2026 (S. 5171)), SB 1111 (AI and digital replicas), SB 903 (AI in mental health therapy), AB 2545 (worker impact data assessment), AB 2713 (amendments to the California AI Transparency Act). Score 4–5 each; only AB 412 fit inside the cap. These expire as live-thread gaps on 13 August — whichever clear the suspense file become law-track pages and whichever die become historical ones. Highest priority for the next run.
• analysis/track-record — in-degree 11, confidence: high on sources_count: 1. The confidence-versus-sourcing mismatch flagged as a folder-wide defect on 08-04 and 08-05 recurs in slice 10, though here it may be structural: the page's subject is the record's own resolved forecasts, which have no external source. Worth a decision rather than another deferral.
• analysis/ carries four pages with sources_count: 0 — open-questions (in-degree 6), embedding-retrieval-architecture (3), source-recommendations-2026-04-14 (3), missing-coverage-2026-04-14-residual (3), plus crystallize-session-2026-04-13 (7) and manual-ingest-queue (0). Several are operational artifacts sitting in mainspace: manual-ingest-queue (191 words, in-degree 0), crystallize-session-2026-04-13 and missing-coverage-2026-04-14-residual read as _meta/ material under the v4.3 content/operational split. A placement question, not a content gap.
• analysis/missing-coverage-2026-04-14 — status: superseded but still carrying an in-degree of 10 from live pages. Inbound links should point at the successor.
• MeitY has no entities/ page and Minister of State Jitin Prasada no entities/ page, both now named on the new India page. Per "Where does X go?", a ministry that regulates AI belongs in entities/. Score 3.
• OMB M-25-21, "Accelerating Federal Use of AI through Innovation, Governance, and Public Trust" — named on the expanded Federal AI Adoption — Patterns and Tensions and the successor to omb-m-24-10 and omb-m-24-18, both of which have pages. Score 3.
• Agent Skills specification (agentskills.io/specification) — the second component format Agent Plugins packages, with no page and no alias in concepts/. Score 3.
• entities/samuel-weinbach and entities/ilhan-scheer (in-degree 2 each), entities/clayton-christensen and entities/adam-smith (in-degree 2 each) — carried unchanged for the eighth consecutive run; genuine broken targets, all below the reference-count threshold.
• concepts/inverse-cooking-problem and concepts/inverse-trust-problem (in-degree 2 each) — still parked in needs-review/ as coined terms; not re-actioned.
• [[_meta/briefings/weekly-2026-W19]] referenced from two mainspace pages (AI Macro-Prudential Policy and index.md) — an operational target linked from content, contrary to the v4.3 split. Carried from 08-03.
• The 14 dated dashboard-rebuild-failed notes from June still sit unresolved in needs-review/. Carried.
• Recent-window items already handled by the nightly fold, not gaps: Alibaba's Qwen revenue-share plan, Apple's Qwen guide for Macs in China, ByteDance's 10-trillion-parameter pre-training run, the Nvidia–Lancium investment, the AMD–Taalas acquisition, the Stripe–OpenRouter talks, the Harvey round, the New Mexico v. Meta decree, the Washington State AI task force wind-down, OpenAI's Astra delay and its audit principles, and Adam Cassady's confirmation — each already has a page updated within the window.

Verification pass

A read-only critic reviewed all six pages against house-style/SKILL.md and the schema before the run closed. Twenty-four defects were found and repaired: sources_count understated on two pages (agent-plugins 5 → 8, delangue 4 → 5); a verbatim Delangue quotation carried only a cross-reference where companies/hugging-face already held the CNBC URL, now restored; five instances of "the sources retrieved" and one "as of this page's date" — process language describing the record's own research state rather than the subject — rewritten; the Frontier Security Institute's ## Sourcing and confidence section, which stated curation metadata as article prose ("The page should be revisited"), rewritten as a ## Public record section; three coined framings removed ("visible-disclosure layer / machine-readable traceability layer", "distribution standard rather than a capability standard", "this collection"); one uncited "first federal instrument" superlative softened; and "This page surveys" removed from the expanded analysis lead.

The critic also caught two defects in Federal AI Adoption — Patterns and Tensions that were not introduced this run: "AG inquiries" had been dropped from the lead at some earlier revision and is now restored, and the page rendered GSA OneGov Program and USAi Platform (August 2025)'s "up to $1B in federal IT savings" as "roughly $1 billion in AWS savings" — a modality-and-attribution shift of exactly the kind source-fidelity/SKILL.md names. Corrected to match the source page.

Post-repair bin/lint-scan.py: banned_headers=0 date_headers=28 self_refs=1 predictions=0 header_hype=0 broken_distinct=256 unprocessed_devlog=1. broken_distinct is unchanged from the 2026-08-08 run, confirming no page created or edited this run introduced a broken link. Dashboard rebuilt: 1,885 article pages.

Two style findings were reviewed and not changed, with reasons: the five-bolded-span paragraph in the procurement section and the bare-parenthetical attributions ("(ProPublica, June 2025)", "(WIRED, August 14, 2025)") both predate this run and sit outside the sections touched; repairing them is a lint job, not a gap-scan edit, and doing it here would put un-diffed changes into an expansion. Both are logged here so the next lint pass has them.

One-line summary

Seven gaps actioned: five pages created live (India's IT Rules amendments, California AB 412, Agent Plugins 1.0.0, Clément Delangue, the Frontier Security Institute), the slice-10 anchor Federal AI Adoption — Patterns and Tensions expanded from 758 to 1,630 words with eight stale relationship links repaired, and four foundational sources verified and queued for the user's review — with the India digest item's six-month date error and the missing G.S.R. number both recorded rather than silently carried.