Gap Scan — 2026-08-07

Scanned

Recent window: 4 dev-log files (2026-08-05 08:09 and 22:12; 2026-08-06 08:10 and 22:12) and 31 pages with last_updated in 2026-08-06/07. Rotation slice: 8 — litigation/ + standards/ + government/ (74 pages measured for word count, in-degree, confidence, and sources_count).

Broken-link pass: 259 distinct broken targets across 292 instances wiki-wide. After alias resolution, none with inbound count ≥ 2 resolved to an existing page under another name — the recurring count-2 targets are genuinely-missing author entity pages (Eric Goldman, Orin Kerr, Alondra Nelson, Kevin Klyman, Lee Anne Fennell, Sacha Altay, Joseph Bernstein, Michael J. D. Vermeer, Clayton Christensen, Adam Smith) plus non-AI institutional stubs from entities/imf (Federal Reserve, ECB, Bank of England). No link-alias errors were available to auto-fix this run. The top-ranked "broken target" (wikilinks, 4 inbound) is a false positive from literal bracketed text in provenance notes; three instances were removed as part of this run's edits.

Two alias checks resolved against declaring a gap: SpaceX (53 inbound mentions, no companies/spacex) is fully covered by companies/xai, which documents the February 2026 merger and the July 2026 SpaceXAI rebrand; Terafab is covered at companies/tesla and companies/intel. Neither is a gap.

Gaps actioned (8 of 21 found)

New pages created (live)

• Kids Online Safety Act (KOSA, S. 1748) — highest-scoring gap of the run (score 10). KOSA was named by name across 16 content pages — concepts/reverse-federalism, concepts/state-level-ai-regulation, concepts/algorithmic-speech-doctrine, legislation/chat-act, legislation/youth-ai-privacy-act, legislation/kids-internet-digital-safety-act, legislation/illinois-sb-315, entities/marsha-blackburn, companies/openai, index.md and others — with no page behind it, while the Senate Commerce Committee ordered it reported on August 5, 2026. Built from the Congress.gov record and CRS summary (S. 1748, 119th Congress, Blackburn, introduced 2025-05-14), the committee's own markup announcement and results release, and contemporaneous reporting. Records the duty-of-care collision with the House-passed Kids Internet and Digital Safety Act (KIDS Act, H.R. 7757) and the unresolved 91-3 / 93-1 discrepancy on the 2024 Senate vote.
• CCIA — Computer & Communications Industry Association — dangling reference explicitly marked "create if missing" on entities/chamber-of-progress, named on 7 content pages (concepts/ai-federalism, legislation/illinois-sb-315, legislation/eu-ai-act, legislation/youth-ai-privacy-act, standards/nist-ai-rmf, companies/anthropic), and live in the August 4 Senate Commerce opposition letter. Consolidates CCIA's stated AI-regulation and liability positions, the CCIA Research Center's $600B federal-preemption estimate, the Illinois SB 315 audit-capacity objection, the April 2026 fair-use amicus for Anthropic, NIST agent-security comments, the March 2026 White House joint letter, and CCIA Europe's EU AI Act position.
• Evo 2 (Arc Institute / NVIDIA) — topical hole from the live thread: the August 6 dev-log carried the first complete AI-designed genomes, and the wiki had zero mentions of Evo 2 or the Arc Institute despite holding concepts/ai-biosecurity, concepts/ai-for-science, concepts/open-weight-frontier-models, and models/alphagenome. Covers architecture, training corpus, the open release, the eukaryotic-virus training exclusion, the August 2026 bacteriophage design work, and the Hie / Johns Hopkins Center for Health Security disagreement over open release. [Depends on two queued sources: the Nature Evo 2 paper and the Science phage paper.]

Pages expanded (live)

• ICE — Immigration and Customs Enforcement (AI Deployer) — thin anchor, slice 8: 223 words, in-degree 8, sources_count: 1, confidence: high (unsupportable at one source), untouched since 2026-06-06. Expanded to ~1,600 words across deployment scale, lead generation (Palantir ELITE), field biometrics (Mobile Fortify), tip processing, social-media monitoring, procurement figures, and oversight/litigation. Every prior sourced fact retained, including the OpenAI/Palantir Information report and the predictive-policing criticism. 10 new (Source: URL) cites; sources_count 1 → 11; confidence corrected high → medium.
• Character.AI Litigation (umbrella) — thin anchor with correctness failures, slice 8: 648 words, in-degree 10, sources_count: 0 under confidence: high, untouched since 2026-06-06 while its own child pages moved. The umbrella recorded Garcia as "Active; consolidated discovery" although litigation/garcia-v-character-ai records it as settled in January 2026, and claimed Pennsylvania was the first state-AG-led suit in the cluster when Kentucky filed on January 8, 2026. Rewritten with a private-suits strand and a state-enforcement strand; adds Kentucky v. Character Technologies, A.F. v. Character Technologies, the Google settlement exposure, and the May 1 vs May 5 Pennsylvania filing-date discrepancy. 8 new cites; sources_count 0 → 8; confidence corrected high → medium. An unattributed forecast paragraph (pseudo-cited as "M.D. Fla. proceedings, 2024-10" and "pattern of pre-verdict AI-tort settlements, 2026-05") was removed as retired-## Predictions residue, with its one resolved forecast recorded as fact.

Link aliases fixed

• [[entities/ccia]] on entities/chamber-of-progress — target created this run; the "create if missing" scaffolding was cleared (1 instance). [[entities/netchoice]] on the same line remains open and is deferred below.

Queued — foundational sources

• "Generative design of bacteriophages with genome language models" (King, Driscoll, Li, Guo, Merchant, Brixi, Wilkinson & Hie), Science 393(6811), 6 August 2026 — the primary text behind Evo 2 (Arc Institute / NVIDIA)'s genome-design and biosafety sections and behind the live AI-biosecurity thread. Verified: AAAS DOI landing page, 10.1126/science.aec2657, HTTP 200; title, 8-author list, date, publisher and abstract opening confirmed; authors independently corroborated by Stanford Report, findings by Guardian and BBC. Saved: URL-only (AAAS paywall — full text to be fetched at ingest). Queued: INGEST-king-hie-generative-design-bacteriophages-2026.md. Verification record: queue/gap-scan/proposed-sources/king-hie-generative-design-bacteriophages-2026.md. Flagged for ingest: related DOI 10.1126/science.aej8512, likely the Inglesby / Hanke commentary.
• "Genome modelling and design across all domains of life with Evo 2" (Brixi et al., 60+ authors), Nature 652(8112):1349–1361, published online 4 March 2026 — the technical paper behind Evo 2 (Arc Institute / NVIDIA) and the source for the single stated biosafety control. Verified: Springer Nature landing page, 10.1038/s41586-026-10176-5, HTTP 200, open access; volume, issue, pagination, online date and abstract passages confirmed; corroborated by the Arc Institute release and Stanford Report. Saved: URL-only (open-access PDF to be fetched at ingest). Queued: INGEST-brixi-hie-evo-2-nature-2026.md. Verification record: queue/gap-scan/proposed-sources/brixi-hie-evo-2-nature-2026.md. Flagged for ingest: the Evo 1 predecessor, Nguyen et al., Science 386 (2024), DOI 10.1126/science.ado9336.

Authenticity-verification failures

None. Both pulled sources verified against the publisher's own DOI landing page with independent corroboration of authorship and findings.

One retrieval failure, not an authenticity failure: ccianet.org/about/ returned HTTP 404, so CCIA's founding date, governance, leadership, and full membership roster could not be sourced. Rather than assert them from background knowledge, they are recorded as an open question on CCIA — Computer & Communications Industry Association and the page is held at confidence: medium.

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

• DHS — Department of Homeland Security (AI Deployer) expand — 385 words, in-degree 17 (the highest in-degree thin page in slice 8), untouched since 2026-06-06; the ICE expansion this run supplies most of the material. Score 3.
• entities/netchoice — dangling with an explicit "create if missing" marker; litigation/netchoice-v-bonta exists with no organization page behind it. Score 3.
• entities/consumer-technology-association — live thread (August 6 FTC AI-policy-statement comment seeking a good-faith safe harbor and federal preemption), but only 1 inbound reference. Score 3.
• Defense Innovation Unit (DIU) expand — 483 words, in-degree 13. Score 2.
• Five Eyes Joint Guidance on Secure Deployment of AI Agents (May 2026) expand — 511 words, in-degree 12. Score 2.
• Taiwan Ministry of Digital Affairs (moda) expand — 586 words, in-degree 11, sources_count: 2, untouched since 2026-06-06. Score 2.
• DOT — Department of Transportation (AI Deployer) expand — 397 words, in-degree 7, untouched since 2026-06-06. Score 2.
• NSF — National Science Foundation (AI research funder) expand — 262 words, confidence: low, sources_count: 1; the NSF 26-513 State and Regional AI Infrastructure Hubs program ($100M, 10 hubs, proposals due November) landed August 6 and the page is still a stub. In-degree 0. Score 2.
• Cloudflare Agent Access Model (Matt Silverlock, August 5, 2026) — proposes a named framework (short-lived task-scoped sender-constrained credentials on RFC 8693 / RFC 9449, plus a "Trust Ratchet"); borderline foundational. companies/cloudflare already folded it as a supporting source. Score 2.
• CBP — Customs and Border Protection (AI Deployer) expand — 199 words, the thinnest page in slice 8; the Clearview AI contract and Mobile Fortify material gathered this run applies directly. Score 2.
• entities/koray-kavukcuoglu — named as incoming Google DeepMind lead (August 5, 2026); companies/google-deepmind and entities/demis-hassabis were both updated August 6 without an entity page. Score 3.
• entities/brad-carson (Americans for Responsible Innovation president, quoted on the classified White House AI framework) — parent org page exists. Score 1.
• Missing source-author entity pages, ranked by inbound count (all 2): Eric Goldman, Orin Kerr, Alondra Nelson, Kevin Klyman, Lee Anne Fennell, Sacha Altay, Joseph Bernstein, Michael J. D. Vermeer, Clayton Christensen. Score 1 each.
• litigation/runlayer-v-rippling (180 words, confidence: low, sources_count: 1) and litigation/ani-v-openai (181 words) — thinnest litigation pages in slice 8, both in-degree ≤ 1. Score 1.

Note for the next cycle

bin/lint-scan.py re-run after this scan's edits: banned headers 0, standing Predictions sections 0, header hyperbole 0, wiki self-references 1 (the deliberate index.md lead), dated section headers 28 (unchanged), broken distinct wikilinks 341, down from 342 despite five substantially expanded pages — the entities/ccia creation and the literal-bracket cleanup more than offset the new outbound links.

One item outside this skill's scope: New Developments Log/2026-08-06-2212-ai-developments.md is unprocessed — no matching Operation: Developments-Log entry in the audit log. Gap-identifier does not fold dev-log files; this run read it for the recent-thread scan only. The August 6 22:12 items (OpenAI's Black Hat disclosure on agent coordination through Artifactory, the classified White House AI oversight framework and its open-model exclusion, the Gemini 3.5 Pro cancellation, the Replit agent-deployment figures, the Weil strict-liability argument) still need the developments-log pass.

One-line summary

Three new pages created live (KOSA, CCIA, Evo 2), two stale high-in-degree anchors expanded and their confidence ratings corrected downward, one alias cleared, and two peer-reviewed primary papers verified and queued for review — the Character.AI Litigation (umbrella) rewrite is the one that needs the user's eye, since it corrects two substantive errors the umbrella had been carrying since June.