Gap Scan — 2026-09-06

Scanned

Recent window: 4 New Developments Log/ files covering 2026-09-04 and 2026-09-05; 74 wiki pages with an mtime inside 48 hours. Rotation slice: 10 — analysis/ (all), 25 pages.

bin/lint-scan.py and bin/primary-doc-scan.py --days 3 were both run rather than re-derived. Lint: banned_headers=0 date_headers=26 self_refs=6 predictions=0 header_hype=0 broken_distinct=212 unprocessed_devlog=1. Primary-doc scan: 6 digests, 8 cited primary-document URLs, 3 with no Raw Sources/ file, 0 possible matches under another address, 4 keyword pre-filter hits.

Broken-link lane is empty again. Of 212 distinct broken targets, 0 are referenced 2+ times; all 212 are the single-reference long tail the skill excludes from candidate generation. This is the fifth consecutive run with an empty referenced-2+ lane. Reported without re-auditing the tail, per the settled finding in the Autonomy rules.

Gaps actioned (9 of 24 found)

New pages created (live)

• X.AI LLC v. Ellison (challenging Minnesota's nudification-technology statute) — score 8 (live thread +3, uncaptured primary document +3, wiki core area +2). The case was tracked in prose on xAI and Minnesota HF 1606 (Nudification Technology Prohibition) but had no litigation/ page, against a folder of 60 case pages that already includes the parallel xAI LLC v. Weiser (challenging the Colorado AI Act) and SpaceXAI v. Bonta (AB 2013 challenge). Alias resolution found nothing under ellison, nudification, minnesota or xai in litigation/. Built primarily from the September 4 memorandum opinion captured this run, which supplies the docket number (0:26-cv-03425-DWF-DTS), the judge, the July 31 TRO denial, the reasoning, and the fact — absent from the digest — that xAI had already disabled its nudification tool in Minnesota, which is part of why the monetary penalty was held recoupable. confidence: medium, 5 sources. Depends on queued source: Memorandum Opinion and Order, X.AI LLC v. Ellison.

Pages expanded (live)

• Grok CSAM suits (Doe plaintiffs v. X / xAI) — score 5. Carried confidence: low with an explicit note that "party names, courts, and filing dates not yet verified against dockets." The Ellison opinion, a primary court document, identifies three of the actions by caption, docket and venue: Doe 1 v. X.AI Corp., No. 5:26-cv-2246 (N.D. Cal., amended July 7, 2026), and Does v. X.AI, Nos. 4:26-cv-750 and 4:26-cv-772 (E.D. Ark., both amended August 17, 2026). Added as a table with an explicit statement that the correspondence to the March/July/August reported suits is not asserted. Also folded: the SpaceX $500m+ litigation reserve for Grok Imagine's "Spicy" and "Unhinged" modes, and xAI's own account-suspension and NCMEC-report figures from its Minnesota complaint. Raised low → medium; sources_count 3 → 4; 2 new typed relationships; court frontmatter updated from "venues per individual suit" to name the two identified districts. 1 new source, 2 new citation instances. 601 → 818 words.
• US AI Regulatory Approaches Compared — score 2 (in-degree 34, the highest in the slice). 1,930 words and 20 sources, so not thin in body; it had no ## Relationships section, leaving the most-linked page in the slice with no typed outbound edges. Added 11 typed links. No prose changed.
• EU vs. US AI Regulation: A Deep Comparison — score 2 (in-degree 11). Same defect; added 8 typed links. No prose changed.
• Track Record — score 2 (in-degree 10). Carried confidence: high on sources_count: 1, which the decay table permits for no claim type, unreinforced since 2026-07-25, and no ## Relationships. Lowered to low, added 4 typed links, and added a ## Schema status section recording the finding below. The roll-up and per-source sections were not altered.

Link aliases fixed

None. Fifth consecutive run of zero, and consistent with the empty referenced-2+ lane — the remaining broken targets are author surnames, institution shorthand and deliberate forward references.

Queued — foundational sources

• Memorandum Opinion and Order, X.AI LLC v. Ellison (D. Minn., Sept. 4, 2026) — score 6. Lane-1 hit from the 2026-09-05 08:05 digest; the instrument the day's litigation item is about. Verified: Thomson Reuters legal-documents mirror, provenance flagged (no RECAP copy of Doc. 54 located), corroborated by a byte-identical copy at Courthouse News, by two independent reports of the docket number and filing date, by Reuters on the disposition and quoted language, and by the Minnesota Attorney General's own announcement. Saved in full: Raw Sources/X.AI LLC v. Ellison - Memorandum Opinion and Order Denying Preliminary Injunction (D. Minn., 2026-09-04).md (14 pages, complete extracted text). Queued: INGEST-xai-v-ellison-pi-order-2026.md.
• Microsoft memorandum in support of summary judgment, Authors Guild v. OpenAI, Inc. (S.D.N.Y., Sept. 4, 2026) — score 8. Lane-1 hit from the 2026-09-05 22:04 digest. Verified: RECAP/CourtListener, docket gov.uscourts.nysd.606655 entry 1211, 42 pages; first four pages retrieved and identity confirmed against The Verge's account of the same filing. URL-only — see the capture compromise note below. Queued: INGEST-microsoft-msj-memorandum-authors-guild-2026.md.
• Prove2Me: An Open Collaborative Platform for Scaling Math Formalization (arXiv:2608.28433) — score 6. Named under ## Primary documents in the 2026-09-05 08:05 digest as the platform behind the Fermat formalization; no raw file, no source page. Verified: arXiv, the canonical host; all five authors, the identifier and the arXiv-issued DataCite DOI match the digest's citation exactly. URL-only, current version fetched at ingest. Queued: INGEST-prove2me-formalization-platform-2026.md.
• Anthropic, "Formalizing Fermat's Last Theorem" (Sept. 4, 2026) — score 6. Same digest, same lane. Verified on Anthropic's own domain; title and content match. URL-only. Queued: INGEST-anthropic-formalizing-fermats-last-theorem-2026.md. Should be ingested together with the Prove2Me paper — one is the application, the other the platform.

Routed to needs-review

• Track Record depends on a section type schema v4.3 retired. The page's entire intake mechanism is defined against standing ## Predictions sections: it specifies their six-column table and has lint scan them each cycle for resolved entries. v4.3 (2026-06-06) retired that section type, and this run's lint reports predictions=0 wiki-wide. The page has therefore described a live process with no input for three months, and its ## Maintenance claims about lint and weekly-synthesis cannot be true as written. Three dispositions are plausible — retire, rebuild the intake against attributed inline forecasts, or freeze as a historical record — and two of them touch SKILL.md files, which under the multi-machine rule must ship alone. Recorded at Wiki/_meta/queue/gap-scan/needs-review/2026-09-06-track-record-vs-v4.3-predictions-retirement.md. The mechanical fixes (confidence, Relationships) were made; the disposition was not.

Authenticity-verification failures

None. Four documents verified, four passed. One provenance flag recorded rather than a failure: the Ellison order was reachable only through mirrors, not through PACER or RECAP.

Dedupe drops (all correct)

• US statement of interest at gov.uscourts.nysd.640396.1682.0.pdf — reported missing by lane 1, but this is the second docket copy of the statement of interest already held as Raw Sources/Statement of Interest of the United States - In re OpenAI Copyright Infringement Litigation (DOJ, 2026-09-01).md and summarized at Statement of Interest of the United States — In re OpenAI, Inc. Copyright Infringement Litigation (September 1, 2026), captured from the 612697 address. Same document, second docket — the known false-positive shape recorded in the v4.15 changelog.
• Brinkema remedies memorandum (lane-2 keyword hit, 2026-09-03 digest) — open INGEST- task since 2026-09-03, not_before: 2026-09-16 while the opinion is sealed.
• Seattle Times / Newsday complaint (lane-2 hit) — The Seattle Times Co. and Newsday LLC v. OpenAI and Microsoft and its source page were both created by the nightly fold on 2026-09-05.
• Two further lane-2 keyword hits read and dropped: the White House polysilicon proclamation, which the Wacker Chemie item quotes but is not about, and the GPT-6 Astra system card, which the evaluation-revision item cites at section level while being about the announcement page and its Internet Archive snapshots. Neither is an instrument the item is about.

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

• Live-thread items owned by the nightly fold, not by this skill (score 3 each). The 2026-09-05 22:04 digest is unprocessed at scan time — the fifth consecutive report of an unprocessed evening digest, now unambiguously a standing condition rather than a run of coincidences. Its page-level work belongs to developments-log: the GPT-6 Astra post-publication benchmark revisions on Astra; Microsoft's summary-judgment motion on Authors Guild v. OpenAI and NYT v. Microsoft, OpenAI et al.; Project Zenith on Microsoft; the AI BioDesign initiative; and the Chapel Hill G20 protest material for Data Center Siting / AI Power Politics. From the 09-05 morning file: the OpenAI "wiki incident" acknowledgment, the Anthropic IPO timing shift, the US–China mid-September safety talks, the CCW autonomous-weapons consensus text, the TCS/HyperVault Telangana campus, and the Wacker Chemie polysilicon item. All were left alone. The four primary documents behind those items were captured or queued instead, which is this skill's lane and does not collide with the fold.
• CCW autonomous-weapons consensus document, Geneva, September 5, 2026 (score 5), new this run. The 128 states party agreed a non-binding text that could open the door to treaty negotiations. Autonomous Weapons exists and was edited inside the window; there is no entities/ page for the CCW or its Group of Governmental Experts, and no legislation/ page for the text. Deliberately not created, per the quality gate: the digest carries the outcome entirely through Reuters and a Stop Killer Robots quotation characterizing the definition and civilian-harm measures as "substantially diluted," and no copy of the agreed text was sought this run. A page built from one wire report plus one advocacy characterization would inherit that framing. The text itself is the right first capture.
• entities/keith-ellison (score 3), new this run. Named on at least three content pages and now the defendant of record in a tracked case. Not created: no first-party material was pulled this run.
• models/grok-imagine (score 3), new this run. The product at issue in X.AI LLC v. Ellison and in the CSAM suits, reached across the vault only through Grok (xAI) and litigation pages. A model page needs capability and release material this run did not gather.
• analysis/ slice-10 remainder — the healthiest slice audited to date. The standard diagnostic returns 5 of 25 pages, against 5 of 87 in slice 9, 7 of 92 in slice 8 and 84 of 198 in slice 4; on a per-page rate the slice is mid-pack, but no page in it is word-thin, and every flag was a missing ## Relationships section or a confidence rating rather than absent content. Four of the five were worked. Remaining: Missing Coverage: Companies, People, Models, Legislation, Concepts (revised) (in-degree 9, sources_count: 0, confidence: high, last_updated: 2026-06-06) and Crystallize Session: Wiki Gaps & User Knowledge Map (2026-04-13) (in-degree 6, 612 words, sources_count: 0). Both are session artifacts from April 2026 that sit in mainspace — working records of a crystallize session and a coverage audit, not encyclopedia pages. Whether they belong in analysis/ at all, or under _meta/, is the same content/operational question v4.3 settled for logs and briefings and did not reach for these. Raised here rather than actioned; it is a placement decision.
• The confidence: high on thin sourcing pattern — fourth consecutive report, now confirmed in a sixth folder. Named 2026-09-03 as fifteen legislation/ pages; found in litigation/ and government/ on 09-04; in models/ on 09-05; and here in analysis/ (Track Record, lowered this run; Missing Coverage: Companies, People, Models, Legislation, Concepts (revised), not reached). Six folders is not a per-page miss. A single confidence-audit pass would close more of this than one page a day, and this skill's cap structurally cannot get there. Recommend it be scheduled as its own operation.
• models/deepseek-v3 — sixteenth run. The 2026-09-05 report recommended removing it from this backlog and running model-robustness-check on it directly, having established that it passes every mechanical thin-anchor test and that its problem is depth against importance, which this skill does not measure. That recommendation stands and is repeated once here; it is listed for the last time.
• entities/adam-thierer (score 3) — ninth consecutive deferral, structural.
• Three slice-3 pages, eighth deferral — structural: Standards-Industrial Complex, Three Theories of Victory (US / EU / China AI Governance), Transaction Costs (Coase) and the Agentic Economy. Eight runs without reaching the top ten. They need a direct instruction or a decision that they are not gaps.
• Slice carryover. Slice-4 (sixth): American Civil Liberties Union (ACLU), Access Now. Slice-5 (fifth): Helen Toner, International Monetary Fund (IMF), Henry Farrell. Slice-6 (fourth): Stanford Center for Research on Foundation Models (CRFM), Robert Chesney, Sen. Elizabeth Warren. Slice-7 (fourth): the fifteen high-confidence-thin-sourcing legislation/ pages. Slice-8 (third): Getty Images v. Stability AI (US), ISO/IEC 42005 — AI Impact Assessment, Tremblay v. OpenAI, GEMA v. Suno (Munich Regional Court), Sony Music v. Udio (second suit, 2026) — the last two still need a docket check rather than a web pass, and the CourtListener MCP is unauthenticated in this session for the fourth consecutive run, which is now a standing capability gap rather than a transient one. Slice-9 (second): Helix (Figure AI Vision-Language-Action model), S1 (Skild AI), Llama 4 (Meta), Llama 3 (Meta).
• Twenty-one source-robustness-check candidates across slices 11–13 — fourteenth consecutive report, structural. 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.
• Date-stamped headers in lint-report.md (score 1) — twenty-first cycle, structural, holding at 26. Six self-references, unchanged.
• Carried unchanged. entities/nightingale; Sydney Von Arx, Cormac Slade Byrd, Spencer Kitts, Thomas Larsen; legislation/caadca (the CAADCA statute itself, reached only through the litigation challenging it); entities/david-feith and entities/hudson-institute; Marvell (no company page, since 2026-08-21); California AB 1856; the Texas OAG / Meta Webster–Mahoney exchange; the draft executive order on a frontier-developer self-regulatory organization; the AI Evaluator Forum / AEF-1 standard; ExploitBench flag-level disclosure; the "Carolina Principles"; OpenAI's "Our agreement with the Department of War"; the House E&C markup memo; the METR/Redwood September 1 Hugging Face analysis; the New York City Public Schools generative-AI policy; the Agentic Commerce Protocol / Model Context Protocol placement question; OpenAI's August 31 response to Rep. Casar (Google Drive host, curator decision); the July 2026 statement from more than 1,000 AI-firm employees.

Run integrity

• No broken outbound links introduced. All five touched content pages and all five queue files were resolved against the full page set after writing: zero unresolved on every one. Two hits were caught and fixed during that check — both were the literal word wikilinks written inside a bracketed example in a provenance note, which the resolver correctly reads as a link target. Rephrased so neither parses as a link.
• Style scan clean, second pass. All five touched content pages checked programmatically against the banned-header, hyperbole, self-reference, date-led-header and standing-Predictions patterns: zero hits on each. Note that Track Record now contains the string "Predictions" in prose but carries no standing section of that name.
• Revision fidelity verified mechanically. Grok CSAM suits (Doe plaintiffs v. X / xAI): all 4 original source URLs, all 4 original wikilinks and all 8 spot-checked original figures and dates present after revision; 601 → 818 words; nothing removed. The three analysis/ edits were additive except for two frontmatter values (last_updated on all three, confidence lowered on one); no prose was deleted from any of them.
• Confidence set honestly, and lowered once. The new page is medium on 5 sources, not high, despite resting on a primary court document — a fresh page does not meet the independent-3-source bar. Grok CSAM suits (Doe plaintiffs v. X / xAI) rose low → medium on primary-sourced docket identification, with a rewritten confidence note stating exactly what remains unverified. Track Record was reduced high → low.
• A contradiction was recorded rather than silently resolved. The opinion cites Minn. H.J. 6139-40 (2026) for a Minnesota House vote on April 23, 2026; Minnesota HF 1606 (Nudification Technology Prohibition) records April 28, 2026 at 132–1 from CBS News Minnesota. Both appear on the new page, unreconciled, with the journal citation identified as the primary reference. The legislation page was not edited — correcting it belongs to the ingest that reads the opinion in full.
• A source-fidelity flag was raised on a queued item. The digest describes Prove2Me as designed by "Anthropic researcher Tianyi Peng and collaborators at Columbia University." The arXiv abstract page lists no affiliations, so the primary record confirms neither attribution. Recorded in the ingest task rather than repeated as fact.
• Cap respected. Nine gaps actioned against a 6–10 cap; 24 candidates found. Five scrapes and one search; no crawls; no failures.
• One capture compromise recorded. Three of the four queued documents are URL-only rather than saved in full: the 42-page Microsoft memorandum, because this sandbox could not read back retrieved text at that size, and the two Fermat-related documents, which are better fetched current at ingest time. All three carry a complete verification record and the sanctioned large-document pattern applies, but the first is a capability limit rather than a judgment call and is stated so it is not mistaken for one.

One-line summary

One litigation page created and four pages expanded live from a primary court document that also supplied docket identification for a second case; four foundational sources verified and queued; and Track Record found to depend on a section type schema v4.3 retired three months ago, routed to the curator with the mechanical fixes made and the disposition left open.