Gap Scan — 2026-09-16

Scanned

Recent window: 4 New Developments Log/ files (2026-09-14 08:09 and 22:05; 2026-09-15 08:05 and 22:05) and 4 content pages edited in the last 48 hours (Overview, Coordinated slowdown proposals compared, Machine-produced mathematics: claims, verification and credit, Legal Services — AI Deployment). Rotation slice: 6 — entities/ O–Z, 135 pages.

bin/lint-scan.py: 211 distinct broken wikilinks, 0 referenced 2+ times, 211 referenced once. The referenced-2+ candidate lane has now been empty for eight consecutive scans; per the standing rule the single-reference long tail was not re-audited. bin/primary-doc-scan.py --days 3: 6 digests, 4 cited primary-document URLs, 1 with no raw file, 2 possible match under another address, 1 keyword-lane hit.

34 candidates generated · 8 dropped at dedupe or alias resolution · 26 surviving · 7 actioned · 19 deferred.

Gaps actioned (7 of 26)

New pages created (live)

• Collaboration on Adversarial Threats and Security Risks Act (S. 5105) — score 9 (live thread +3, uncaptured primary document +3, US AI regulation core +2, lint/crystallize backlog +1). Gap types 1, 3 and 9 together, and the item the 2026-09-15 report called "the week's most consequential live legislative thread … blocked on identifiers, not on judgment." Both identifiers were found on the first search: S. 5105, 119th Congress, introduced July 23, 2026 by Sen. Schiff for himself and Sen. Banks, referred to Senate Judiciary, no action since. The page is built from the bill text rather than from coverage: all nine definitions in section 2, both limbs of the section 3 exemption, the four bounds on it, and section 4's injunctive-relief carve-out. confidence: medium on 4 sources, a > **Provenance note:** recording that citations upgrade once the queued ingest runs, three ## Open questions, and six typed relationships including a contradicts: edge carrying the Khan and Calidas objections rather than narrating them.

The date is the substantive correction. The 2026-09-14 22:05 digest carries the bill in a September news cycle in the present perfect ("have introduced"), with no date. It was introduced seven weeks earlier. A page written from the coverage would have placed it in the week of the pacing debate rather than seven weeks ahead of it — which inverts the causal story, since the instrument predates the proposal it is now being read against.

• Protect Democracy — score 5 (live thread +3, US AI regulation core +2). Gap type 3. The organization is named on AI Pre-Release Vetting and Office of Science and Technology Policy (OSTP) with no page behind it, and it is the sole external source of information about the administration's undisclosed frontier-model review framework. Built from the organization's own case page and the Ars Technica account of the complaint: the four defendant agencies, the September 1 filing, the September 30 production date sought, the GOLD EAGLE allegations, the undefined term "covered frontier model," and the CISA-extension argument. confidence: medium on 4 sources, three ## Open questions, four typed relationships.

Alias resolution first: sources/who-decides-ai-war-hai is an unrelated HAI source and does not cover this organization; no entities/ page covers it under another name.

Pages expanded (live)

• John Thune — score 6 (in-degree 9 +1, US AI regulation core +2, +3 for the specific staleness below). The page stopped at August 3, 2026 and asserted a two-senator negotiation. By September the negotiation had three principals — Thune, Klobuchar and Cruz — with Cantwell weighing in, and the fight had moved from whether to legislate to who runs the tests. Added a ### Negotiations and committee timetable subsection carrying the September 11 Reuters and Politico accounts of the impasse, the Nextgov account of the self-testing dispute and the "primarily a voluntary standard type situation" characterization, the reported state-preemption clause, the cancelled pre-recess markup and Cruz's stated reason, the September 23 markup and its September 16 notice deadline, the likely inclusion of the Young–Cantwell bill on the roster, Speaker Johnson's September 17 adjournment intention, and the 2023 Thune–Klobuchar precedent. 2,452 → 6,837 body chars; sources_count 3 → 8; 10 new cites; two typed relationships added including contradicts: [[entities/maria-cantwell]], qualified in place as an open disagreement inside one negotiation rather than a settled split. confidence held at medium — the draft is unpublished and every account of its contents is from people familiar with the negotiation. Nothing removed.

• Maria Cantwell — score 7 (in-degree 5, US AI regulation core +2, live thread +3, +2 for the gap below), and the highest-scoring expansion candidate carried into this run: the 2026-09-15 report deferred her and Thune explicitly, on the ground that the digest carrying the thread was unprocessed and the nightly fold owned it first, and named them "the first expansion candidates next run." The fold has since landed on that digest. The page recorded her position abstractly — "a pre-deployment regime under which the government would vet frontier models" — and did not have the September specification: testing by scientists and experts at the national laboratories, in her own words on X. Added that, a ## Role in the 2026 Senate negotiation section giving the precise element she objects to (company self-testing with results presented to the Commerce secretary for deployment approval) and the state of her counter-language, and a ## Future of AI Innovation Act section on the Young–Cantwell measure that would give CAISI a statutory basis and testbeds at DOE, NSF and NIST — the institutional counterpart to her testing position, which the page had no record of at all. 2,264 → 5,622 body chars; sources_count 3 → 6; 5 new cites; two typed relationships added. confidence held at medium.

• Robert Chesney — score 4 (in-degree 9 +1, national-security core +2, +1 backlog). The slice's one thin anchor under 1,800 characters, and a thirteenth-deferral carryover. Two additions, one a correction. (1) The page described him as "a professor at the University of Texas at Austin School of Law"; Lawfare's own contributor page and UT Law's news page both give Dean of the University of Texas School of Law. (2) More consequentially, the page did not name his single most AI-relevant work. Added a ## Deep-fakes scholarship section on Chesney and Citron, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security (SSRN 3213954, posted 2018-07-14; California Law Review), which introduced the liar's dividend — a term with its own page at Liar's Dividend that the Chesney page never linked to, in either direction. Also recorded that the paper predates the generative-model wave and was written about manipulated audio and video, so later citations extend rather than apply it. 1,254 → 2,843 body chars; sources_count 2 → 6; 5 new cites; two typed relationships added including supports: [[concepts/liars-dividend]]. confidence held at medium.

Link aliases fixed

Zero, and not for want of looking. lint-scan.py reports 0 targets referenced 2+ times for the eighth consecutive scan. This is the settled finding recorded since 2026-08-18 and was not re-derived.

Queued — foundational sources

• S. 5105, Collaboration on Adversarial Threats and Security Risks Act, introduced text (2026-07-23) — score 9. Verified: congress.gov, HTTP 200, Dublin Core metadata self-corroborating (dc.number S.5105, dc.creator Sen. Schiff, Adam B. [D-CA], dc.coverage 2026-07-23), independently corroborated by the GovInfo record BILLS-119s5105is and by both sponsors' press releases of the same day. Captured complete — sections 1 through 4, every definition and subsection. Saved: Raw Sources/Collaboration on Adversarial Threats and Security Risks Act, S. 5105, 119th Congress (Introduced in Senate, 2026-07-23).md. Queued: INGEST-catsr-act-s5105-2026.md.

Six source-fidelity divergences recorded in the raw file. The two that most change a claim: (1) the digest states the bill "would require labs to give the attorney general advance notice of frontier-pacing plans" — section 3(a)(2) requires notice to the Assistant Attorney General in charge of the Antitrust Division, a different officer; the Attorney General appears separately in section 4 as the officer who may seek an injunction. (2) The rule of construction preserving the prohibitions on price-fixing and market allocation is drafted against subsection (a)(1) only — the information-sharing limb — not against (a)(2), the coordinated-delay limb, which is instead bounded by the notice requirement, the exclusive-purpose test and the affirmative-defense burden. Endorsement statements collected in the press releases describe those safeguards as applying to the exemption generally.

• Complaint, UMG Recordings, Inc. v. DistroKid, LLC, No. 1:26-cv-01156 (D. Del., 2026-09-15) — score 6 (live thread +3, uncaptured primary document +3). Gap types 3 and 9. The 2026-09-15 22:05 digest carries the suit through one trade account citing no primary-document URL, so primary-doc-scan.py surfaced it only in the keyword pre-filter lane. Verified through the CourtListener MCP: docket 74792008, RECAP document 493638703, PACER path gov.uscourts.ded.94617, complaint 52 pages with a 36-page Exhibit A–B set, jury demand, assigned to Judge Maryellen Noreika, answers due October 6, 2026. Distinctive passages confirmed present in the fetched document. Corroborated independently by five accounts published September 15. Queued with the verified URL rather than transcribed, per the large-PDF rule: INGEST-umg-v-distrokid-complaint-2026.md.

Three corrections to the coverage are recorded in the ingest task. The complaint's lead volume example is "Lofi Chill," at 4,562 tracks in a single twelve-month period — a figure no vault-held source carries; the coverage leads with "Chill Flow Radio" and "Mellow Vibes Radio" and says "thousands." The deception count is pleaded about Delaware consumers and artists specifically, which is the likely explanation for the forum and which the coverage omits. And the false-representation theory is tripartite, the second limb resting on the Music Fights Fraud Alliance, an organization named in the complaint and absent from the wiki entirely.

• State Bar of New Mexico Ethics Advisory Opinion on generative AI, Bar Bulletin Vol. 63, No. 11-D (2024-11-27) — score 5, second deferral, now closed. The only precise-lane hit from primary-doc-scan.py — the one cited primary-document URL across three days of digests with no Raw Sources/ file. Verified: sbnm.org, the State Bar's own domain, HTTP 200, application/pdf, 54 pages, masthead and ISSN self-corroborating, and the opinion's location confirmed twice within the file (cover contents and table of contents both give page 13). Queued rather than captured: the opinion occupies roughly pages 13–16 of a 54-page bulletin that also carries two unrelated New Mexico Supreme Court opinions, and a whole-file capture would put ~50 pages of notices and advertising into Raw Sources/ under an AI-ethics title. Verification record: Wiki/_meta/queue/gap-scan/proposed-sources/nm-state-bar-ethics-advisory-opinion-generative-ai-2024.md. Queued: INGEST-nm-state-bar-ethics-advisory-opinion-ai-2024.md, with the page range as an explicit capture instruction.

Authenticity-verification failures

None. Three sources were pulled or verified and all three passed.

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

New this run

• litigation/umg-v-distrokid (score 6). The case page, as distinct from the complaint capture actioned above. Deferred on the standing ground that the 2026-09-15 22:05 digest is unprocessed and the nightly developments-log fold owns the litigation folder first; the queued ingest will also want to create it. If neither has by tomorrow, this is the first new-page candidate.
• litigation/protect-democracy-v-oncd (score 5). Protect Democracy Project v. Office of the National Cyber Director, No. 1:26-cv-03064. The entity page created today records the case in prose; the "Where does X go?" rules put an individual action in litigation/. Deferred on the cap, not on merit — and the CourtListener connector working again (below) makes the docket retrievable for the first time.
• Citron & Chesney, Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security (score 5). Gap type 1, surfaced by today's Chesney expansion. Liar's Dividend carries depends-on: Citron & Chesney 2018/2019 (**not in wiki**) — origin in its own relationships section, which is a dangling foundational reference the page states about itself. Canonical hosts located: SSRN abstract 3213954 and the Boston University institutional repository (scholarship.law.bu.edu/faculty_scholarship/640). A high-confidence concept page resting on an unheld origin text.
• entities/who-decides (score 5). The nonprofit Alex Bores launched September 15, 2026 with Anna Myers, having raised $10 million and targeting $20 million more in 2027, working through organizations in 11 battleground states. Alex Bores exists; the organization does not. Deferred: it rests solely on the unprocessed 2026-09-15 22:05 digest.
• entities/center-for-american-progress (score 4) and entities/bilal-chughtai (score 3). CAP called for a US–China frontier-AI pacing agreement at an upcoming ministerial and a Trump–Xi summit; Chughtai resigned from Google DeepMind on September 14 and is carried in two consecutive digests. Neither has a page.
• entities/simon-willison lacks ## Relationships at in-degree 7 (score 3). The only page in the 135-page slice above the in-degree-6 bar without the section — 1 of 80. Three consecutive slices have now returned essentially complete graph wiring.
• Four further slice-6 single-source pages at in-degree ≥6 (score 2–3 as a set). Sanjog Misra (in-degree 9), Project Liberty Institute (7), Rakshit Trivedi (7), Sir Stephen Fry (7), Stewart Baker (6) — all medium on sources_count: 1. All five carry last_updated: 2026-06-06, and 41 of the slice's 135 pages carry that same date. This is the fourth consecutive slice to produce the signature the 2026-09-13, -14 and -15 reports identified; it is one import, not dozens of independent gaps, and it will keep generating candidates until it is treated as one job.
• Nine slice-6 pages at confidence: high on sources_count ≤ 1 (score 2 as a set). Packy McCormick, Ray Kurzweil, Reid Hoffman, Rishi Bommasani, Ross Douthat, Sen. Chuck Grassley, Stanford Cyber Policy Center, Toby Ord, Tyler Cowen — 9 of 135, against 7 of 148 in slice 5 and 5 of 205 in slice 4. The pattern outside sources/ is now confirmed in three consecutive slices and is trending up rather than holding.

Carried forward

• Young–Cantwell "Future of AI Innovation Act" (score 5) — second deferral, half-closed. The Schiff–Banks half of the 2026-09-15 pair was closed today; this half was not. No bill number or text URL has been located, and the measure is named in coverage only. A June 2026 commentary identified S. 4178 of the 118th Congress with a 119th-Congress successor at S. 3952, which the next run should test directly against congress.gov rather than re-searching generally — the same instruction that worked for S. 5105 today.
• DeepSeek-V4.1-Flash technical report (score 5) — third deferral. Reported by primary-doc-scan.py as "possible match under another address"; the cited Hugging Face blob path has no raw file, and Raw Sources/ holds the V3 technical report but nothing for V4.1. The possible-match state needs a human look, not another scan.
• Anthropic August 2026 Risk Report (score 3) — recurring false positive, stated once. The second "possible match" hit. The vault holds it under its canonical landing page while the digest cites the CDN PDF path; this is the exact case schema v4.15 documented when the scanner was written. Not a gap.
• H.R. 10315 (Gottheimer, AI labs in K-12 schools), score 6 — fifth deferral. Standing caveat unchanged: the digest cites an all-actions URL, not bill text.
• California SB 1119 and AB 1864 bill texts (score 6 each) — sixth deferral. Text-capture gap; leginfo URLs are in the 2026-09-10 22:05 digest.
• The eleven other California child-safety bills signed September 10 (score 5 as a set) — sixth deferral. The placement question is a decision, not research, and the cap will keep deferring it until someone makes it.
• Anthropic Institute economic scenarios working paper (score 5) — third deferral, and a named hole in The Anthropic Institute's own ## Open questions.
• entities/rum-group (score 4) — third deferral, now with more support than before: the counterparty identification moved out of "reported but unverified" into a named September 14 report giving the 120 MW Maysville figure and the 51-million-share option. Still two companies neither confirming. Held at the quality gate one more run.
• Chen Yixin article, China Cyberspace (score 4) — second deferral; South Korea's expanded espionage law (score 5) — third; Epoch AI, The Nvidia-sized hole in US GDP statistics (score 5) — fourth; 404 Media / Marshall Project DHS targeting teams (score 5) — third; Terence Tao, "A Severe Misalignment of AI in Mathematics" (score 5) — fifth; Nvidia–Groq DOJ antitrust investigation (score 5) — sixth; entities/bureau-of-land-management (score 4) — third.
• Eighth and ninth deferrals: entities/tech-transparency-project and the TTP CSAM-advertising report (score 6); the expanded Anthropic Max-tier class action (score 6); legislation/ pages for the AI OVERWATCH Act, Chip Security Act and MATCH Act (score 5 each, placement undecided); the OpenAI Navier–Stokes proof artifacts; the 葬AI / FuneralAI ChinaTalk essay; entities/supreme-peoples-court and the SPC companion Q&A; the Clement-Jones amendment and the Sanders–Casar press release; OpenAI's Daybreak for Frontline Defenders and entities/ms-isac; entities/alex-sobel and the Canadian half of the ControlAI coalition; entities/cosmos-institute (second deferral, and it did not appear in the referenced-2+ lane as the last report predicted — see below).
• entities/adam-thierer (score 3) — nineteenth consecutive deferral, structural. Reported once, not re-listed.
• Three slice-3 pages, eighteenth deferral — structural: Standards-Industrial Complex, Three Theories of Victory (US / EU / China AI Governance), Transaction Costs (Coase) and the Agentic Economy. source-robustness-check work.
• Slice carryover. Slice-4 (sixteenth): American Civil Liberties Union (ACLU), Access Now. Slice-5 (fifteenth): Helen Toner, International Monetary Fund (IMF), Henry Farrell; plus Maria Cantwell and John Thune, both closed today. Slice-6 (fourteenth): Stanford Center for Research on Foundation Models (CRFM) and Sen. Elizabeth Warren carry forward; Robert Chesney is closed today after thirteen deferrals. Slice-8: GEMA v. Suno (Munich Regional Court), Sony Music v. Udio (second suit, 2026). Slice-9: Grok Imagine (eleventh; still model-robustness-check work). Slice-10: Missing Coverage: Companies, People, Models, Legislation, Concepts (revised), Crystallize Session: Wiki Gaps & User Knowledge Map (2026-04-13).
• Twenty-one source-robustness-check candidates across slices 11–13 — twenty-third consecutive report, structural. Out of this skill's lane.
• Date-stamped headers in lint-report.md — thirtieth cycle, structural. Now 25, unchanged. Self-references at 6, unchanged.

Findings that are not gaps

• The CourtListener MCP is authenticated and working, after twelve consecutive runs reporting it unavailable. Every report from 2026-09-04 onward recorded it as present but requiring an OAuth flow the session could not run, and the 2026-09-15 report said twelve runs was "well past a transient failure." It answered on the first call today and delivered, for the DistroKid matter alone, the case number, court, filing date, assigned judge, three defendant entities, the full docket entry list, page counts for the complaint and its exhibits, and a literal search inside the complaint text confirming two distinctive passages. Three items that had been blocked solely on this connector are now retrievable: the two slice-8 litigation gaps (GEMA v. Suno (Munich Regional Court), Sony Music v. Udio (second suit, 2026)) and the Anthropic Max-tier class action, which the last four reports recorded as blocked on an unrecorded court and case number. They are left in the backlog rather than actioned, because the cap was already met — but the reason they were stuck no longer applies.
• The FRONTIER Act resolves by alias and is still not a missing page. Named again in the 2026-09-15 08:05 digest, where OpenAI endorsed its independent-verification provision and gave it a bill number (H.R. 9925). It resolves to Frontier Act / Great American AI Act (Obernolte–Trahan), whose ## Introduced bill: the Frontier Act (July 2026) section carries it. This was determined on 2026-08-28 and again on 2026-09-12; recorded a third time because the bill number is new information that does not change the resolution. The curator rename that section describes is still outstanding.
• Gemini 3.8 Live resolves by alias. Gemini 3 / Gemini 3 Pro is the family page and its own description already runs "through … the Flash point-releases running to Gemini 3.8 Flash (September 2026)." The September 15 release of Gemini 3.8 Live and 3.8 Live Extended Thinking is an update to that page, which the nightly fold owns, not a new page.
• entities/cosmos-institute did not enter the referenced-2+ lane. The 2026-09-15 report predicted it would, on the ground that The Anthropic Institute added a second reference to a previously single-reference target. lint-scan.py still reports 0 targets at 2+ references. The prediction was reasonable and the mechanism was right; the count did not move, which is worth one line so the next run does not go looking for it.
• Slice-6 thin anchors are now zero. The slice held exactly one page under 1,800 body characters at in-degree ≥6 — Robert Chesney — and it was expanded. Eighteen pages in the slice sit under 1,800 characters in total, all of them below the in-degree bar.
• Slice-6 graph wiring is effectively complete: 1 of 80 pages at in-degree ≥6 lacks ## Relationships, against 0 of 148 in slice 5 and 3 of 87 in slice 3.

Verification notes

• No broken outbound links introduced. All five created or expanded pages were resolved programmatically against the full content-page set after writing: zero unresolved wikilinks on each. Two links were corrected during writing rather than left: [[companies/google]] does not exist and was pointed at Google DeepMind, the vault's Google landing page; [[concepts/coordinated-slowdown]] does not exist and was pointed at Coordinated slowdown proposals compared.
• Style scan clean. All five checked programmatically against the banned-header, hyperbole, self-reference, date-led-header and standing-Predictions patterns: zero hits on each, with one exception that is not a violation — unprecedented appears once on Maria Cantwell, inside a pre-existing verbatim quotation from Cantwell, where house style permits it. Section headers on the two expanded senator pages are named for their subjects ("Negotiations and committee timetable", "Role in the 2026 Senate negotiation", "Future of AI Innovation Act"), not for dates.
• Revision fidelity. No prose sentence, figure, name, quotation, citation or existing section was removed from any of the three expanded pages. Every (Source: …) present before the run survives: Thune 3 → 13, Cantwell 3 → 8, Chesney 2 → 5. The Chesney edit is the one that changes an existing claim rather than only adding — "a professor at" became "Dean of" — and the change is made against two institutional sources both cited in place, so it is auditable.
• Confidence not raised anywhere. Two pages created at medium, never high, per the autonomy rule. Three expanded pages held at medium despite rising source counts: the Thune and Cantwell subjects rest on an unpublished draft reported through people familiar with the negotiation, and Chesney's new material is biographical and bibliographic rather than independently corroborated on three sources.
• Capture completeness stated, not assumed. S. 5105: complete, sections 1–4, with congress.gov page furniture omitted and no substantive text truncated. UMG complaint: not captured, queued with the verified URL under the large-PDF rule, page counts recorded (52 + 36). NM Bar Bulletin: not captured, with the reason and the exact page range recorded in its verification record — a first fetch returned pages 1–12 and stopped one page short of the opinion.
• Cap respected. Seven gaps actioned against a 6–10 cap; 26 candidates after dedupe. Eight web pulls (four Firecrawl searches, three Firecrawl scrapes, one PDF parse) and three CourtListener calls; no crawls.
• Lock held in file-only mode throughout, with heartbeats between phases. No git command was run from this session; the changeset is left uncommitted for the host sync loop.

One-line summary

Closed the week's blocked legislative thread by finding S. 5105 and building the page from the bill text rather than the coverage — which moved the introduction date back seven weeks and corrected who receives the statute's advance notice; created Protect Democracy; brought John Thune and Maria Cantwell forward from early August to the live impasse over who runs the safety tests; closed Robert Chesney after thirteen deferrals; and queued three foundational sources — with the run's standing finding being that the CourtListener connector is working again after twelve reported failures, which unblocks three backlog items the curator no longer needs to authorize anything for.