Gap Scan — 2026-09-03

Scanned

Recent window: 4 dev-log files (2026-09-01 08:11 and 22:10; 2026-09-02 08:05 and 22:05). Rotation slice: 7 — legislation/ (all), 153 pages. Lint baseline at scan start: 212 distinct broken wikilinks, 0 referenced 2+ times; 26 date-stamped headers; 6 self-references; zero banned headers, standing predictions or header hyperbole; 1 unprocessed digest (2026-09-02 22:05 — the nightly fold had not run at scan time). primary-doc-scan.py --days 3: 6 digests, 6 cited primary-document URLs, 1 with no raw file, 1 possible match under another address (the Anthropic August 2026 Risk Report, already held under its canonical landing page), 3 lane-2 keyword hits.

Gaps actioned (7 of 19 found)

New pages created (live)

None. The live-thread page candidates in the 2026-09-02 22:05 digest — Gemini 3.8 Flash and 3.8 Flash Cyber, Muse Spark 1.3, the New York City schools moratorium, Broadcom's Q3 and the Microsoft segment change — are items of the unprocessed digest, which the nightly developments-log fold owns; creating pages here would collide with that run. No slice-7 gap required a new legislation/ page: every statute, order and framework named in the four digests already has one, including H.R. 10152 and H.R. 9129 from the September 1 House Energy and Commerce markups.

Pages expanded (live)

• Amazon v. Perplexity AI — score 8 (live thread +3 via the SIIA amicus item in the 2026-09-02 08:05 digest, dangling foundational reference +3, core area +2). The page carried two recorded defects: an unreconciled August 4 / August 5 date for the appellate ruling, and the explicit statement that "the retrieved account does not state what the ruling does to the preliminary injunction, and the disposition is therefore not recorded here." Both are now resolved from the published opinion, retrieved from the Ninth Circuit's own datastore: the opinion is dated August 4, 2026 and vacates the preliminary injunction and remands. Added the panel composition (M. Smith writing, Tung, Hinderaker by designation), the holding on the "access" element under both the CFAA and the CDAFA with the panel's own language, the rule-of-lenity reasoning, the equitable-factors analysis absent from the page entirely, the opinion's four express self-imposed limits, Amazon's August 18 en banc petition, the substance of the SIIA brief's four arguments and its Agentic Commerce Protocol / Model Context Protocol framing, and the full merits-stage amicus list. Corrected the court, case_status narrative, ## Current status and description frontmatter, which all still described the injunction as in force on appeal. A > **Provenance note:** records that the opinion citations should upgrade once the queued ingest runs. [6 new (Source: URL) cites; sources_count 6 → 9; confidence held at medium — the March 2026 district-court orders are still unretrieved, and the page says so.]
• Colorado AI Act (SB 24-205) — score 4 (in-degree 100, second-highest in slice 7 +2; US AI regulation core +2). Content-rich at 2,165 words but with no ## Relationships section, so the highest-in-degree deregulation story in legislation/ had no typed outbound edges. Added five typed links wiring it to the ADMT concept the SB 26-189 rewrite adopts, the governor's workgroup, the constitutional challenge, both sponsors' pages, and the two state frontier-safety statutes it is habitually contrasted with. No prose claim added or removed.
• New York RAISE Act — score 4 (in-degree 83 +2, core +2). Missing sources_count entirely (frontmatter defect, now 6) and no ## Relationships. Added typed links: depends-on California SB 53, which the signing statement names; contradicts the December 2025 state-preemption executive order; related Illinois SB 315 for the third-party-audit contrast the page already draws in prose, plus the DFS, Bores, rogue-deployment and safety-framework anchors. Purely additive.

Frontmatter repairs

• California SB 53 — highest in-degree page in the slice (149) and missing sources_count; set to 6 from the page's own citation count. No other change; the page's depth is a source-robustness-check matter for the sources/ page, which remains on the structural backlog.

Link aliases fixed

None. The referenced-2+ lane was empty before and after the run — the fifth consecutive scan to report zero. The single-reference long tail was not re-audited, per the standing finding.

Queued — foundational sources

• United States, Statement of Interest, In re OpenAI, Inc. Copyright Infringement Litigation (2026-09-01) — score 8 (live +3, uncaptured primary document +3, US regulation core +2). Gap type 9: the 2026-09-02 22:05 digest carries the filing entirely through Reuters, and no content page names it. Verified: CourtListener RECAP, HTTP 200, 20 pages; ECF stamp Case 1:25-cv-03483-SHS-OTW Document 316 Filed 09/01/26; MDL caption 25-md-3143 (SHS)(OTW); signed Woodward, Shumate and Weisbuch, dated September 1, 2026; Wired links the identical RECAP URL and Reuters quotes a sentence that appears verbatim. Saved: Raw Sources/Statement of Interest of the United States - In re OpenAI Copyright Infringement Litigation (DOJ, 2026-09-01).md. Queued: INGEST-us-statement-of-interest-openai-mdl-2026.md. Figures and passages the ingest should carry that the digest lacks: the § 517 basis and the footnote 11 statement that the argument applies to "all parties in this litigation and the related cases"; the licensing-competition argument that fees would function "primarily as large subsidies for old mainstream media companies"; and footnote 17, which says the Register of Copyrights' Part 3 report "does not warrant deference" under Loper Bright.
• Ninth Circuit, Amazon.com Services, LLC v. Perplexity AI, Inc., No. 26-1444 (2026-08-04) — score 8 (live +3, dangling foundational reference +3, core +2). Gap type 1: named on Amazon v. Perplexity AI and relied on there, with no Raw Sources/ file and no sources/ page. Verified: cdn.ca9.uscourts.gov, the court's own opinion datastore, HTTP 200, 21 pages, FOR PUBLICATION; Justia's docket page carries the same August 4 date; the SIIA brief quotes the slip opinion at pages 7-18 word for word. Saved: Raw Sources/Amazon.com Services LLC v. Perplexity AI Inc - Ninth Circuit Opinion No. 26-1444 (9th Cir., 2026-08-04).md. Queued: INGEST-ca9-amazon-v-perplexity-opinion-2026.md.
• SIIA, Brief as Amicus Curiae Supporting Rehearing En Banc (2026-08-28) — score 6 (live +3, primary document +3). Verified: siia.net's own file, HTTP 200, 25 pages; ECF header Case: 26-1444, 08/28/2026, DktEntry: 67.2; SIIA's own August 31 announcement links this exact PDF; the panel opinion independently lists Manatt's Tod Cohen as SIIA's counsel in the case. Saved: Raw Sources/Brief of the Software and Information Industry Association as Amicus Curiae Supporting Rehearing En Banc - Amazon v. Perplexity (SIIA, 2026-08-28).md. Queued: INGEST-siia-amicus-amazon-v-perplexity-2026.md. The ingest should decide whether the Agentic Commerce Protocol and the Model Context Protocol warrant standards/ or concepts/ pages; neither exists today, and both are named as industry permissioning baselines in a court filing.

Three verification records written to Wiki/_meta/queue/gap-scan/proposed-sources/.

Authenticity-verification failures

None. All three pulls resolved to a primary host on the first attempt: the court's own datastore for the opinion, CourtListener RECAP for the docketed statement of interest, and the filing organization's own domain for the amicus brief. No secondary copy or mirror was used, and the PublicNow mirror of the SIIA brief that surfaced in search was declined in favour of siia.net.

Resolved from the deferred backlog

• OpenAI, "pacing model development" post — carried at score 5 on 2026-09-02 as "not in Raw Sources/ by URL grep". It is in fact held as Raw Sources/OpenAI - Pacing model development in an era of cyber-critical capabilities (openai.com, 2026-08-18).md. The earlier check missed it; dropped from the backlog, no pull needed.
• ITIF / Public First data-centre polling (2026-08-26 digest item) — held as Raw Sources/Public First and ITIF - The opposition to data centers (polling deck, August 2026).md. Not a gap.
• METR / Redwood Hugging Face analysis — the 2026-09-02 08:05 digest describes an analysis "made public on September 1"; Raw Sources/Brief Independent Investigation of Agents' Behavior in the OpenAI - Hugging Face Incident (METR and Redwood Research, 2026-08-26) is held under an August 26 date. Whether the September 1 publication is the same document under a later release date or a fuller successor was not resolved in this run and should not be assumed either way; carried to the backlog below rather than closed.

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

• Live-thread items owned by the nightly fold (score 3 each). The 2026-09-02 22:05 digest is unprocessed at scan time. Its page-level work — Gemini 3.8 Flash and 3.8 Flash Cyber and the Fairwind Program on Gemini or a new model page, Muse Spark 1.3, the New York City schools generative-AI moratorium (a candidate for Education — AI Deployment), OpenAI's "automated shutdown capabilities" letter to Reps. Casar and Matsui, the Broadcom Q3 TPU 8i and Jalapeno deployment figures, and Microsoft's Azure segment redefinition — belongs to developments-log. This is the second consecutive report of an unprocessed evening digest at scan time.
• METR / Redwood, September 1 Hugging Face analysis (score 5) — needs a titles-and-dates comparison against the August 26 raw before any pull; if it is a distinct fuller report, it is a gap type 9 with a live thread behind it.
• OpenAI letter to Reps. Greg Casar and Doris Matsui, September 2, 2026 (score 4) — the instrument the day's safety item is about, but Reuters describes it as "reviewed by Reuters" and no public copy was located in this run. Candidate for a targeted search of the members' own press pages next run; do not queue from coverage.
• New York City Public Schools generative-AI policy (score 3) — the nyc.gov item is a press release (supporting). Whether a chancellor's regulation or policy document exists separately was not checked.
• Agentic Commerce Protocol and Model Context Protocol (score 4) — named as industry permissioning baselines in the SIIA brief with no standards/ or concepts/ page for either. Raised this run; routed to the SIIA ingest to decide placement, since a page built from a litigant's characterisation would inherit that characterisation.
• Slice-7 thin anchors under the standard diagnostic (in-degree ≥6 with sources_count ≤ 2, missing sources_count, no ## Relationships, or under 350 words): 25 of 153 pages. Worked: colorado-ai-act, new-york-raise-act, california-sb-53. Next by score: Seoul Frontier AI Safety Commitments (2024) (in-degree 55, sources_count: 2, high); Bletchley Declaration (AI Safety Summit, 2023) (54, 584 words, sources_count: 2, high); BIS Framework for AI Diffusion — Interim Final Rule (RESCINDED) (39, sources_count: 1, high); China — Interim Measures for the Management of Generative AI Services (37, sources_count: 1, high); AI LEAD Act (S. 2937) (34, 679 words, no ## Relationships); Executive Order 14179 — Removing Barriers to American Leadership in Artificial Intelligence (27, 607 words, sources_count: 1, high); California SB 1047 (VETOED) (25, sources_count: 1, high). Fifteen slice-7 pages carry confidence: high on sources_count: 1, which the decay table does not permit for any claim type; that is a systematic pattern in this folder rather than a set of individual misses, and is better addressed by a single sourcing pass than one page per day.
• Slice-6 carryover (second deferral): Stanford Center for Research on Foundation Models (CRFM) (in-degree 14, high on 2 sources), Robert Chesney (9, 137 words), Sen. Elizabeth Warren (9, 319 words).
• Slice-5 carryover (third deferral): American Civil Liberties Union (ACLU) (24, sources_count: 2) and Access Now (20, sources_count: 2).
• Three slice-3 pages, fifth deferral — structural: Standards-Industrial Complex, Three Theories of Victory (US / EU / China AI Governance), Transaction Costs (Coase) and the Agentic Economy. Five runs without reaching the top ten; they will not surface on score alone and need either a direct instruction or a decision that they are not gaps.
• entities/adam-thierer (score 3) — sixth consecutive deferral, structural.
• Twenty-one source-robustness-check candidates across slices 11–13 — eleventh consecutive report, structural. Led by California SB 53 — Transparency in Frontier AI Act (in-degree 254) and TAKE IT DOWN Act — Source Summary (97).
• models/deepseek-v3 — thirteenth run, structural. Routed to model-robustness-check on 2026-08-22, unaddressed.
• Date-stamped headers in lint-report.md (score 1) — eighteenth cycle, structural, holding at 26. Six self-references also appear in this run's lint output.
• Carried unchanged. Marvell (no company page, since 2026-08-21); California AB 1856 (text not located); the Texas OAG / Meta Webster–Mahoney exchange; the draft executive order on a frontier-developer self-regulatory organization (no public text); the AI Evaluator Forum / AEF-1 standard (score 4); ExploitBench flag-level disclosure (score 3); the "Carolina Principles" (no document exists); OpenAI's "Our agreement with the Department of War" (verified URL held at score 2); the House E&C markup memo (staff memo, supporting).

Verification notes

• No broken outbound links introduced. All 22 wikilink targets added across the four edited pages and the three INGEST- task bodies were resolved programmatically against the content folders; all resolved on the first check.
• Style scan clean. The edited pages were checked against the banned-header, hyperbole, coined-term and self-reference lists: zero hits. The panel's own phrases are quoted and attributed rather than restated as the page's voice.
• Revision fidelity. Every pre-existing sentence, table row, citation and "Sources" entry on the four edited pages survives, except for three passages on Amazon v. Perplexity AI that were superseded because the primary document contradicted or completed them: the unreconciled August 4 / August 5 date, the statement that the disposition could not be recorded, and the confidence note asserting that no primary court order had been retrieved. Each replacement carries the primary-source citation. The two legislation/ expansions are purely additive.
• Source fidelity. Dates were taken from the documents' own faces, not from the digests or the coverage: the statement of interest from its ECF stamp (09/01/26), the opinion from its filing line (August 4, 2026), the brief from its ECF header (08/28/2026). Where coverage and document disagreed — Reuters' August 5 for the opinion — the document governs and the discrepancy is recorded on the page rather than silently corrected.
• Confidence set honestly. Amazon v. Perplexity AI stays medium: the appellate record is now primary-sourced, but the March 2026 district-court orders are still known only through news coverage and as quoted in the opinion. No page was raised to high in this run.
• Raw-source fidelity. The three raws are full extracted texts of 20, 21 and 25 pages; footnotes are preserved in place, nothing was summarised or trimmed, and no editorial gloss was added inside the document bodies.
• Cap respected. Seven gaps actioned against a 6–10 cap; 19 candidates found. Three web pulls plus two searches; no crawls.
• Dashboard rebuilt — see the audit-log entry for the article count. The broken-link count is unchanged; the referenced-2+ lane is still empty.

One-line summary

Three primary documents behind the current litigation thread — the DOJ's statement of interest in the OpenAI copyright MDL, the Ninth Circuit's Amazon v. Perplexity opinion, and SIIA's en banc amicus brief — are verified and queued for review; the Amazon v. Perplexity page's unreconciled ruling date and missing disposition are corrected from the opinion itself, and the three highest-in-degree pages in the legislation/ slice gained the typed relationships and sources_count values they lacked.