Gap Scan — 2026-08-11

Scanned

Recent window: 4 New Developments Log/ files (2026-08-09 08:09 and 22:10; 2026-08-10 08:11 and 22:10). Rotation slice: 12 — sources/ I–P (140 pages).

Broken-link baseline at the start of the run, mainspace only: 251 distinct broken targets across 279 instances, after normalizing the escaped-pipe artifact. That artifact — targets captured with a trailing backslash from [[page\|Alias]] inside markdown tables — accounted for 45 phantom targets across 75 files and is not a defect in the wiki. [[page\|Alias]] is the correct Obsidian syntax for an aliased wikilink inside a table cell, where an unescaped pipe would break the table. Prior runs have flagged this as a bin/lint-scan.py capture bug for eleven consecutive cycles; the fix belongs in the scanner's regex (.rstrip("\\")), not in the pages. It is recorded here again only because Wiki/_meta/lint-report.md still reports the inflated figure of 252 distinct broken targets.

The strict thin-anchor rule (in-degree ≥6 and under 350 words) produced zero hits in slice 12, the same result as slice 11 yesterday. Every high-in-degree sources/ page in I–P runs to 700 words or more. Eight pages trip the weaker medium-confidence-with-sources_count: 1 test at in-degree ≥6; they are recorded in the deferred backlog rather than actioned, since the right instrument for a single-source sources/ page is source-robustness-check, not a supporting-source expansion.

Eleven candidates survived alias resolution and deduplication against the open queue, the seven gap-scan reports since 2026-08-04, and existing Raw Sources/ files. Seven were actioned, at the low end of the 6–10 cap; ten web pulls were made.

Gaps actioned (7 of 11 found)

New pages created (live)

• Tim Fist — named in plain text on four mainspace pages (Recursive Self-Improvement (RSI), Open Problems in Technical AI Governance, Pacing the Frontier (statement from employees of frontier AI companies, July 2026), Institute for Progress (IFP)) with no page behind him, and reinforced by the 2026-08-09 evening dev-log. The existing queue task INGEST-fist-khan-pacing-ai-self-improvement.md had explicitly flagged him as an unanchored candidate. Built from IFP's own author page plus the wiki's existing sourced material. confidence: medium. [Depends on queued sources: the IFP report and the Noahpinion guest post — provenance note recorded on the page.]
• Saif M. Khan — same window and same three mainspace pages. IFP's own author page supplies a substantial public-service record (Counselor for Critical and Emerging Technologies to the Secretary of Commerce; NSC Director for Technology and National Security; senior advisor to the House Select Committee on the CCP; CSET research fellow) that the wiki carried none of. confidence: medium. Note entities/lina-khan already exists and is a different person.
• GPT-5.6-Cyber — OpenAI's purpose-trained cybersecurity model, released 2026-08-10 and carried in that evening's dev-log. Zero mainspace mentions before this run. Alias resolution confirmed it is not covered by GPT-5.6 (Sol, Terra, Luna): that page documents the Sol/Terra/Luna family and mentions Daybreak only in passing, whereas GPT-5.6-Cyber is a separate training run with its own Preparedness classification, its own benchmark set and a restricted release channel. Built from OpenAI's own announcement, with the Axios partner list folded as a second source. confidence: medium.
• In re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation (MDL 3047) — MDL 3047 is named substantively on three mainspace pages (Content vs Architecture Theory of Social Media Harm, Inside My Advanced Topics Class 4: Everyone Described Harm (Farahany, February 2026), Inside My Advanced Topics Class 4.2: The Shield — Section 230 (Farahany, February 2026)) with no case-tracking page, and the Ninth Circuit ruled in it on 2026-08-10. The page records the JPML transfer order of October 6, 2022, the design-defect pleading strategy around Section 230, the March 2026 Los Angeles bellwether verdict, and the August 10 holding that Section 230 supplies a defense to liability but not immunity from suit — with the panel (Nguyen, Bennett, and Matsumoto of E.D.N.Y. by designation), the caption and docket number (People of the State of California v. Meta Platforms, Inc., No. 24-7032), and the denial of Meta's bid to postpone the 29-state-attorneys-general trial. confidence: medium.

Pages expanded (live)

None. No page in rotation slice 12 met the strict thin-anchor rule. Three pages received targeted corrections instead (below) rather than supporting-source expansion.

Link aliases fixed

No alias errors were found. The 45 apparent trailing-backslash targets are correct Obsidian table syntax, not broken links (see Scanned). Three link-and-accuracy corrections were made instead, all connecting the two new entity pages into the graph and correcting one title against the primary source:

• Recursive Self-Improvement (RSI), Pacing the Frontier (statement from employees of frontier AI companies, July 2026), Institute for Progress (IFP) — "Tim Fist" and "Saif M. Khan" wikilinked to the new pages (they had been plain text on all three), and Khan's title corrected from "a distinguished fellow" to "a distinguished technology fellow" per IFP's own author page. No sourced fact was dropped and no unsourced claim added.

Queued — foundational sources

• Americans for Responsible Innovation, "Responsible Innovation at the Frontier" (2026-08-10) — a structured federal legislative blueprint from an organization the wiki already tracks, published in the last 48 hours with nothing of it in mainspace. Verified: ari.us (the publisher's own domain), fetched live, HTTP 200; authorship corroborated against ARI's own team pages; distinctive provisions confirmed present. Saved: Raw Sources/Americans for Responsible Innovation - Responsible Innovation at the Frontier (2026-08-10).md (full text). Queued: INGEST-ari-responsible-innovation-frontier.md.
• Mark Zuckerberg, "The Future is for Everyone" (2026-08-10) — a dangling foundational reference: named on four mainspace pages with no sources/ page and no Raw Sources/ file. Verified: about.fb.com (Meta's own newsroom); title, author and date corroborated against the Guardian and Reuters; four distinctive passages confirmed verbatim against independently published quotations. No raw file saved by design — the newsroom page renders body text with inter-word spacing stripped in extraction, and a capture in that state would break quote verification at ingest. Queued with the verified canonical URL and a preferred alternative first-party host (meta.com/thefutureisforeveryone/): INGEST-zuckerberg-future-is-for-everyone.md.
• Institute for Progress, "How Should the US Prepare for Increasingly Automated AI R&D?" (2026-08-06) — the underlying report behind the guest post already queued on 2026-08-10. Verified as to identity, authorship and date against two independent first-party IFP author pages. Long-form report; body deferred to fetch at ingest. Queued: INGEST-ifp-preparing-for-ai-research-automation.md, with instructions to process it alongside — not instead of — the existing INGEST-fist-khan-pacing-ai-self-improvement.md.

Source-fidelity findings

Two date-and-title discrepancies were found between the digests and the primary sources, both in the schema v4.7 failure classes. Neither is corrected in mainspace by this run beyond the title fix; both are recorded in the queue tasks so the ingest run resolves them against the primary text.

1. A restatement's date displacing the underlying document's. Three mainspace pages date the Fist–Khan pacing argument to August 9, 2026, taken from the Noahpinion guest post. IFP's own site dates the underlying report August 6, 2026. These are two artifacts, not one, and the wiki currently records only the later restatement. Recorded in INGEST-ifp-preparing-for-ai-research-automation.md note 4.
2. A title compressed away from the source. The same three pages give Saif M. Khan's IFP title as "a distinguished fellow"; IFP gives Distinguished Technology Fellow. Corrected on all three this run.
3. A page count reported as fact. Muse Glimmer (Meta Superintelligence Labs) and Mark Zuckerberg both describe Zuckerberg's essay as a "14-page essay," a Reuters figure the 2026-08-10 evening dev-log had already corrected to about 6,500 words. The fetched primary text is consistent with the word count, not the page count. Left in place pending the ingest, with the reconciliation recorded in INGEST-zuckerberg-future-is-for-everyone.md note 4 — correcting it now would decouple two pages from the source that will supersede both.

One consistency check passed: the Advanced Cybersecurity Completion Rate figures in the 2026-08-10 evening digest (95.0% / 57.3% / 2.0% / 1.5%) match OpenAI's own post exactly, including the assignment of 2.0% to Daybreak Blue and 1.5% to safeguards-enabled Sol.

Authenticity-verification failures

None. Three sources were pulled and all three passed the protocol. Three limitations are recorded rather than waved through:

• No formal identifier on any of the three. A think-tank blueprint, a corporate newsroom letter and a think-tank report carry no DOI, arXiv ID, Federal Register citation or docket number. In each case the identifier check rests on the canonical URL, the first-party domain and the site metadata. This is the same limitation recorded for the OpenAI newsletter post on 2026-08-08 and is a standing weakness of the protocol for this source class.
• ARI date discrepancy. The page is dated August 10, 2026; the companion PDF filename encodes 26.08.06. Both recorded; reconciliation deferred to ingest.
• Zuckerberg render artifact. The authentic full text was retrieved but with inter-word spacing stripped. Content authenticity is not in doubt — four passages match independently published quotations verbatim once spacing is restored — but the capture was withheld rather than saved in a degraded state.

Deferred backlog (re-surfaces next run)

The first four entries are the deferred scored candidates that make up the 11-found / 7-actioned arithmetic. The remainder are standing carries and resolved non-gaps, recorded so they stop re-scoring.

• companies/theseus-infrastructure (score 5) — the Anthropic / Macquarie Asset Management / GIC data-centre venture announced 2026-08-10. Live thread and a core compute topic, but a single Bloomberg report supplies only the formation, the equity split and Anthropic's consumer-electricity-price commitment. Withheld under the quality gate; the facts belong in Anthropic via the nightly fold until there is enough for a page.
• Near-threshold thin anchors in slice 12 (score 3) — eight sources/ pages at in-degree ≥6 with confidence: medium and sources_count: 1, none under the 350-word bar: lawfare-sun-export-controls-cannot-reach-2026-05 (in-degree 19), marcus-astra-oversold-2026 (17), lawfare-frazier-rozenshtein-dominance-by-understanding-2026-05 (14), maga-loved-ai-safety-transformer (12), newman-anecdotes-evidence (12), kimi-k3-open-weights-escalation-lambert (11), legora-roi-report (9), openai-model-spec (9). openai-model-spec is the strongest single candidate — 576 words for a heavily cited primary text — and is a source-robustness-check job, not a gap-identifier expansion.
• sources/protecting-wellbeing-of-users (score 2) — 177 words, confidence: low, sources_count: 0, in-degree 0. Carried from 2026-08-10, where it was already flagged alongside sources/when-the-interface-is-neural. Two stub sources/ pages with no inbound links and no recorded source; the standing note is at Wiki/_meta/queue/gap-scan/needs-review/2026-08-10-two-stub-sources-pages-superseded.md.
• Person pages implied by this window, none created — Iskandar Haykel and Morgan C. Plummer (ARI, authors of the queued blueprint); Tao Burga, Arthur Tellis, Ben Schifman, Jonah Weinbaum and Olivia Scharfman (IFP co-authors); Jared Atkinson (SpecterOps CTO, quoted in the OpenAI release). All below the reference threshold; revisit after the ingests land.
• Organizations named in the window with no page — Theseus Infrastructure, Macquarie Asset Management, GIC, SpecterOps. Standing carries from prior runs, still below threshold: entities/steven-adler, entities/consumer-technology-association, entities/epic, entities/cpsc, entities/national-academies, entities/seth-lazar, entities/vector-institute, MeitY.
• concepts/spiralism — re-surfaced from the 2026-08-09 morning window and dropped again on dedupe, correctly. It resolves to Parasitic AI / Spiral Personas, which defines Spiralism inside Adele Lopez's framework; the Verge coverage is already folded there and at Adele Lopez. Not a gap. Recording the resolution again so it stops re-scoring.
• Recent-window items the nightly fold already handled, not gaps — Muse Glimmer, Muse Spark 1.2, Astra, Terafab, Microsoft's Maia 300, Nvidia's $500 billion financing package, the Anthropic Riemann-zeta result, Inkling, Discovery Loop. Each already has a page or a folded home.

One-line summary

Eleven gaps found, seven actioned: four pages created live (Tim Fist, Saif M. Khan, GPT-5.6-Cyber, In re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation (MDL 3047)), three foundational sources verified and queued (ARI's federal blueprint with full text saved; Zuckerberg's "The Future is for Everyone" and IFP's automated-AI-R&D report queued by verified URL), and three pages corrected. Needs the user's review: the three INGEST- tasks, and in particular whether INGEST-ifp-preparing-for-ai-research-automation.md and the pre-existing INGEST-fist-khan-pacing-ai-self-improvement.md should be processed as one ingest or two — they are the same argument in two artifacts published three days apart, and the wiki currently records only the later one.