Gap Scan — 2026-08-02

Scanned

Recent window: 4 New Developments Log/ files (2026-07-31 08:10 and 22:09; 2026-08-01 08:09 and 22:12) and 60 wiki pages edited in the last 48 hours. Rotation slice: 3 — concepts/ P–Z (81 pages).

Link scan over 1,710 content pages found 262 distinct broken [[wikilink]] targets, down from the 361 a naive scan reports. The 99-target difference is the escaped-pipe artifact this scan has flagged on prior runs: targets written [[companies/anthropic\|Anthropic]] are captured as companies/anthropic\ by a regex that stops at | without stripping the escape. Handling the trailing backslash removes the entire top of the false-positive list — companies/anthropic\, companies/openai\, companies/meta\, claude-fable-5\ and the rest. This is the same defect logged against bin/lint-scan.py on five consecutive runs, and the lint report's headline figure of 345 broken links carries it.

Firecrawl was available for this run, in contrast to the 2026-08-01 22:12 developments file, where every scrape returned "Insufficient credits." Six searches were made; no bulk crawl. Every claim added below rests on a source read this run.

Gaps actioned (9 of 23 found)

New pages created (live)

• Anil Seth — broken [[entities/anil-seth]], in-degree 2, cited from Luiza Jarovsky and \"Conscious AI\" as an AI Safety Issue — Luiza Jarovsky (April 2026) as the author of the strongest available counter to broad functionalism about machine consciousness. Alias resolution found no existing page under any variant. Built from his own bio page, Wikipedia, the Noema essay and the Berggruen Prize announcement; confidence: medium, sources_count: 4. The essay — "The Mythology of Conscious AI," winner of the English-language category of the 2025 Berggruen Prize Essay Competition — supplies the page's substance, including its central conditional and its argument that collective belief about machine sentience is consequential independent of the underlying fact.
• Rogé Karma — broken [[entities/roge-karma]], in-degree 2, cited from Lila Shroff and So, About That AI Bubble — Rogé Karma (The Atlantic, May 2026). Atlantic staff writer, author of the "Work in Progress" newsletter, previously senior editor of The Ezra Klein Show. confidence: medium, sources_count: 3. Short by design: the page carries what two publisher bios and the wiki's existing source page support, and no more.

Pages expanded (live)

• Three Privacy Problems AI Creates — the strongest genuine thin anchor in slice 3: in-degree 17, sources_count: 1 (accurate, not a metadata error), last touched 2026-06-06. Everything on it derived from one lecture transcript. Three changes:
  • Attribution repaired. The Stanford HAI white paper it adapts was credited only to the institution. It is now named in full — Rethinking Privacy in the AI Era: Policy Provocations for a Data-Centric World, by Jennifer King and Caroline Meinhardt, February 22, 2024 — with both authors' affiliations. Leaving an institution where a source names its authors is one of the four failure classes in source-fidelity.
  • New section, "Regulatory attention to the agent cascade." Of the framework's three problems, the agent cascade is the one data-protection regulators have taken up since the page was written. Added: the UK ICO's January 2026 Tech Futures report on agentic AI, including its finding that explicit consent may be unobtainable in the agentic context absent a genuine choice; Spain's AEPD guide to agentic AI; and Beduschi's 2026 article. Folded thematically, not as a dated timeline. 2 new (Source: URL) cites plus 2 corroborating URLs.
  • Four malformed typed relationships fixed. introduced-by: and adapts: are not in the CLAUDE.md vocabulary and were rewritten to depends-on:; a bare-prose contradicts: with no link target now points at Three Theories of Consent Failure (Information / Capacity / Design); and the dead [[concepts/inference-explosion]] (planned subset) was removed. sources_count 1 → 5, last_updated → 2026-08-02, confidence held at medium.

Frontmatter repaired (live)

• 44 of 81 slice-3 concepts/ pages had understated or missing sources_count. Recomputed from distinct [[sources/…]] wikilinks plus distinct (Source: URL) cites and corrected upward; no value was lowered. The two largest errors sit on the two most-relied-upon pages in the slice: Scaling Laws, in-degree 90 — the highest of any page in the wiki — declared 1 while citing 11; and Recursive Self-Improvement (RSI), in-degree 49, declared 1 while citing 11. Others of scale: regulatory-typology 2→16, san-francisco-consensus 2→16, risk-based-ai-regulation 6→16, regulating-under-uncertainty 4→15, sycophancy-and-hallucination 4→15, treacherous-turn 4→15, state-level-ai-regulation 13→22, reward-hacking 17→22. Techno-Federalism was missing the required field entirely (now 9).

This finding reframes the slice's thin-anchor list. Of the ten pages the mechanical scan flagged as thin anchors, eight were flagged only because their sources_count was wrong — scaling-laws at 806 words with 11 real citations is well-supported, not underbuilt. After repair, one genuine thin anchor remained in slice 3 (three-privacy-problems-ai, expanded above) and one is still open (specification-hazard, in-degree 13, genuinely 2-sourced). The same audit is worth running against slices 1 and 2, whose 2026-06-06 cohort has now been flagged on three consecutive runs.

last_updated was deliberately not bumped on the 44 repaired pages. A count correction is not a content revision, and bumping the date would reset each page's confidence-decay window and mask real staleness.

Citation-format defects fixed

• [[sources/nyt-cohere-aleph-alpha-merger-2026-04-24]] → (Source: <https://www.nytimes.com/2026/04/24/business/cohere-aleph-alpha-ai-merger.html>) in Aleph Alpha and Runway Started by Helping Filmmakers — Now It Wants to Beat Google at AI (Bellan, TechCrunch, May 2026) (2 instances). This was not a missing-source gap. The Aleph Alpha reference labels the article "supporting" in its own text, and CLAUDE.md directs supporting sources to (Source: URL), never to a sources/ wikilink — so the page was citing a sources/ page the rules say should never exist. The article itself was verified as real (Adam Satariano, "A.I. Start-Ups From Canada and Germany Merge to Take On Silicon Valley," The New York Times, April 24, 2026; corroborated by Satariano's NYT author page and Techmeme's 2026-04-24 archive) and is now cited with byline, title and date in place of the broken link. This is the fourth consecutive run to flag citation-format defects of this shape.

Queued — foundational sources

• Ana Beduschi, "Data protection in the era of agentic artificial intelligence" (Elsevier, July 2026) — a peer-reviewed law article on whether data protection law can accommodate agentic AI. It is the first academic anchor found for what Three Privacy Problems AI Creates calls the agent cascade, a concept carrying in-degree 17 on thin sourcing. Verified: ScienceDirect (publisher of record, PII S2212473X26000830) plus an open-access deposit in the University of Exeter repository; corroborated by the author's Exeter research-outputs listing and a Global Legal Insights item quoting her. DOI and journal title were not established and are recorded as unresolved rather than guessed. Queued: INGEST-beduschi-agentic-ai-data-protection-2026.md. Verification record: queue/gap-scan/proposed-sources/beduschi-agentic-ai-data-protection-2026.md.
• Nikhil Suresh, "AI Mania Is Eviscerating Global Decision-Making" (Hermit Tech) — surfaced by the 2026-08-01 22:12 developments file. Foundational on the borderline heuristic: it advances an original argument from the author's own primary observation across roughly 300 professional conversations, not a restatement of reporting. Verified: hermit-tech.com is the author's own domain and the original place of publication; corroborated by his LinkedIn announcement, a Tildes discussion thread, and Cory Doctorow's 2026-08-01 Pluralistic essay quoting it by title. Distinctive-passage check passed on four section headings and two named anecdotes. Queued: INGEST-suresh-ai-mania-decisionmaking-2026.md. Verification record: queue/gap-scan/proposed-sources/suresh-ai-mania-decisionmaking-2026.md.

Both were queued by verified URL rather than saved to Raw Sources/: the Beduschi copy to capture is the open-access Exeter deposit, and the Suresh capture returned only in part this run. Saving a fragment as though it were the raw source is precisely the failure source-fidelity was written against.

Authenticity-verification failures

None. Four documents were verified this run (the Seth essay, the NYT merger report, the Beduschi article, the Suresh essay); all four passed. Two fields on the Beduschi record — DOI and journal title — are logged as unresolved rather than inferred; this is an incomplete record, not a failed verification.

Deferred to needs-review

• Situational Awareness LP (score 8: +3 live thread, +3 dense 48h coverage, +2 core area) — declined on the quality gate and written up at queue/gap-scan/needs-review/2026-08-02-situational-awareness-lp-placement.md. Two curator questions block it: the fund is already carried substantively on Leopold Aschenbrenner, so a separate page may violate the no-duplication rule; and entity_type has no value for a for-profit non-AI-developer, the same schema gap already parked under nonprofit placeholders on SoftBank Group, Sequoia Capital, Constellation Energy, NextEra Energy and Dominion Energy. Five placeholders makes this a schema decision, not a per-page workaround.

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

• concepts/specification-hazard — in-degree 13, genuinely 2-sourced after repair, 616 words. The strongest remaining slice-3 thin anchor. (score 6)
• Slice-3 thin anchors not reached, in reliance order: two-conditions-three-weapons-manipulation (in-degree 9), surveillance-technology (9), standards-as-litigation-evidence (7), peat-framework (6). All now correctly counted but still lightly sourced. (score 5–6)
• OpenAI "Astra" model family — named in the 2026-08-01 08:09 file, but sourced to three people briefed on unreleased plans, with no confirmed name, date or capability set. Declined on the quality gate; a models/ page built on an unannounced product is speculation. Revisit on announcement. (score 6)
• Nexus Data Centers — in advanced talks to raise $15 billion for a Texas campus tied to Anthropic (2026-07-30). Single source, transaction not closed, and the company has no other coverage in the wiki. (score 5)
• entities/samuel-weinbach and entities/ilhan-scheer — Aleph Alpha co-founder and co-CEO, in-degree 2 each, alias resolution confirms neither exists. Genuine missing-page gaps, below the cap. (score 4)
• entities/clayton-christensen (in-degree 2) and entities/adam-smith (in-degree 2) — both genuine broken targets, both historical figures cited for a framework rather than for AI-policy positions. Worth a curator ruling on whether the wiki wants biography pages for pre-AI thinkers at all before either is created. (score 3)
• concepts/inverse-cooking-problem and concepts/inverse-trust-problem (in-degree 2 each) — not re-actioned; already parked at needs-review/2026-07-10-inverse-cooking-trust-coined-terms.md as coined terminology, which house style bars absent attribution to a cited source. Carried unchanged.
• bin/lint-scan.py escaped-pipe target-capture bug — sixth consecutive run flagging this. It inflates the reported broken-link count by roughly 38% (345 reported vs. 262 actual) and puts five nonexistent targets at the top of the list. A one-line .rstrip('\\') on the captured target fixes it. Outside gap-identifier's remit; needs a curator or lint fix.
• The 14 dated dashboard-rebuild-failed notes from June still sit unresolved in needs-review/. Carried.
• Recent-window items behind the nightly fold, not gaps: Microsoft and Amazon Q2 results (snapshot rows), Kimi K3 architecture detail, MiniMax H3, the Nano Banana 2 rollback, the Codex Security CLI release, the EU AI Act transparency deadline of 2026-08-02, and the Anthropic containment-incident detail. Pages exist for each; the developments-log cycle owns them.
• Recent-window items already handled: Reddit Inc v. SerpApi LLC (Reddit's data-scraping suit against Perplexity AI), Inkling-Small, Gemini Robotics 2, CHAT Act and CHAT Act 2.0, ATOMIC Act (H.R. 9965), Irregular, Defense Innovation Unit (DIU) all exist and were edited within the window.

One-line summary

Two entity pages created, one high-in-degree concept page expanded and its four malformed relationships fixed, 44 slice-3 sources_count errors corrected — including in-degree-90 Scaling Laws, which had been misreporting 11 sources as 1 — two citation-format defects repaired, and two verified foundational sources queued for the curator's ingest; the slice's apparent thin-anchor problem turned out to be a frontmatter-accuracy problem, and the lint-scan.py escaped-pipe bug now has six consecutive flags.