Gap Scan — 2026-08-24

Scanned

Recent window: 4 New Developments Log/ files (2026-08-22 08:10 and 22:05; 2026-08-23 08:11 and 22:12) and 35 wiki pages with last_updated in the last 48 hours. Rotation slice: 11 — sources/ A–H.

Candidate generation: bin/lint-scan.py (authoritative broken-wikilink report — 235 distinct targets across 259 refs at the start of the run, 24 referenced 2+ times), bin/primary-doc-scan.py --days 3, the slice-11 anchor-thinness pass, and the deferred backlogs of the 2026-08-21 through 2026-08-23 reports.

Alias resolution was run against the referenced-2+ list before any missing-page gap was declared. It produced one genuine finding this run, described under Link aliases fixed below.

Gaps actioned (6 of 28 found)

New pages created (live)

• Commodity Futures Trading Commission (CFTC) — score 5 (live thread +3, US regulation core area +2). The Commodity Futures Trading Commission was named on three content pages (Kalshi, Private Credit & AI Infrastructure, index.md) with no page, and on 2026-08-19 it took the first US federal regulatory action treating compute as a commodity underlier. Built from the Commission's own copy of the request for comment, Release 9286-26, the Federal Register publication, and the regulations.gov docket: the CEA section 3 mandate, the DCM Core Principle framework in a table (3, 4, 5, 9, 11), the self-certification and approval routes under 17 CFR 40.2/40.3, the Appendix C cash-settlement standard, and a full account of the compute request including the three questions specific to this underlier. 1,125 words, confidence: medium, 4 sources. Depends on queued source: the request for comment itself.

• Compute derivatives — score 5 (live thread +3, compute core area +2). A search for "compute derivative" across the content folders returned zero matches, despite Private Credit & AI Infrastructure already carrying the June 2026 Goldman/JPMorgan exploration and Kalshi sitting under the same DCM framework. Covers the CME Group–Silicon Data contracts (announced May 12, 2026; October 5, 2026 launch set August 11; H100 and B200 rental-index futures, cash-settled, NYMEX rules), the CFTC's preliminary finding on fungibility, standardization and liquidity, the three underlier-specific problems the request identifies (reflexivity of the reference price, unobservable bilateral inputs, feedback into the cash market), the electricity and Merrill Lynch v. Curran antecedents the Commission cites, and the Bandi scale estimate. 1,321 words, confidence: medium, 6 sources. Depends on queued source.

• Air Space Intelligence (ASI) — score 5 (live thread +3 — FAA — Federal Aviation Administration (AI Deployer) and Palantir Technologies were both edited in the window — plus government AI procurement core area +2). Deferred on 2026-08-23. Named on two content pages with no page of its own, despite holding the largest AI-specific FAA award. Built from the company's June 22, 2026 announcement, Reuters, Forbes and the existing Verge citation: the $875M/12-year FMDS and SMART award, the displacement of Palantir and Thales, the fall 2026 initial SMART target, the Flyways platform and its airline and Air Force deployments, the DIU prototype OTA, and the National Grid collaboration. 771 words, confidence: medium, 4 sources.

• Carnegie Mellon University — score 3 (broken target ×2, +1 in-degree band; academic AI research and safety evaluation core area +2). Built as a distributed-footprint entity page on the Johns Hopkins University model rather than a stub: "Carnegie Mellon" appears on 22 content pages. Covers the Newell–Simon origin (Logic Theorist 1956, IPL, General Problem Solver, the 1975 Turing Award, Soar), the institutional build-out from the 1956 computation center to the 1988 school, DARPA's early funding, Tom Mitchell and the Machine Learning Department, Cosma Rohilla Shalizi, and the two current CMU research outputs the wiki already relies on — the ExploitBench benchmark used in UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities (July 2026) and the "Silver Bullet or Security Theater" red-teaming paper — plus the CMU-trained founders and chief scientists at OpenAI and Moonshot AI. 800 words, confidence: medium, 4 sources.

• Vercel — score 3, on its fourth consecutive deferral and named structural on 2026-08-22. 13 content mentions with no page: Agent Plugins 1.0.0 records that Vercel initiated the specification, Financial Services — AI Deployment uses AI Gateway traffic as a market-share indicator, AI Coding Agents carries the $1M sandbox-escape bounty, and gpt-oss (OpenAI open-weight models) and Grok (xAI) list it as a launch-day deployment partner. The page was created rather than deferred a fifth time. Covers Next.js, the AI SDK, the AI Gateway, v0, the Agent Plugins and sandbox-security work, and a financial table (Series F $300M at $9.3B post-money, September 30, 2025; ~$340M ARR estimated February 2026; ~$863M raised). 842 words, confidence: medium, 5 sources. Third-party ARR estimates are labelled as such rather than presented as disclosures.

Pages expanded (live)

None as a distinct action. The run's budget went to five new pages, their graph wiring, and the queued primary document. The slice-11 finding is a routing decision rather than an expansion — see below.

Graph wiring

Nine inbound links added across nine existing pages so the new pages are not orphans and named-but-unlinked references now resolve: Palantir Technologies and FAA — Federal Aviation Administration (AI Deployer) → companies/air-space-intelligence; Kalshi and Private Credit & AI Infrastructure → entities/cftc; Private Credit & AI Infrastructure and Compute Governance → concepts/compute-derivatives; AI Coding Agents, Agent Plugins 1.0.0 and Financial Services — AI Deployment → companies/vercel; Tom Mitchell and Herbert A. Simon → entities/carnegie-mellon. FAA — Federal Aviation Administration (AI Deployer) additionally gained the award's date, term, and value from the primary announcement, which the page previously carried only through the Verge's account.

Link aliases fixed

One, and it is the second confirmation of the 2026-08-20 reframe. [[entities/carnegie-mellon]] was a broken target referenced twice from Cosma Rohilla Shalizi. It was not an alias of an existing page and no rewrite could fix it — the institution page did not exist. Creating it resolved both references.

This is the same pattern as entities/sais → entities/johns-hopkins on 2026-08-20, and it holds the reframe recorded there: the institution-shorthand group is missing-page candidates that each clear a broken target on creation, not permanently dead links. The wiki-wide broken count fell from 235 distinct / 259 refs to 234 / 257 — the first reduction since 2026-08-20, and the second in six runs. entities/santa-fe-institute, entities/bank-of-england, entities/ecb, entities/federal-reserve remain in the same category; entities/uw remains an abbreviation better rewritten in the citing prose than created as a page.

Queued — foundational sources

• CFTC, Request for Comment on the Listing of Compute Derivatives Contracts, RIN 3038-AF77 (issued 2026-08-19) — gap type 9, and the only lane-1 hit from bin/primary-doc-scan.py --days 3. Verified against the issuing agency's own copy at cftc.gov/media/14496/RFC_ComputeDerivates081826/download: full 19-page text fetched live, title/RIN/CFR parts/issuance line corroborated across the Federal Register landing page (91 FR 54259–54264, published 2026-08-21, FR doc 2026-17163), the regulations.gov docket CFTC-2026-1850, and Release 9286-26; two distinctive passages confirmed verbatim. Queued URL-only for a fresh fetch at ingest rather than saved as a raw file — transcribing a primary legal text out of a scrape buffer risks introducing the fidelity errors the authenticity protocol exists to prevent, the same disposition used for the Moody v. NetChoice slip opinion on 2026-08-20. Verification record: queue/gap-scan/proposed-sources/cftc-compute-derivatives-rfc-2026.md. Queued: INGEST-cftc-compute-derivatives-rfc.md. Not ingested.

Two provenance findings are recorded in the queue file rather than left implicit. First, the URL the digest cited is the wrong one: public-inspection.federalregister.gov/2026-17163.pdf is the pre-publication staging copy and is expected to stop resolving now that the document has published, so a future reader following the digest's citation would find a dead link. Two stable canonical addresses are recorded in its place. Second, the digest's claim that "Chairman Selig voted in the affirmative and no commissioner voted in the negative" appears in the Federal Register presentation, not in the Commission's own copy of the approved document, which carries only the Secretary's issuance line; the ingest task carries that correction so the fold sources it accurately.

Authenticity-verification failures

None. One document was pulled and verified; it passed.

Slice-11 finding: sources/ pages are the wrong shape for the thin-anchor rule

The rotation slice was sources/ A–H, and the mechanical anchor-thinness test returned 61 pages — a number that overstates the problem, because sources_count: 1 is the correct value for a single-source summary page and the test reads it as under-support. Filtering to word count against the underlying raw file leaves a real and narrower finding:

These are the two highest-in-degree source pages in the wiki, and California SB 53 — Transparency in Frontier AI Act renders a 7,116-word statute in 678 words while 254 pages point at it. Routed to source-robustness-check, not expanded here — the work they need is a faithful rewrite against the raw file, which is that skill's lane, and the same reasoning that sent models/deepseek-v3 to model-robustness-check on 2026-08-22. Both are last_updated: 2026-06-06.

Recording the routing rather than re-scoring these as thin anchors every fortnight is the point; a sources/ page's depth is measured against its raw file, and gap-identifier does not read raw files.

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

• Live-thread items from the 2026-08-23 22:12 digest (score 3 each), deliberately left to the nightly fold. That digest is the one unprocessed New Developments Log/ file, so developments-log has not yet run on it. Brazil's Bill 4356/25 on AI in the Unified Public Security System (approved in committee August 20; Brazil AI Bill (PL 2338/2023) exists but is a different instrument); Japan's Cabinet Office draft "Principles Code" on IP protection and transparency in generative AI, a soft-law layer under the Act on Promotion of Research and Development and Utilization of AI-Related Technology (Act No. 53 of 2025), which folds into Japan AI Promotion Act (Act on Promotion of Research, Development and Utilisation of AI-Related Technologies, 2025); Australia's Joint Select Committee on Artificial Intelligence, appointed August 20 with a November 30, 2026 reporting deadline, which folds into Australia Voluntary AI Safety Standard (2024). Also the DOJ Antitrust Division's Seismic–Highspot closing statement of August 19 — notable because the Division's stated basis was that AI-native entrants already pressure incumbents, with Deputy AAG Beller cautioning that merging parties "often claim that AI is a disruptive force and rationale for consolidation." No litigation/ page; a merger-review closing is a candidate fold for a competition page rather than a case page.
• The Nvidia–Perplexity discussion (score 3). A round valuing Perplexity above $30B, with annualized revenue reported above $750M from under $250M at the start of the year. Both pages exist; this is a Snapshot fold for the nightly run, not a gap.
• Slice-11 routing, above — two source-robustness-check candidates, California SB 53 — Transparency in Frontier AI Act first.
• The remaining 23 referenced-2+ broken targets — tenth consecutive run, but the framing has changed. Twelve are author entity pages implied by sources/ pages at in-degree 2 (entities/kevin-klyman, entities/lee-anne-fennell, entities/michael-j-d-vermeer, entities/sacha-altay, entities/joseph-bernstein, entities/clayton-christensen, entities/samuel-weinbach, entities/ilhan-scheer, entities/dina-powell-mccormick, entities/adam-smith, and others). The stated instrument since 2026-08-15 has been a single batch pass over the author-page set, and it has not been run. The institution group is now demonstrably tractable — two cleared in five days — so the author set is the one genuinely stuck item.
• Concept targets named but pageless (score 2 each), ninth run. concepts/inverse-cooking-problem and concepts/inverse-trust-problem (parked under needs-review/2026-07-10 as coined-term questions), concepts/a-vision-of-democratic-ai, concepts/data-broker-regulation, concepts/talent-flow-china-us, concepts/economic-possibilities-for-artificial-intelligence, concepts/ai-and-language-models, concepts/algorithmic-decisionmaking (parked under needs-review/2026-08-19).
• Carried unchanged. Marvell (4 mentions, no company page); models/deepseek-v3 (in-degree 49 on 608 words, routed to model-robustness-check); the remaining slice-9 thin model anchors; the slice-10 stale analysis pages; Amir Salek; the International Catholic Legislators Network; the White House release of 2026-07, instrument still unidentified, carried from the 2026-08-20 primary-document baseline.
• The 28 date-stamped headers in lint-report.md (score 1). Down one from 29. Unchanged in substance for eight cycles; the report still does not distinguish legitimate procedural histories from timeline dumps, which remains the finding.

Checks run before closing

• No broken outbound links introduced. Every wikilink on the five new pages and the ten edited pages was resolved against the full page set after writing; zero unresolved.
• Broken-link count fell. 235 distinct / 259 refs → 234 / 257. The reduction is entities/carnegie-mellon, created this run. This is the first run in five where an actioned gap was also on the broken-link list.
• Style scan clean after two fixes, both caught programmatically and both mine. Commodity Futures Trading Commission (CFTC) carried a wiki self-reference ("the prediction-market venues whose AI-related contracts the wiki tracks"), rewritten to describe Kalshi's regulatory status directly; Compute derivatives used "unprecedented," replaced. Post-fix, all five new pages return zero on the banned-header, hyperbole, self-reference, standing-Predictions and date-header checks. Wiki-wide totals: banned 0, self-refs 0, predictions 0, header hype 0, date headers 28.
• Revision fidelity. All ten edits to existing pages were additive — a wikilink wrapped around existing text, or a sentence appended. No page was rewritten, so no _revision-backups/ copy was required and no citation or fact could be dropped.
• Confidence set honestly. All five new pages are confidence: medium. None met the three-independent-source bar. Commodity Futures Trading Commission (CFTC) and Compute derivatives additionally depend on a source not yet ingested and cite it inline as (Source: URL) pending the fold.
• Cap respected. Six gaps actioned against a 6–10 cap; 28 candidates found. Five web pulls and four searches, no crawls.

One-line summary

Five pages created live — Commodity Futures Trading Commission (CFTC), Compute derivatives, Air Space Intelligence (ASI), Carnegie Mellon University, Vercel — the CFTC compute-derivatives request for comment verified and queued for ingest with a correction to the digest's citation and its voting claim, and the broken-link count reduced for the first time in five runs; the curator's attention is wanted on the author-entity batch pass, stuck for ten runs, and on routing California SB 53 — Transparency in Frontier AI Act to source-robustness-check.