AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-08-13

Daily gap hunt — what the wiki is missing and how each gap was triaged.

Contents

Scanned

Recent window: 4 New Developments Log/ files (2026-08-11 08:12 and 22:05; 2026-08-12 08:05 and 22:09) and 76 content pages with last_updated inside the last 48 hours.

Rotation slice: 0 — companies/ (all 146 pages), scanned for in-degree-weighted thinness, frontmatter completeness, and confidence-versus-sourcing mismatch.

Broken-link analysis: 265 distinct broken targets by this run's own scan; bin/lint-scan.py reports 251 distinct across 279 refs. Twenty-nine broken targets carry an in-degree of 2 or more; all twenty-nine went through alias resolution.

Queue state at start of run: empty (Wiki/_meta/queue/ held only README.md). Every task filed on 2026-08-12 has been processed and archived.

Run-order note: New Developments Log/2026-08-12-2209-ai-developments.md was unprocessed at the start of this scan — it carries no matching Operation: Developments-Log entry. This skill is specified to run after the nightly developments-log fold, and did not. Three of the eight gaps actioned below (Grok 4.6, Decart, the Colorado rulemaking) come from that unfolded file, which is why they scored as untouched live threads. The file remains unprocessed; the nightly fold should still run over it, and may find the same items already handled here.

Gaps actioned (8 of 26 found)

New pages created (live)

  • Surveillance Pricing — score 6 (live thread +3, wiki core area +2, named in prose on ≥2 pages +1). The highest-scoring genuine missing page of the run. "Surveillance pricing" appears 12 times across four content pages (Maryland Protection from Predatory Pricing Act, New York Algorithmic Pricing Regime and AG Enforcement, overview.md, index.md) with no page behind it, and — the reason it is a real gap rather than an alias — Algorithmic Pricing and Antitrust does not contain the phrase at all. That page's own "Distinction from personalized pricing" section explicitly scopes itself to collusion and hands the personalized-pricing question off to a page that did not exist. Built from the FTC's January 2025 6(b) staff findings and issue spotlight, the Section 6(b) tool-and-data taxonomy, and both rounds of Rep. Frank Pallone's Energy and Commerce inquiry (25 corporations in May 2026; eight airlines on 12 August 2026, answers due 25 August). 1,339 words, confidence: medium, 6 sources.
  • Colorado HB 26-1263 (Chatbot Safety Act) — score 5 (live thread +3, US AI regulation +2). Colorado's Chatbot Safety Act, codified at C.R.S. § 6-1-1708, had zero mentions anywhere in the wiki despite the wiki carrying Colorado SB 26-189 (2026 — replaces 2024 Colorado AI Act) and Colorado AI Act (SB 24-205) and despite the 12 August digest covering the rulemaking that implements it. Built from the signed-act PDF (10 pp.) plus the General Assembly bill page and the Department of Law's statement of basis: full definitions including the twelve exclusions from "conversational artificial intelligence service", the age-estimation trigger, the minor-protection duties, the disclosure-format rules (persistent visible disclaimer / intermittent audio / every three hours), the crisis protocol and its express exclusion of law-enforcement referral, the professional-equivalence prohibition, and the annual-reporting duty from 1 July 2027. 2,486 words, confidence: medium, 4 sources. Depends on two queued sources; carries a > **Provenance note:** line.
  • Decart — score 5 (live thread +3, frontier-lab governance +2). Zero prior mentions. Reuters and Bloomberg reported on 12 August that Anthropic is in early-stage talks to acquire it at a possible ~$6B. The company itself is substantial independent of the rumour: $300M in May 2026 at ~$4B led by Radical Ventures with Nvidia as a new investor, >$450M total raised, three product lines (DOS, Lucy, Oasis), Amazon as a strategic customer. 892 words, confidence: medium, 4 sources.

Pages expanded (live)

  • Grok (xAI) — score 5 (live thread +3, frontier models +2). The page's lead, description, and infobox all stopped at "Grok 4 (July 2025)" as the current flagship while the lineage table already ran to Grok 4.5, and Grok 4.6 — released 12 August 2026 — had zero coverage. Added: a lineage row; a ### Grok 4.6 subsection carrying the full ten-row developer benchmark table against Grok 4.5, GPT-5.6 Sol Max and Fable 5 Max; the training description (longer supplemental run, Grok 4.5 used to regenerate the SFT trajectories, agentic RL across kernel optimization / web dev / CAD); the safety paragraph, which records that the announcement names no evaluation, threshold or third party and that no system card accompanied it; and availability and pricing. Frontmatter repaired: description, last_updated, sources_count 9 → 10. 1 new (Source: URL) cite across five insertions.
  • Crusoe — score 5 (rotation slice +2, compute core area +2, in-degree 6–9 +1). The strongest slice-0 thin anchor: 292 words, in-degree 7, confidence: low, sources_count: 1, untouched since 2026-06-06, and carrying zero inline citations — the whole funding history reduced to "It has raised venture capital to scale its operations." Rebuilt to 1,183 words with a Snapshot block (Valuation and Compute sub-tables) and six sources: the full Series E investor list from Crusoe's own newsroom, the Abilene 1.2 GW expansion, the $11.6bn phase-two financing inside a $15bn joint venture, the Sacra account of the Crusoe-builds / Oracle-operates / OpenAI-rents structure and its revenue estimates, the July 2026 ~$3bn / ~$30bn talks, the Atero acquisition, and the joint 1.0 GW Childress campus with Lancium. confidence low → medium. Every fact and wikilink from the original survives.

The three new pages were wired into the existing graph rather than left as leaves, and one pre-existing frontmatter defect was repaired:

No alias fixes were made. All twenty-nine multi-inbound broken targets were resolved individually; none has an existing page under another name (see Deferred).

Queued — foundational sources

  • Colorado Department of Law, "Automated Decision-Making Technology & Chatbot Safety Act Proposed Rules — Statement of Basis, Specific Statutory Authority, and Purpose" (4 CCR 904-6, filed 2026-08-11) — score 8 (live thread +3, dangling foundational reference +3, core area +2). The implementing instrument for two statutes the wiki tracks, with no sources/ page and no raw file. Verified: coag.gov, the Department's own domain, linked from its rulemaking page. Saved: Raw Sources/Colorado Department of Law - ADMT and Chatbot Safety Act Proposed Rules Statement of Basis (4 CCR 904-6, 2026-08-11).md (full text). Queued: INGEST-colorado-admt-chatbot-rules-2026.md, which names three companion documents (the proposed rules themselves, the notice of hearing, the June pre-rulemaking considerations paper) for fresh fetch.
  • Colorado HB 26-1263, Signed Act — score 8 (same scoring). Verified: leg.colorado.gov bill-file store, reached from the official bill page. Queued URL-only, no raw file saved: the PDF parse dropped the introductory clause of § 6-1-1708(2)(c), the introductory clause of subsection (5), part of exclusions (VII)–(VIII), and subsections (6)(b)–(6)(c). The document is authentic; the extraction is not complete, and the authenticity protocol forbids saving an excerpt in place of full text. Queued: INGEST-colorado-hb-1263-signed-act.md with six specific recovery targets.

Source-fidelity findings

Five discrepancies surfaced this run. None was silently resolved.

  1. Two official Colorado sources disagree on when HB 26-1263 was signed. The Attorney General's rulemaking webpage states 1 July 2026. The General Assembly bill history records "Governor Signed" on 29 May 2026, and the Department of Law's own statement of basis — filed 11 August — states "On May 29, 2026, Governor Polis signed House Bill 26-1263." The July date is corroborated nowhere. Both are recorded on the page, attributed, with the outlier identified.
  1. A CMS date artifact on crusoe.ai would have produced a wrong year. The Series E newsroom page renders "August 4, 2026" as its page date while the press-release body datelines "October 24" and every independent source (Reuters, techfundingnews, private-markets profiles) puts the round in October 2025. Taken at face value the page date would have dated a $10B valuation sixteen months late. The October 2025 date was used.
  1. Selective compression in the 12 August digest's Grok 4.6 item. The digest recorded the two benchmarks on which Grok 4.6 trails its peers (DeepSWE v1.1, Terminal-Bench v3.0) and the one on which it leads (Harvey LAB), but omitted that it also leads both named peers on GDPVal-AA v2 and AA-Briefcase. The full ten-row table is now on the page, with the developer-assembled provenance of the comparison stated.
  1. Decart round metadata conflicts across sources. The company describes the May 2026 round as a Series B; some trade coverage calls it a Series D. CTech puts the prior August 2025 round at $100M, the Wall Street Journal at $153M, both at $3.1B. Founder counts also differ (two per CTech, three per WSJ). All recorded as attributed conflicts rather than resolved by picking one.
  1. The Colorado statement of basis mis-cites its own predecessor statute. It refers to the Anti-Discrimination in AI Act once as "SB 24-204" and twice as "SB24-205". The Colorado AI Act is SB 24-205. Preserved verbatim in the raw file with an explicit instruction not to normalize at ingest.

Authenticity-verification failures

None. Two sources pulled, two verified. One (the HB 26-1263 signed act) was queued URL-only rather than saved, for extraction incompleteness rather than authenticity failure — recorded in proposed-sources/colorado-hb-1263-signed-act.md.

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

  • The escaped-pipe scanner bug is FIXED — closing a ten-run carry. Flagged on every gap scan from 2026-07-26 through 2026-08-08 as an outstanding one-line defect in bin/lint-scan.py. It is no longer present: the current lint-report.md contains zero targets ending in a backslash, and its top-targets list matches this run's own post-strip scan exactly. The 334-target figure that circulated early in this run came from /tmp/scan.py, a stale scratch script left over from a previous session, not from the wiki's tooling — the two were conflated in this report's first draft and the claim is retracted. The remaining 251-versus-265 spread between lint-scan and this run's scan is methodological: lint-scan resolves _meta/ targets by basename and this scan does not. No action needed; this item should stop re-surfacing.
  • Slice-0 thin anchors not reached (score 3–4 each). Four remain from the five the slice surfaced: companies/coinbase (in-degree 10, 328 words, and the folder's sharpest confidence mismatch — confidence: high on sources_count: 2); companies/liquid-ai (8, 592, medium/1); companies/fairly-trained (8, 303, low/1, and separately parked in needs-review/ since 2026-07-16 as misfiled); companies/baseten (6, 485, medium/1). All four carry last_updated: 2026-06-06.
  • companies/riot-platforms (score 3) — carried from 2026-08-12. The 20-year, 191 MW Rockdale supply agreement Bloomberg identified as a $9.1bn Anthropic contract still has no company page.
  • Compute-as-collateral and GPU rental futures (score 3). Nvidia's 10 August memorandums with six firms targeting >$500bn against compute collateral, and CME Group and Silicon Data's 5 October launch of H100 and B200 rental-index futures, together describe compute becoming financeable and hedgeable. concepts/private-credit-ai-infrastructure, concepts/circular-financing-ai and concepts/ai-tokenization-finance each touch a corner; none covers the exchange-traded instrument. companies/silicon-data has no page.
  • The twenty-nine multi-inbound broken targets, and why none is a page gap this run. Thirteen are author entity pages implied by sources/ pages (Lee Anne Fennell, Kevin Klyman, Orin Kerr, Eric Goldman, Michael J. D. Vermeer, Alondra Nelson, Sacha Altay, Joseph Bernstein, Clayton Christensen, Adam Smith, Samuel Weinbach, Ilhan Scheer, Dina Powell McCormick), each at in-degree 2 — the same set deferred on 2026-08-09 and 2026-08-12, and a standing question for the curator: does the wiki want author pages for every cited author, or not? Seven are institutions named in passing (carnegie-mellon, santa-fe-institute, sais, uw, federal-reserve, ecb, bank-of-england) at in-degree 2. Five are concepts forward-referenced as "(planned)" (talent-flow-china-us, data-broker-regulation, a-vision-of-democratic-ai, economic-possibilities-for-artificial-intelligence, ai-and-language-models). Two (inverse-cooking-problem, inverse-trust-problem) remain curator-blocked in needs-review/ as coined terms. Alias resolution confirmed no existing page covers any of them.
  • Named in the 48-hour window with no page (score 1–2 each). Organisations: HateAid (German digital-rights group, criminal complaint over Meta's Ray-Ban glasses), Silicon Data, Dream (Israeli security firm, Taiwan agent-attack disclosure), Snyk, MATS Research, NEC. People: Frank Pallone (now cited five times on the new surveillance-pricing page), Caitlin Kalinowski (OpenAI robotics lead to Anthropic), Michael Gordon (Crusoe COO/CFO, now on the expanded page), Brad Lightcap (departing OpenAI), Sean Camacho and Javier Mabrey (HB 26-1263 prime sponsors), Michael Dalton.
  • The 29 date-header pages in lint-report.md (score 1). Unchanged for three cycles. Some are legitimate procedural histories; the report does not distinguish them, which is itself the finding.
  • Recent-window items the nightly fold already handled, not gaps. Anthropic's EU AI Act watermarking (folded 08-12, with EU Code of Practice on Transparency of AI-generated Content created that run), Nemotron 4 (page created 08-12), OpenAI's $7bn tender at $852bn, Gemini's 1 billion MAU, Manus separating from Meta, CoreWeave's Q2 (page updated 08-12), Intel's upsized $20bn offering, Palantir's GSA objection, the Rounds–Hickenlooper robotics bill, the NIST NVD request for information, the Google leadership reshuffle.

One-line summary

Three pages created and two expanded live — the Colorado Chatbot Safety Act had zero prior coverage despite an active rulemaking, and concepts/surveillance-pricing closes a hand-off that Algorithmic Pricing and Antitrust had left dangling in its own text; two Colorado primary texts are queued for the curator's ingest; the ten-run escaped-pipe scanner carry is closed as already fixed; and New Developments Log/2026-08-12-2209-ai-developments.md still needs its nightly fold.

Categories: Briefings | briefing | gap-scan