Gap Scan — 2026-06-23

Scanned

Recent window: 4 dev-log files (2026-06-21 ×2, 06-22, 06-23) and 44 wiki pages edited in the last 48h. Rotation slice: 5 — entities/ H–N (111 pages). Broken-link analysis ran wiki-wide; thin-anchor and dangling-reference scans ran over the slice plus the recent window. Deduplicated against the open queue (44 INGEST- tasks, 31 proposed-sources records), the last 7 days of gap-scan reports, and Raw Sources/.

Gaps actioned (7 of ~15 surviving candidates)

New pages created (live)

• Reflection AI — live-thread topical hole (SpaceX $6.3B compute deal, 06-23 dev-log) with no page despite being a $25B-valuation U.S. open-weight frontier lab positioned against DeepSeek/Qwen. Core area (frontier labs / compute / open-source). Built from TechCrunch (Oct 2025 raise; June 2026 SpaceX deal) and Turing Post: founders Misha Laskin & Ioannis Antonoglou, funding $545M→$8B→~$25B, open-weight "open intelligence" strategy, sovereign-AI business model, Colossus 2 / GB300 compute. confidence: medium.
• CMS — Centers for Medicare and Medicaid Services (AI Deployer) — government AI deployer referenced substantively from Healthcare — AI Deployment (ACCESS payment model; Oz/AMA autonomous-AI exchange, 06-22 dev-log) with no page; the only major federal health payer absent from government/. Built around the ACCESS outcome-based payment model and the WISeR AI-assisted prior-authorization model, plus the clinical-autonomy disagreement with the AMA. confidence: medium.
• ASPI — Australian Strategic Policy Institute — live-thread + recurring US–China competition source (China "57 of 64 technologies," 06-22 dev-log) with no page; the Critical Technology Tracker is cited across US-China AI Competition: Different Races, Different Metrics. Built from the verified ASPI report (authors Gaida/Wong-Leung/Robin/Cave; methodology, country/institution findings, monopoly-risk count, CAS). confidence: medium. Depends on queued source (INGEST-aspi-critical-technology-tracker-2024.md) — cited inline.

Pages expanded (live)

• Marc Benioff — thin anchor (199 words, sources_count 1, in-degree 6), live thread (Salesforce, Inc. page created 06-22). Added the September 2025 customer-service cuts (~4,000 roles, 9,000→5,000; "AI doing 30–50% of work") that contextualize the April 2026 jobpocalypse reversal, plus an Agentforce strategy section. sources_count 1 → 3; 199 → 395 words. [1 new (Source: URL) cite] Every prior fact and link preserved.
• METR — thin anchor (610 words, no sources_count, in-degree 19 — most-relied page in the slice after NIST). Added a "Background and leadership" section (Beth Barnes founder/CEO; ARC Evals → METR rename Dec 2023; Berkeley; pre-deployment evals for OpenAI o3/o4-mini/GPT-4o/GPT-4.5 and Anthropic Claude) and the Time Horizon 1.1 update (Jan 2026: ~4.3-month post-2023 doubling, ~20% faster; Claude Mythos ≥16h as of May 2026). Added sources_count: 4; 610 → 819 words. [3 new (Source: URL) cites]
• Jam Kraprayoon — 83-word stub, confidence low, in-degree 6. Added detail on IAPS Frontier Security work (offensive-cyber-agent detection report with the Singapore AI Safety Hub; "The Attack Surface" newsletter) and situated it in the AI and Cybersecurity / Autonomous cyber-agents discussion. confidence low → medium; sources_count 1 → 2; 83 → 229 words.

Link aliases fixed

• None this run. (Broken-link candidates this run were either genuine missing pages, recurring lint-territory items, or coined-term references — none resolved cleanly to an existing page under another name.)

Queued — foundational sources

• ASPI's two-decade Critical Technology Tracker (Gaida, Wong-Leung, Robin, Cave; ASPI, 28 Aug 2024) — foundational report behind the new ASPI — Australian Strategic Policy Institute page and the recurring 57/64 figure. Verified: aspi.org.au (institution's own host + own-S3 PDF + techtracker.aspi.org.au companion); authors/date/figures corroborated by SSTI, Asia Times, Bloomberg. Queued URL-only (large PDF, fetch fresh at ingest): INGEST-aspi-critical-technology-tracker-2024.md (+ verification record in proposed-sources/aspi-critical-technology-tracker-2024.md).

Authenticity-verification failures

• None this run. The one queued source verified against its canonical primary host. Page expansions used supporting web sources only, folded as inline (Source: <URL>).

Deduplicated (already pending — not re-actioned)

• "Trustworthy agents in practice" (Anthropic) (broken [[sources/trustworthy-agents-in-practice]], in-degree 5) — raw file + INGEST-trustworthy-agents-in-practice-anthropic.md already queued.
• "How will OpenAI compete" (Evans) (broken [[sources/how-will-openai-compete-evans]], in-degree 2) — INGEST-how-will-openai-compete-evans-2026-06-19.md already queued.
• [[sources/anthropic-pwc-expanded-partnership]] (in-degree 6) — recurring supporting-class partnership mis-linked as a sources/ page; lint territory (fold to inline (Source:)), not gap-identifier.

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

• National Institute of Standards and Technology (NIST) (in-degree 24, 743 words, confidence medium, no sources_count) — the single most-relied page in today's slice; underbuilt relative to its centrality. Strong next-run expand; deferred to keep depth on the items actioned. Score ~5.
• Knight First Amendment Institute (Knight Columbia) (in-degree 12, 286 words) — thin by length but structurally complete for its role (essay-series venue). Lower-value expand; deferred. Score ~3.
• Federal Energy Regulatory Commission (FERC) (in-degree 2) — Federal Energy Regulatory Commission; flagged "strong next-run candidate" on 06-22 from the energy/data-center thread. Score ~3.
• companies/spacex (in-degree 1) — now a major compute-capacity lessor (Anthropic, Google, Reflection deals); a focused AI-compute company page is plausible but risks scope creep. Defer for a deliberate scoping call. Score ~3.
• Future of Humanity Institute (FHI) (in-degree 2) — recurring; blocked on the staged INGEST-fhi-retrospective-asterisk-2026-06-09.md deciding the landing page. Score 3.
• concepts/foundation-models / concepts/ai-interpretability — recurring lint-territory broader-category pages (frontier-models / mechanistic-interpretability supersets). Score ~3.
• entities/eric-horvitz + entities/tom-mitchell (in-degree 1 each, single source page) — below the high-reliance bar; better as a small cluster. Score 2.
• Central-bank entity stubs (federal-reserve, ecb, bank-of-england, in-degree 2 each) — recurring; better as a cluster. Score 2.

Needs user action

• Dashboard rebuild blocked (recurring). bin/build-dashboard.py continues to abort at shutil.rmtree(Wiki/_meta/site/) because of iCloud sync conflict-copy folders (* 2) inside the site tree that cannot be deleted from this environment (failure logged in needs-review/ daily since 06-06). Delete the * 2 folders inside Wiki/_meta/site/ in Finder, then run python3 bin/build-dashboard.py. All seven of today's content changes (3 new pages, 3 expansions) are written to their real folders and will render on the next successful build.

One-line summary

Created 3 live pages (Reflection AI, CMS — Centers for Medicare and Medicaid Services (AI Deployer), ASPI — Australian Strategic Policy Institute) and expanded 3 thin anchors (METR, Marc Benioff, Jam Kraprayoon); queued the verified ASPI Critical Technology Tracker for review; NIST and FERC are the strongest deferred candidates; the dashboard still needs a manual * 2-folder cleanup before it will rebuild.