Gap Scan — 2026-07-21

Scanned

Recent window: 4 dev-log files (07-19 morning/evening, 07-20 morning processed by the nightly cycles; the 07-20 evening file left for tonight's cycle per precedent), ~77 pages edited in last 48h. Rotation slice: 5 — entities/ H–N (~110 pages).

Gaps actioned (5 of ~24 found)

New pages created (live)

• None this run. The window's pageless candidates (a possible Tennessee v. Meta litigation page; the Alphabet "Frozen v2" chip; the Zvi Mowshowitz / Nathan Lambert / Gary Marcus Kimi K3 essays) all sit in the unprocessed 07-20 evening dev-log and belong to tonight's fold first — deferred below. Alias resolution closed the one live broken-target candidate (companies/mistral — a link error, not a missing page).

Pages expanded (live)

• Mustafa Suleyman — the slice's top thin anchor: in-degree 33 at 492 words, stale since 06-06, with no coverage of the year's central development in his role. Added the MAI Superintelligence Team (formed November 2025, headed by Suleyman), the revised Microsoft–OpenAI agreement he described as being "set free" to pursue superintelligence with Microsoft's own researchers, data pipelines, and custom silicon, the Build 2026 seven-model launch under his leadership (MAI-Thinking-1 as first reasoning model; "the basic building blocks of a superintelligence"; the "truly self-sufficient… at the absolute frontier" quote; fourth-frontier-lab positioning), the Mayo Clinic healthcare foundation-model collaboration and healthcare-as-next-adoption-area prediction, and a new "Humanist superintelligence" position subsection with the human-progress-test quote. Facts kept consistent with the sourced record on Microsoft (a conflicting eWeek context-window figure was not folded). Every prior fact and citation preserved; Microsoft wired into Relationships. confidence stays high, now reinforced; sources_count 3 → 7. [5 new (Source: URL) cites]
• Howard Lutnick — slice thin anchor #2 (in-degree 15 at 353 words) sitting directly on the window's hottest live thread (Commerce vs. Chinese open-weight models; the June export-control episode). Added the legal anatomy of the June 12 Anthropic directive from Mayer Brown's analysis — the Is-Informed Letter form, its ECRA §4817(b)(1) + EAR §744.22(b) grounds, the first-ever treatment of an AI model itself as controlled technology, the departure from three BIS advisory opinions on remote access, the pending Remote Access Security Act, and the Legion v. United States (Anthropic export-directive challenge) challenge — plus the WSJ-reported Amazon-CEO trigger, the Bloomberg-published letter, the June 26 trusted-partner exemption letter (Mythos 5 only), and the June 30 withdrawal. New "Chip export policy and enforcement" section: the early-2026 approval of higher-tier Nvidia chip exports to China, the Moolenaar/Cotton–Huizenga/Liccardo congressional letters, and his April 2026 BIS enforcement-funding and higher-fines testimony. Every prior fact and citation preserved. confidence stays medium; sources_count 3 → 10. [10 new (Source: URL) cites]

Link aliases fixed

• [[companies/mistral]] → [[companies/mistral-ai]] (1 instance on Open-Weight Frontier Models; the stale (stub) marker retired — the target is a full page, high/9, 1,067 words). Remaining ≥2-inbound broken targets are _meta cross-namespace links, table pipe-link artifacts, already-queued sources, or standing carries — consistent with the 07-14→07-20 runs.

Queued — foundational sources

• Ben Thompson, "Who's Afraid of Chinese Models?" (Stratechery, July 20, 2026) — the 07-20 lint report's suggested source #1 and a type-1 dangling foundational reference on the live open-weight-policy thread: named by title on Ben Thompson with a supporting cite only, while its cybersecurity argument is load-bearing on Defensive AI Paradox and Open-Weight Frontier Models. Verified: canonical host stratechery.com, published_time 2026-07-20; free weekly Article, full text retrieved. Saved: Raw Sources/Who's Afraid of Chinese Models.md. Queued: INGEST-whos-afraid-of-chinese-models-2026-07-21.md.
• Stanford HAI, "The Commercial Landscape of AI Sovereignty Offerings" (Meinhardt, Pava, Yee, Landay; issue brief, July 15, 2026) — the 07-20 lint report's suggested source #2, matured: the 07-20 nightly cycle folded it onto AI Sovereignty citing only the companion news article. Verification surfaced a naming correction: the wiki's inline "AI Sovereignty Paradox" label is the news article's headline; the primary document's canonical title is as above. Verified on hai.stanford.edu (landing page + PDF on the institutional host; Landay's "calibrating interdependence" passage corroborated). Queued URL-only per the large-PDF pattern: INGEST-stanford-hai-commercial-landscape-ai-sovereignty-2026-07-21.md.

Authenticity-verification failures

• None. Both queued sources verified on their canonical hosts. One recurring non-failure: congress.gov returned an empty page body for H.R. 9619 (People-First Chatbot Act) for the fourth consecutive attempt (07-16, 07-18, 07-19, 07-21); last human-readable check (07-16) showed "text has not been received." Carried.

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

• Unprocessed 07-20 evening dev-log items — tonight's fold first, then gap-scan candidates: Bartz v. Anthropic $1.5B final approval with the $101.6M fee cut (Bartz v. Anthropic update); State of Tennessee v. Meta Instagram-addiction trial + the parallel four-state $1.4T federal case (litigation page candidate, score ~6 once folded); CAISI director Chris Fall resignation (NIST CAISI (Center for AI Standards and Innovation) fold); the administration's revived de facto-ban push on Chinese open models (Axios/WSJ; folds onto Open-Weight Frontier Models / US-China AI Competition: Different Races, Different Metrics); Alphabet "Frozen v2" chip (2028; companies/google-deepmind or a compute page); Z.ai 1-GW all-Chinese-chip data center; AMD Helios rack-scale shipments; Netflix ~300 AI-assisted titles; Kimi K3 weights public release scheduled July 27. Score ~2–6 each once folded.
• Kimi K3 essay cluster — Zvi Mowshowitz "On Kimi K3", Nathan Lambert "Kimi K3: The Open Weights Escalation" (Interconnects), Gary Marcus "China has all but caught up" — foundational queue candidates once tonight's fold names them (the Lambert "6 months to live" precedent). Score ~4 each.
• Slice-5 residue (two-per-run pacing): Lennart Heim (high/2 at in-deg 22 — misrating pattern), International Committee of the Red Cross (ICRC) (high/1, in-deg 15), Marsha Blackburn (high/1, in-deg 12), Mariano-Florentino Cuéllar (high/1), Knight First Amendment Institute (Knight Columbia) (286w at in-deg 16), Matthew Prince (high/2, 388w), Jake Sullivan (medium/2, 477w), Nathan Lambert (medium/2, 371w — will mature with tonight's K3 fold), Jam Kraprayoon (229w), Horizon Institute for Public Service (sources_count: 0 — lint frontmatter territory). Score ~3 each.
• entities/sean-plankey (in-deg 1, from Overview) — below the person-page threshold; joins the in-degree-2 person carries.
• H.R. 9619 (People-First Chatbot Act) primary text — carried (congress.gov unfetchable ×4). Score ~3.
• Substack-redirect URL upgrades (Japan AI Basic Plan revision, Australia mandatory standards, Malaysia consultation — 07-17 ingest Notes). Score ~2.
• Orphan-check sources-layer backlog (07-15): six state chatbot-act primary texts + the Model Spec blog post. Batch decision when the queue drains.
• Standing carries: claude-code/claude-cowork placement (curator-blocked), inverse-cooking/inverse-trust coined terms (needs-review 07-10), companies/reflection/reflection-ai duplicate (needs-review 07-15), companies/fairly-trained misfiling (needs-review 07-16), entities/cdao/government/cdao duplicate (lint), concepts/foundation-models disambiguation, concepts/safety-training-methodologies and concepts/a-vision-of-democratic-ai planned umbrellas, person/org pages at in-degree 2 (pam-bondi, alondra-nelson, eric-horvitz, orin-kerr, palmer-luckey, soufan-center, masahiro-mori, anil-seth, recalc-academy, et al.). Score ~2–3.

One-line summary

Two slice-5 thin anchors on live threads brought current (Suleyman's superintelligence turn; the legal anatomy of Lutnick's export-control episode), one stale alias fixed, and both of the 07-20 lint report's top suggested sources verified and queued — the Stratechery essay saved in full, the Stanford HAI brief queued under its corrected canonical title; foundational ingests await curator review.