Gap Scan — 2026-07-08

Scanned

Recent window: 4 dev-log files (07-06 morning/evening, 07-07 morning/evening; the 07-07 evening file is not yet processed by the nightly cycle), ~59 pages edited in last 48h. Rotation slice: 6 — entities/ O–Z (~88 pages).

Gaps actioned (6 of ~18 found)

New pages created (live)

• Muse Image and Muse Video (Meta Superintelligence Labs) — top live-thread page gap (score ~5: +3 live, type-3 missing model page). The 07-07 evening dev-log led its frontier-models section with Meta's launch of Muse Image and preview of Muse Video — the first media-generation models from Meta Superintelligence Labs and the lab's second release overall — with no wiki coverage (the only Muse page was Muse Spark (Meta Superintelligence Labs)). Built from Meta's own announcement (capabilities, Arena No. 2 text-to-image/editing Elo claims as of July 5, availability across Meta AI/meta.ai/Instagram Stories US/WhatsApp, the Content Seal watermarking system and detection-tool preview, the log-linear combined-compute Elo observation) plus SiliconANGLE's coverage of the built-in coding and search capabilities. Back-linked from Muse Spark (Meta Superintelligence Labs). confidence: medium.

Pages expanded (live)

• Paul Christiano — slice thin anchor: 330 words at in-degree 17, high confidence on sources_count: 3, last_updated 2026-06-06, and a "Current role" that still pointed only at the renamed AISI. Added education (MIT 2012, Berkeley PhD 2017), OpenAI tenure dates (2017–2021), ARC founding year (2021), the NIST role description from the primary host (nist.gov/people/paul-christiano), the March 2024 staff-opposition episode (attributed), the continuation of his Head of AI Safety role through the CAISI transition, ELK / AI Safety via Debate / Deep RL from Human Preferences research contributions, and a new attributed "Positions on AI risk" section (2023 estimates). Every prior fact and citation preserved. sources_count 3 → 5. [2 new sources, 5 new (Source: URL) cites]
• US AI Safety Institute (NIST AISI) — the slice's most-relied thin-frontmatter anchor: confidence: high on sources_count: 1 at in-degree 56 (the stochastic-parrots pattern). Corroboration research surfaced a factual error: the page dated the AISI→CAISI renaming to February 14, 2025; the announcement was June 2025 (Commerce Secretary Lutnick), per nist.gov/caisi and FedScoop. Corrected with sources and an inline correction note, added the techpolicy.press safety-to-security framing (attributed), added Christiano's 2024 appointment as head of AI safety with the NIST primary-host cite, and added the missing [[entities/nist-caisi]] successor link to Relationships. sources_count 1 → 4 — the high rating now has source support. [4 new (Source: URL) cites]

Link aliases fixed

• None this run. The remaining high-count broken targets are the standing backlog: legacy pre-v4.3 _meta/ pointers ([[lint-report]], [[log]] — lint territory), [[claude-code]] (needs the user's products/-schema call), and in-degree-2 person/org targets deferred pending substance (below).

Queued — foundational sources

• "A Global Workspace in Language Models" (Anthropic, July 6, 2026) — type-1 dangling foundational reference on a live thread (score ~8): both 07-07 dev-logs carry it (the evening log adds Zvi Mowshowitz's "No Space Like J-Space" analysis), and Mechanistic Interpretability and Anthropic Interpretability Team (both edited <48h) describe the J-space/J-lens findings with only a supporting anthropic.com cite. Verified: anthropic.com research post scraped in full; paper at transformer-circuits.pub; open-source implementation and Neuronpedia demo confirmed; Dehaene/Naccache/Nanda commentary matches the dev-logs. Queued: INGEST-anthropic-global-workspace-j-space-2026-07-08.md (+ verification record in proposed-sources/).
• FLI AI Safety Index — Summer 2026 (July 7, 2026) — type-1 dangling foundational reference on a live thread (score ~8): Future of Life Institute (FLI) (edited <48h) carries the summer grades on an Inside AI Policy supporting cite, while the Winter 2025 edition is already an ingested foundational raw — the wiki tracks the series by name. Verified: futureoflife.org index page scraped in full; grades match the dev-log exactly; nine companies, 37 indicators, panel and June 3 evidence cutoff recorded. Report PDF → fetch fresh at ingest. Queued: INGEST-fli-ai-safety-index-summer-2026.md (+ verification record).
• "The Data Center Water Problem Is Soluble" (Robin Gaster, ITIF, July 6, 2026) — type-1 dangling foundational reference on a live thread (score ~6): named institutional report in the data-center environment cluster, cited only as a supporting URL on AI Environmental Impact (edited <48h; Data Center Siting / AI Power Politics also in window). Verified: itif.org publication page with full citation metadata (author, date, publisher); distinctive passages confirmed. Queued: INGEST-itif-data-center-water-soluble-2026.md (+ verification record).

Authenticity-verification failures

• None. All three queued sources verified on their canonical hosts (anthropic.com/transformer-circuits.pub, futureoflife.org, itif.org); the new and expanded pages were built from primary or established secondary hosts folded as (Source: URL) (ai.meta.com, siliconangle.com, nist.gov, en.wikipedia.org, fedscoop.com, techpolicy.press, reuters.com).

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

• Slice thin anchors below the top two — entities/rand (high on sources_count: 2 at in-deg 22, 602w), entities/stuart-russell (high/2, in-deg 17), entities/pauseai (sources_count missing, in-deg 12), entities/scsp (980w but src 1, in-deg 8), entities/pete-hegseth (high/2, in-deg 13). Score ~4 each.
• Low-confidence / stub slice pages at in-deg 6–9 — susie-wiles, sanjog-misra, seth-lazar, robert-chesney, sean-cairncross, project-liberty-institute, rakshit-trivedi, sen-warren, stanford-cyber-policy-center, ted-cruz, ted-lieu, radha-plumb (low/1), rishi-bommasani, packy-mccormick, stewart-baker, pim-de-witte, tom-brown, opc-canada (src 0). Confirm substance before expanding. Score ~3 each.
• American AI Exports Program 78-bids item (07-07 evening log) — Executive Order 14320 — Promoting the Export of the American AI Technology Stack exists; the fold is tonight's developments-log work, not a gap. Score n/a.
• d-Matrix (Nvidia partnership; Corsair in volume production June 9) — companies/ candidate (hardware), single news cycle so far; the Thrive Holdings rule — create if references recur. Score ~4.
• Kokotajlo retrospective ("Grading AI 2027's 2025 predictions") — [[grading-ai-2027-2025-predictions]] broken at in-deg 2 on Situational Awareness: A One-Year Retrospective — Nathan Delisle (LessWrong, June 2025); a queue candidate (essay) rather than an alias fix, but not live and below the cap. Score ~3.
• California AI safety commission bill — carried from 07-07; bill number still unidentified in available coverage. Score ~4.
• Duplicate pages entities/cdao.md / government/cdao.md (in-deg 57) and companies/reflection.md / companies/reflection-ai.md — merge/supersede remains lint territory, open since 06-28/07-06. Score ~4.
• [[claude-code]] (in-deg 7, broken) — perennial; needs the user's products/-schema call. Score ~4.
• concepts/foundation-models (in-deg 2, broken) — still needs disambiguation vs. Frontier Models / General-Purpose AI (GPAI). Score ~2.
• sources/anthropic-pwc-expanded-partnership (in-deg 6, broken) — supporting press release; lint/fold task, no sources/ page warranted. Score ~3.
• Person/org pages at in-degree 2 — pam-bondi, masahiro-mori, anil-seth, eric-horvitz, orin-kerr, alondra-nelson, palmer-luckey, tom-mitchell, soufan-center, federal-reserve, santa-fe-institute, bank-of-england, ecb. Confirm substance before creating. Score ~2–3.

One-line summary

Meta's Muse Image/Video got the models/ page its launch-day thread lacked; the slice surfaced (and fixed) a wrong AISI→CAISI rename date on the wiki's 56-in-degree AISI page while giving it and Paul Christiano the sourcing their ratings claimed; and three live-thread foundational texts — Anthropic's J-space paper, the FLI Summer 2026 Safety Index, and ITIF's data-center water report — were verified on primary hosts and queued for ingest review.