AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-07-19

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

Contents

Scanned

Recent window: 4 dev-log files (07-17 morning/evening, 07-18 morning/evening; the 07-18 evening file is unprocessed — tonight's fold first, per precedent), ~80 pages edited in last 48h. Rotation slice: 3 — concepts/ P–Z (74 pages).

Gaps actioned (6 of ~22 found)

New pages created (live)

  • None this run. The window's pageless actors were either created by the 07-18 nightly fold (Etched, OpenRouter, OpenEvidence, Sunday Robotics, HumansFirst) or sit in the unprocessed 07-18 evening log (PrismML/Bonsai 27B, Kalshi compute curve, Project Jupiter) — deferred to tonight's fold per the standing run-order precedent.

Pages expanded (live)

  • Fireworks AI — explicit 07-17 ingest-cycle handoff ("confidence low — single source, flagged for gap-identifier enrichment"). Added founding (2022; CEO Lin Qiao, ex-head of Meta's PyTorch team), the full Series D syndicate (led by Atreides, Index, TCV; Nvidia among participants), 40T+ tokens/day with >95% from customer-specialized models, and the named customer list (Uber, Shopify, Doximity, Elastic, GitLab, MongoDB) — resolving one open question and sharpening the Nvidia one. Every prior fact and citation preserved. Confidence low → medium; sources_count 1 → 4. [3 new (Source: URL) cites: company blog, BusinessWire release, TechFundingNews]
  • Sociotechnical AI Risk Governance — slice 3's most-relied misrated thin anchor (in-degree 30, high on sources_count: 1 — the stochastic-parrots misrating pattern, stale since 06-06). Added a "Position in the sociotechnical literature" section situating the Mulligan–Marda–Wang framework against NIST AI RMF 1.0's socio-technical characterization (nvlpubs primary PDF) and Weidinger et al.'s three-layer sociotechnical evaluation framework (arXiv:2310.11986), with the evaluation-layer vs. governance-intervention distinction drawn explicitly. Confidence honestly re-rated high → medium with a confidence note; sources_count 1 → 3. Backup in _meta/_revision-backups/. [2 new (Source: URL) cites]
  • TESCREAL Framework (Gebru and Torres) — slice thin anchor #2 (in-degree 13, medium/1): the page cited only the MIT TR summary, not the framework's fullest statement. Added the peer-reviewed First Monday 29(4) paper (April 2024) with its two documented additional arguments (the "interconnected and overlapping ideologies"/eugenics-lineage characterization; the undefined-system safety-testing argument and defined-tasks recommendation), plus a provenance note tying inline cites to the already-queued INGEST-tescreal-bundle-gebru-torres-2024. sources_count 1 → 2; confidence stays medium. Backup in _meta/_revision-backups/. [2 new (Source: URL) cites]
  • CXMT (ChangXin Memory Technologies) (enrichment carry, in passing) — the 07-18 scan's flagged founded: unknown field closed: founded 2016, Hefei, China's only domestic DRAM maker to reach mass production, with the Xiaomi/Oppo/Vivo/Transsion/Lenovo LPDDR supply-chain detail (SCMP explainer, May 20, 2026 — verified fetch). sources_count 5 → 6.
  • None this run. The ≥2-inbound broken targets remain _meta/ cross-namespace links (basename-resolvable in Obsidian), table-cell \| pipe-link artifacts (targets verified to exist), already-queued sources (trustworthy-agents-in-practice, claude-sonnet-5-system-card, grading-ai-2027-2025-predictions, how-will-openai-compete-evans), or standing carries.

Queued — foundational sources

  • Anthropic Alignment Science, "Agentic Misalignment in Summer 2026" (July 13, 2026) — explicit handoff from the 07-18 ingest cycle's Notes; type-1 dangling foundational reference named by title on Agentic Misalignment: How LLMs Could Be Insider Threats and Anthropic with supporting cites only. Verified on alignment.anthropic.com (full report fetched; authors Lynch/Hughes/Serrano/Kirk/Bowman; four failure modes, harmful-compliance vs. agentic-misalignment distinction, and six-lab model list confirmed verbatim). Queued URL-only (~100K-char report; fetch fresh at ingest): INGEST-agentic-misalignment-summer-2026-2026-07-19.md; trail in gap-scan/proposed-sources/.
  • Xi Jinping, 2026 WAIC keynote ("Joining Hands to Build a Just and Equitable System For Global AI Governance," July 17, 2026) — the 07-18 backlog carry, matured: Chinese AI Policy and World Artificial Intelligence Cooperation Organization (WAICO) cite the mfa.gov.cn transcript URL inline as a supporting cite. Verified on both canonical hosts (Chinese full text mfa.gov.cn, meta-PubDate 2026-07-17; official English translation fmprc.gov.cn); distinctive passages (5,000 training slots, WAICO Shanghai, six partner organizations, MAZU) corroborate the Reuters/WSJ coverage already folded. Full official English translation saved: Raw Sources/Xi Jinping 2026 WAIC Keynote - Global AI Governance.md. Queued: INGEST-xi-waic-2026-keynote-2026-07-19.md.

Authenticity-verification failures

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

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

One-line summary

Two queued foundational sources close the window's loudest dangling references (the Anthropic agentic-misalignment follow-up, handed off by the nightly cycle, and Xi's WAIC keynote captured in full from both MFA hosts); three slice/handoff thin anchors expanded with corroborating primaries (Fireworks low→medium, sociotechnical-risk-governance honestly re-rated high→medium, TESCREAL anchored to its First Monday paper); CXMT's founded field closed — user review needed only on the two queued ingests.

Categories: Briefings | briefing | gap-scan