Gap Scan — 2026-05-24

Live run. The 5 new pages and 1 expansion below were researched and committed directly to their real Wiki/ folders (status: active, confidence: medium). One foundational source — the Robin Nature paper — is verified and queued at Wiki/queue/INGEST-robin-multi-agent-scientific-discovery-2026.md, awaiting a reviewable ingest.

Scanned

Recent window (last 48h): 9 New Developments Log/ files (2026-05-22-0309 through 2026-05-23-2205); ~90 wiki pages with last_updated 2026-05-22 to -24 (the bulk from the 2026-05-23 Ingest-Reflect cycles). Rotation slice: 3 of 14 — concepts/ P–Z (57 pages). Broken-link analysis: 23,910 wikilinks across 1,235 pages → 534 unique broken targets. After alias resolution, doc-artifact filtering, and dropping items already actioned/deferred in the 2026-05-22 and 2026-05-23 (×2) gap-scan reports and items with an existing Raw Sources/ file, ~18 genuine gaps remained. The 9 highest-scored were actioned (7 substantive + 2 alias-fix clusters).

Gaps actioned (9 of ~18 found)

New pages created (live)

• FutureHouse — San Francisco nonprofit AI lab; developer of the Robin scientific-discovery system. Live thread: surfaced in the 2026-05-23-2205 dev-log (the newest file, not yet folded by the nightly Ingest-Reflect cycle, which stopped at 2057). No prior wiki page or even a plain-text mention existed. Confirmed a nonprofit ("philanthropically-funded moonshot," futurehouse.org) → entities/, not companies/. Depends on the queued Robin source — cited inline as (Source: URL) with a Provenance note.
• The Democratic Matrix — gap type 4: the foundational source Building AI for the Democratic Matrix: A Technical Research Agenda for Normative Competence and Normative Institutions (Hadfield + Trivedi + Hadfield-Menell, Knight Columbia, March 3 2026) existed and explicitly listed **supports:** [[concepts/democratic-matrix]], but no concept page sat behind it. The Hadfield-Trivedi-Hadfield-Menell framing of democracy as a "dancing landscape" of co-adapting agents. Core safety/governance area.
• Meaningful Human Review — gap type 4: referenced from Sociotechnical AI Risk Governance (×2) and Deirdre K. Mulligan with no page. A well-established ADM-law standard (GDPR Art. 22, EU AI Act Art. 14, Washington State facial-recognition law, state ADMT laws) and its "false comfort" critique. Core governance area.
• The Long Industrial Revolution — gap type 2: named across AI as Social Technology, Henry Farrell, Cosma Rohilla Shalizi, and the Farrell-Shalizi source page with no page behind it. The periodization concept placing AI in a centuries-long lineage of social technologies. Deferred on the two prior runs.
• Australian National University (ANU) — Australian National University; academic-institution page. 5 live inbound links (Seth Lazar, Anticipatory AI Ethics (Lazar framework, including the Technological Horizon), the Lazar source). Alias-resolution confirmed it is not Anu Bradford. Deferred on the two prior runs.

Pages expanded (live)

• Responsible AI Deployment — thin anchor from the rotation slice: 314 words, sources_count: 0, in-degree 10. Expanded to ~900 words with a developer-vs-deployer rationale, a "Frameworks and regulatory anchors" section (NIST AI 600-1 Generative AI Profile; EU AI Act Article 26 deployer obligations; procurement-as-gate), a downstream-deployer-safeguards practice, and a sharpened open-questions set. 3 supporting (Source: URL) cites folded; sources_count 0→3; last_updated → 2026-05-24. Held at confidence: medium.

Link aliases fixed

• [[sources/claude-mythos-preview-system-card]] → [[sources/claude-mythos-system-card]] — 3 instances across 2 pages (Unverbalized Evaluation Awareness ×2, Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations ×1). The real source page is claude-mythos-system-card (it exists); the -preview- form was a dangling alias.
• [[concepts/jailbreaking]] → [[concepts/jailbreaking-red-teaming]] — 1 instance (IH-Challenge: A Training Dataset to Improve Instruction Hierarchy on Frontier LLMs (OpenAI, March 2026)), with the stale (planned) parenthetical dropped since the target page exists. Alias-resolution confirmed concepts/jailbreaking is not a missing page — the wiki's canonical slug is jailbreaking-red-teaming, which exists and resolves the bulk of "jailbreaking" links.

Queued — foundational sources

• "A multi-agent system for automating scientific discovery" (Ghareeb et al., Nature, 2026-05-19) — the Robin paper. The wiki's most concrete 2026 claim that AI has moved from research assistant to research agent; a foundational anchor for AI for Science. Authenticity verified — PASS: scraped directly from the Nature publisher page (the canonical host for a peer-reviewed paper); DOI 10.1038/s41586-026-10652-y resolvable and CrossMark-verified; 14-author list and FutureHouse/Oxford/Fordham affiliations corroborated across page body and metadata; independently corroborated by FutureHouse's own research-announcement page and the dev-log. Per the large-PDF rule, not saved to Raw Sources/ — the INGEST task carries the verified canonical URL plus the freely-accessible arXiv:2505.13400 preprint. Queued: Wiki/queue/INGEST-robin-multi-agent-scientific-discovery-2026.md. Verification trail: Wiki/queue/gap-scan/proposed-sources/robin-multi-agent-scientific-discovery-2026.md.

Authenticity-verification failures

None. One source verified — PASS (Robin / Ghareeb et al.). No sources rejected; nothing fabricated was saved.

Notes for the next lint / ingest pass

• Claude Opus 4.7 system card is not an open gap. The run-2 evening report (2026-05-23) deferred it as a QUEUE candidate (score 3). It already has a Raw Sources/Claude Opus 4.7 System Card.md file and an ingested sources/claude-opus-47-system-card page — dropped from this run's candidate list per the dedup rule (a Raw Sources/ file exists). The run-2 deferral entry can be retired.
• concepts/autonomy-certificates is flagged by both Levels of Autonomy for AI Agents (Feng-McDonald-Zhang framework) and the Feng source page as a "concept stub … may be merged into levels-of-autonomy-framework." Because the wiki itself is undecided whether it should be a standalone page, it failed the quality gate (genuine ambiguity) and was not created — left for a human merge/keep decision. See Deferred backlog.

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

• concepts/ai-risk-management — gap type 4 (3 inbound; the Mulligan source wants it as an anchor distinct from standards/nist-ai-rmf). Deferred: an INGEST-ai-risk-management-framework-2026-05-14 task is already queued and standards/nist-ai-rmf exists; a generic concept page now risks redundancy and rework once that ingest lands. Score 2.
• concepts/autonomy-certificates — quality-gate ambiguity (stub vs. merge into Levels of Autonomy for AI Agents (Feng-McDonald-Zhang framework)); needs a human call. Score 2.
• concepts/classification-institutions — gap type 4, the Hadfield-Weingast microfoundation; companion to the new democratic-matrix page. Score 2.
• concepts/ai-and-children — 3 inbound (Blackburn "one rulebook" source + Tumbler Ridge Families v. OpenAI); child-safety topical hole. Score 2.
• concepts/seeing-like-a-state — 3 inbound (Scott's analytic lens, used by Farrell-Shalizi). Score 1.
• concepts/chinese-ai-policy — 3 inbound; possible hub, possible alias overlap with China Generative AI Registration (da moxing bei'an / 大模型备案) — needs scoping. Score 1.
• concepts/fifth-amendment-and-ai — 2 live inbound (Farahany neurorights source). Niche. Score 1.
• entities/john-moolenaar — congressman; has a SCALE Act — Rep. John Moolenaar (April 2026) page named for him. Score 1.
• Missing entity pages (3–6 broken links each, carried from prior runs): david-mcdonald, amy-zhang, rakshit-trivedi, nik-marda, victor-zhenyi-wang, herbert-simon, brian-armstrong, jack-dorsey, nathan-delisle, pim-dewitte.
• Lint cleanups (not gap-identifier pulls): sources/anthropic-pwc-expanded-partnership (7 broken — likely supporting, folds into Anthropic); sources/trustworthy-agents-in-practice (6+3 — log.md records the page as both created and deleted). Flagged for lint.
• Carried foundational candidates: Topol Ground Truths review + the Nature Medicine editorial (lint "Suggested next sources"); OpenAI Erdős / unit-distance proof artifact.

One-line summary

Seven substantive gaps actioned: five new pages created live — entities/futurehouse (from the day's newest dev-log), the Knight Columbia concept anchors democratic-matrix and meaningful-human-review, the twice-deferred long-industrial-revolution and entities/anu — plus the thin responsible-ai-deployment anchor tripled in depth and the Robin Nature paper verified and queued; two link-alias clusters fixed, and the stale "Claude Opus 4.7 system card" deferral retired (a Raw Sources/ file already exists).