Wiki Gap Analysis — February 2026

Systematic review of coverage gaps across the wiki's 639 pages, combining crux gaps list output (existing pages needing insight extraction) with manual topic coverage analysis to identify both under-developed existing pages and entirely missing topics.

Executive Summary

The wiki has strong structural coverage (639 pages across 25 categories) but significant gaps in three areas:

Insight extraction: 386 pages (of 519 tracked) lack insights. 203 high-importance pages have zero insights.
Fast-moving topics: Chinese AI labs, test-time compute, frontier model transparency, and recent incidents are under-represented.
Category imbalances: Incidents (2 pages), forecasting (2 pages), and capabilities (11 pages) are notably thin relative to their importance.

Dimension	Coverage	Key Gap
Policy & governance	~70%	Implementation tracking, enforcement status
Safety techniques	~55%	Post-2024 methods, effectiveness evidence
Frontier evaluations	~40%	Specific benchmarks, scaling analysis
AI labs & orgs	~65%	Chinese labs, emerging players
Capabilities	~45%	Vision/multimodal, inference-time compute
Incidents	~5%	Only 2 incident pages despite many documented cases
Open-source safety	~15%	Downstream risks, audit methods
International governance	~40%	Deep coordination, compliance tracking

Part 1: Missing or Under-Covered Pages

A prioritized list of 100 missing pages is maintained in the Suggested Pages dashboard, ranked by numeric priority (1–100) based on how often each topic is mentioned across existing pages and editorial importance to AI safety coverage.

The top 10 gaps by priority:

Priority	Topic	Mentions	Status
100	AI Governance	183	Sub-topics exist, no umbrella page
99	Reinforcement Learning	81	No standalone page
98	Retrieval-Augmented Generation (RAG)	404	No page
97	GPT-4	193	No page
96	Claude (Model Family)	186	No page
95	Frontier Model (Concept)	161	No definition page
94	Training Data	111	No page
93	Safety Evaluations	106	No page
92	Misalignment Potential	110	110 EntityLinks, no page
91	Civilizational Competence	104	104 EntityLinks, no page

See the full list of 100 suggested pages for details.

Part 2: Existing High-Priority Pages Needing Insight Extraction

From crux gaps list, the top 20 existing pages with highest gap scores (importance × quality, zero insights):

Rank	Score	Imp	Qual	Page	Category
1	166	87	91	Intervention Portfolio	responses
2	162	85	91	Scheming & Deception Detection	responses
3	159	85	87	OpenAI Foundation	organizations
4	157	82	91	Capability Elicitation	responses
5	157	82	91	AI Safety Cases	responses
6	157	82	91	Bioweapons	risks
7	157	82	91	Multipolar Trap	risks
8	151	87	73	Intervention Effectiveness Matrix	models
9	151	90	68	Evaluation Awareness	responses
10	151	79	91	Reward Hacking	risks
11	151	85	78	Sleeper Agents: Training Deceptive LLMs	risks
12	149	85	75	AI Control	responses
13	149	90	65	Eval Saturation & The Evals Gap	responses
14	149	78	91	Pause Advocacy	responses
15	149	78	91	Sandboxing / Containment	responses
16	149	78	91	Structured Access / API-Only	responses
17	149	78	91	Compute Thresholds	responses
18	149	78	91	Tool-Use Restrictions	responses
19	149	78	91	Voluntary Industry Commitments	responses
20	149	85	75	Deceptive Alignment	risks

Recommended batch for insight extraction:

Run crux insights extract on these pages to populate the insights data layer. The responses category is most under-extracted (4/136 pages = 3% coverage).

Part 3: Category-Level Coverage Statistics

From crux gaps stats:

Category	Pages	With Insights	Coverage	Priority
responses	136	4	3%	CRITICAL
organizations	106	0	0%	CRITICAL
models	78	0	0%	HIGH
risks	60	0	0%	HIGH
people	40	0	0%	Medium
intelligence-paradigms	16	0	0%	Medium
capabilities	11	10	91%	Low (well covered)
debates	11	3	27%	Medium
metrics	10	8	80%	Low (well covered)
cruxes	5	5	100%	Done
future-projections	5	4	80%	Low
incidents	2	0	0%	HIGH (few pages)
forecasting	2	2	100%	Done (but few pages)

Key finding: 93% of pages (482/519) have zero insights. The insight extraction pipeline should prioritize the responses (136 pages) and organizations (106 pages) categories.

Part 4: Recommended Action Plan

Immediate actions (next sessions)

Create top-priority pages (priority 90+) via crux content create — see Suggested Pages
Expand adversarial-robustness stub via crux content improve
Run insight extraction on top-20 gap-scored pages

Short-term actions

Create priority 70–89 pages: DeepSeek, Hallucination, Multimodal AI, Jailbreaking, etc.
Run insight extraction on all 0-insight pages with importance >= 80 (203 pages)

Medium-term actions

Work through priority 40–69 pages from the Suggested Pages dashboard
Systematic insight extraction across remaining categories

Generated by crux gaps list, crux gaps stats, and manual topic coverage analysis on 2026-02-13.