Entry Types
The database tracks four types of entry, each with specific standards:
Capability
An AI system has demonstrated it CAN do something. Must be supported by peer-reviewed research or established preprints.
Incident
A specific real-world event involving AI systems. Must be documented by credible reporting or official records.
Governance Gap
A verifiable fact about oversight or regulation. Must reference official policy documents, legislation, or authoritative analysis.
Landscape
Who is doing what in AI governance globally. Must reference organisational publications or credible reporting.
Source Verification Threshold
No capability entry should be marked as verified without meeting ALL of these criteria:
- Minimum 3 primary academic sources - peer-reviewed papers or established preprints (arXiv with institutional authors)
- Specific page/section references - not just "this paper says X" but "Section 3.2, p.14" with the relevant finding
- Established paper descriptions - what each paper IS (scope, methodology) and why it's authoritative (venue, authors, citation count)
- Sourced counterarguments - the strongest academic pushback against the claim, with its own citation
If an entry cannot meet this threshold, it must be honestly marked as needing further evidence. Credibility over impressiveness.
Confidence Levels
Evidence Status (Capabilities)
Capability entries are additionally rated by where the evidence sits on the lab-to-real-world spectrum:
- Demonstrated (Real-World) - Observed in deployed AI systems or real-world contexts
- Demonstrated (Lab) - Reproduced in controlled research settings
- Partially Demonstrated - Some evidence exists but results are mixed or limited
- Approaching - Capability is rapidly developing but not yet reliably demonstrated
- Theoretical - Predicted by alignment theory but not yet observed
Counterargument Policy
Every capability entry must include sourced counterarguments - the strongest academic or expert pushback against the claim. This is not optional. An unsourced counterargument is an assertion, not evidence.
Counterarguments are presented with:
- The core claim of the counterargument
- Who makes it (with a source URL where available)
- A response explaining why the counterargument doesn't fully negate the entry (where applicable)
This exists because credibility requires engaging with the strongest objections, not ignoring them. If an entry cannot withstand its own counterarguments, the entry is the problem - not the counterarguments.
Source Credibility Ratings
- Primary - The original research paper, official document, or first-hand source
- Secondary - Credible reporting on primary sources (quality journalism, institutional reports)
- Analysis - Expert commentary, policy analysis, or synthesis of multiple sources
Independence
This site is maintained by Sebastian Wood, a technology professional based in Colchester. It is not affiliated with any political party, AI company, think tank, or government body. It receives no funding from any organisation.
The evidence layer aims to be neutral and rigorous. The advocacy section is explicitly Sebastian's personal position and is clearly separated.
Corrections Welcome
If you find an error, a broken source link, or a missing counterargument, please get in touch. The database is only as good as its accuracy.