Science & evidence
The trust substrate
Everything else on this site rests on this. It is a pillar beneath Tiffany, not the whole of her identity — and it is the part that is complete today.
Evidence databaseClaims, coverage, falsifiers, provenance
MethodologyHow a claim becomes publishable
Status vocabulary85 terms across 16 dimensions, versioned
Policy as code15 rules that run in the build
VerificationWhat it proves, and what it does not
Research indexEmpty by design
What changedIncluding changes that weaken a claim
Worked demonstrationOne claim, stated to withdrawn
The commitments
- The claim is the unit of publication, not the paper. Papers bundle claims of very different strength behind one title and one date.
- No confidence score, ever. Status is a nine-dimension vector whose dimensions regularly disagree, and the disagreement is the information a reader needs.
- Popularity is not evidence. No counter, ranking or engagement metric exists anywhere in the data model.
- Nothing is deleted. Superseded and withdrawn claims keep their URLs and their original bytes, so a citation resolves to what was actually read.
- Negative results are first-class. A falsifier that executed and failed is a result with the same identity, versioning and citation machinery as anything else.
- Cryptographic validity is not scientific validity, and every page keeps them visibly apart.
Open problems and what is missing
The laboratory records what it still owes its own claims — missing replications, unexecuted falsifiers, measurement gaps, unresolved contradictions, untested boundaries — as typed objects rather than as a score. Today the only such record belongs to the demonstration claim, whose final state is that no endpoint currently exists that could measure the quantity it asserted. That is a legitimate scientific result and it is published as one.
On artificial general intelligence
Tiffany is a research programme, and questions about its objective belong in the open. The framing this laboratory uses →