evidence database
The laboratory
The publication infrastructure itself: claim records, evidence coverage, falsifiers, provenance, immutable versions and the deterministic policy that decides what may be published.
What it is
A machine-operable scientific publication system. Claims are the unit of publication, not papers. Each claim has a stable identity, immutable numbered versions, an evidence coverage map, registered falsifiers, typed evidence debt and a citation that resolves to the exact version you read.
Why it exists
Publishing infrastructure built after a first result tends to be shaped by that result. This was built first, and proven on a fixture that ends in withdrawal.
What it lets you do
See a claim decomposed into components, each with what actually supports it — including the parts supported by nothing — plus its registered falsifiers, its open evidence debt and everything that changed since the last version.
Recompute a record's digest from the record itself, check an artifact's bytes against its published hash, or check the whole site against one manifest.
Look up exactly what a status term means, in a versioned document, instead of guessing from the word.
Follow a claim's provenance chain step by step, with input and output digests, and see which steps are sanitised attestations rather than inspectable inputs.
Limitations
- The scientific corpus is empty. No finding has completed release review, so there is nothing to browse but the machinery and one clearly-marked fixture.
- There is no signing key and no transparency log. Both were designed and deferred with stated activation conditions.
Interfaces
| Kind | Reference | Note |
|---|---|---|
| http api | /api/claims.json | Published scientific claims. Empty by design. |
| http api | /api/changes.json | Scientific state changes, including the ones that weaken a claim. |
| document | /methodology | How a claim becomes publishable. |
Related
- exposes api — Public API
- documented by — Developer and API portal
Documentation
Machine record
Licence: CC BY 4.0.
Human and machine records are generated from one source in one pass and are tested to agree on availability, access, release state and limitations.