Evidence database

Claims decomposed into components, each showing what actually supports it — including the parts supported by nothing. Falsifiers, replications, contradictions, provenance and open obligations, all citable.

Available Public 0 published findings

Empty, and working

No finding has completed archival audit, prior-art comparison, independent replication where the claim requires it, reproducibility packaging, and security, privacy, licence and IP review. Until one does, it does not appear.

This is not a placeholder for a corpus that exists elsewhere and has not been copied over. There is no hidden index, and the machine endpoint returns the same empty set this page shows. Manufacturing entries to make the database look populated is the one thing that would make it worthless.

Four modes

Browse

Claims by status, scope and open obligations — never by popularity, which is not evidence and is not recorded anywhere here.

Inspect

Methods, evidence objects, coverage map, falsifiers, replication independence, contradictions, assumptions and limitations.

Trace

Provenance step by step with input and output digests, and an honest marker where a step is a sanitised attestation rather than an inspectable input.

Verify

Recompute digests, check artifacts, download a portable bundle that survives this website.

See it working end to end

One demonstration claim runs the whole length of the machinery. It is deliberately a claim that ends badly: stated too broadly, narrowed after a falsifier is executed and fails, challenged on its measurement, and then withdrawn — because the interesting question is not whether an evidence database can display a success, but whether it can display a retraction without erasing what came before it.

v1 — experimental, one component supported by nothing at all.
v2 — scope narrowed after a cross-machine falsifier failed. The failed component was removed rather than defended.
v3 — withdrawn. The measurement compared the apparatus with itself, so no version of it could have failed.

All three remain published at their original addresses, unedited, with their original digests. It is clearly marked as a fixture and excluded from every scientific index, feed, sitemap and endpoint.

What the database records that most do not

Object	Why it is first-class here
Evidence coverage	Which parts of a claim are carried by nothing. A claim-level badge may never imply support its components lack.
Falsifiers	Preregistered conditions that would refute the claim, with whether each has been run and what happened.
Replication independence	Eight dimensions, recorded and never ranked. Internal reproduction is not independent replication.
Evidence debt	Typed unresolved obligations. Never summed into a score — a missing replication and a known contamination are not interchangeable.
Contradictions	Evidence pointing the other way, kept visible rather than retired quietly.
Negative results	A falsifier that executed and failed is a result, published with the same identity and citation machinery as anything else.

How a claim becomes publishable · The status vocabulary · Machine endpoint