# claimset

> An open format for published facts that a stranger can check. A claims.json
> file records, for each load-bearing fact: the verbatim sentence, the source
> URL it was taken from, the date it was confirmed present, and the date the
> publisher stops vouching for it. A verifier re-fetches every source and
> reports PASS, DRIFT, EXPIRED, WEAK, UNREACHABLE or SUPERSEDED.

The distinguishing property is re-verification: not "is this link alive" and not
"who made this file", but "does the cited source STILL say what the citation
claims". Comparison is substring containment after a defined normalisation
pipeline, so two implementations cannot disagree about a result.

Status: version 1 of the format, draft but stable enough to build against.
Reference verifier: one Python file, standard library only, MIT licensed.
No published package yet; the public repository is publishing soon.

## Pages
- https://claimset.org/: the problem, the file format, the verifier output, quickstart links
- https://claimset.org/spec: the full specification, rendered verbatim from SPEC.md
- https://claimset.org/spec.md: the same specification as raw Markdown
- https://claimset.org/schema/1/claims.schema.json: JSON Schema 2020-12, normative for structure
- https://claimset.org/quickstart: write claims.json, run the verifier, wire it into CI
- https://claimset.org/examples: a worked example verified live against federal sources
- https://claimset.org/faq: how this differs from fact-check verdicts, provenance and link checkers
- https://claimset.org/claims.json: this site's own claimset, verifiable against this origin
- https://claimset.org/license: MIT

## A note on this file
This file is self-attested: I wrote it about my own site, and you have no way to
check it from here. That is exactly the property claimset exists to replace. The
checkable version of this page is https://claimset.org/claims.json — fetch it, run the
verifier, and confirm the claims against the pages that serve them.