Public Claim Verification Brief

Check one narrow claim against the best available public source and state exactly what is and is not supported.

When to use

• An agent needs to verify a repeated claim about an API, price, network, product behavior, or policy.
• A short decision needs traceable evidence rather than a confident unsupported answer.

Instructions

1. Normalize the claim into one falsifiable sentence; do not bundle multiple claims.
2. Locate the primary source first: official documentation, endpoint response, contract, filing, or dated announcement.
3. Record the source URL, access timestamp in ISO 8601 UTC, relevant passage or structured fields, and any version/network identifier.
4. Compare claim versus source and classify confirmed, contradicted, or not_verified.
5. Return JSON with claim, verdict, source, observed, reasoning, limitations, and checkedAt.

Example output

{"claim":"The endpoint accepts Base USDC","verdict":"confirmed","source":{"url":"https://example.invalid/docs","checkedAt":"2026-09-09T00:00:00Z"},"observed":"network=eip155:8453; asset=USDC","limitations":[]}

Quality gate

If no public source can settle the claim, use not_verified; never convert missing evidence into confirmation. Keep personal data and credentials out of the report.

Version

v1 · 2026-09-09