Skip to content

Guide: Check a claim

What is this?

Evidence lookup over extracted page content: retrieve blocks that lexically match a claim string (BM25), attach Merkle membership proofs when a signed root exists, and optionally run a separate heuristic stance pass.

occam_claim_check does not prove truth, falsity, or semantic absence from the page.

When should I use it?

You have a sentence to ground (“Does this source contain text relevant to X?”) and want cited blocks with verifiable membership in the signed extract — before you (or occam_attest) judge support vs refute.

Minimal example

{
  "name": "occam_claim_check",
  "arguments": {
    "claim": "Occam returns typed failures when extraction fails",
    "url": "https://example.com/docs"
  }
}

For many citations with a heuristic stance partition, use occam_attest — not cryptographic attestation; an unsigned aggregate tally over regex-driven classification.

Expected result

Field Honest reading
found: true A block cleared the BM25 relevance floor — read the text; judge support yourself
found: false No block cleared the floor
proven: true (only when found: false) Retrieval-complete negative: the leaf set was not token-truncated and no block matched — not proof the page omits the claim (paraphrase, images, unextracted regions still possible)
proven: null When found: true — field not used
verdict Always not_evaluated — claim_check does not classify support
matches[].proof Merkle proof of block membership in the signed extract, not claim truth
receipt Signed Receipt v1 for the page extract (verifiable via occam_verify)

Workflow

  1. occam_claim_check — retrieve evidence + Merkle proofs; stance not_evaluated.
  2. You read matching blocks and decide support vs refute.
  3. Optional: occam_attest — heuristic status (supported / contradicted / …) for batch gatekeeping; gate on status, not on BM25 score alone.

Next