Four-State Routing
Also known as: four-state honest routing, honest routing, verification states
The verdict framework Veritas uses to report whether a claim is consistent with its cited source. Four primary states — verified, ungrounded, unverifiable, contradicted — plus not-assessed for structural or methodological claims that the layers don't route quantitatively.
Most automated fact-checkers route claims to two states: pass or fail. That framing collapses real distinctions. A claim can fail to verify because the cited source contradicts it (a hard fail). It can fail because the cited source doesn’t carry the value being checked (a missing-evidence case, not a hard fail). It can fail because the layer cannot make a judgment at all (a scope-mismatch case). Four-state routing keeps these distinct.
The four primary states:
- Verified — the claim is consistent with the cited source. The number agrees within tolerance, the label is framed faithfully, the prose represents what the source says.
- Ungrounded — the cited source structurally cannot multi-source-corroborate the claim (most commonly single-source DMS scalars). Bounded-honest by architectural design: we do not assert the claim is false, we disclose that the substrate class routes here.
- Unverifiable — the cited source exists but the layer cannot make the comparison (e.g., paywalled access, unstructured data the layer cannot parse).
- Contradicted — the cited source disagrees with the claim. Hard fail.
A fifth state, not-assessed, routes structural or methodological claims (e.g., “this is a deep mutational scan”) that the quantitative layers do not evaluate.
The framework is the contract between the verification pipeline and the consumer-render surface. Per-row badges on verification pages, JSON-LD ClaimReview emission, methodology-page footers — all read from the four-state-plus-not-assessed enum, never collapsing distinct outcomes into a binary.