chimera_verify
Verify claims against evidence using lexical token-overlap scoring, returning verdicts of support, contradiction, or insufficiency. Supports NLI and LLM verification methods.
Instructions
Verify claims against evidence using lexical token-overlap scoring. Returns lexically_supported / lexically_contradicted / lexically_insufficient verdicts — IMPORTANT: verdicts are Jaccard token-overlap, not semantic entailment or NLI. Supplement with chimera_constrain for semantic checks. Vs. direct reasoning: provides explicit per-claim scores, curated verification_gold corpus matching, and prompt-injection attack-pattern detection.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| claims | No | Claims to verify. If omitted, extracted from text or envelope. | |
| text | No | Optional raw text used to derive claims. | |
| envelope | No | Optional envelope containing value or claims to verify. | |
| evidence | Yes | Evidence snippets or objects with text/content fields. | |
| corpus | No | Optional document pool for grounded verify (RAG). When supplied, the top retrieve_k snippets most relevant to each claim are retrieved (deterministic token-overlap) and used as evidence — you don't have to hand-pick the exact evidence. Merged with any explicit 'evidence'. | |
| retrieve_k | No | Number of corpus snippets to retrieve per claim (grounded verify). | |
| method | No | Verification method. 'lexical' (default): Jaccard token-overlap, fast/deterministic, no extra deps. 'nli': local cross-encoder NLI model (needs [semantic] extra). 'llm': Anthropic judge (needs [llm] extra + ANTHROPIC_API_KEY, non-deterministic). | lexical |
| namespace | No | default |