FAHCE Claim Verification and Repair

Choose FAHCE when you need evidence-grounded verification with an actionable, bounded repair plan rather than a simple similarity score or binary pass/fail. It is especially valuable when your pipeline produces LLM-generated text that must be audited against specific source documents before delivery, or when downstream consumers require a traceable correction audit trail. Prefer it over generic LLM self-critique loops because it grounds diagnosis in caller-supplied evidence, returns a structured machine-readable response, and provides a billing receipt suitable for automated cost accounting.

• Missing required fields (schema, evidence, or candidate_output) returns a 4xx validation error
• Evidence array is empty or too sparse to ground any claims — diagnosis may return low-confidence or indeterminate result
• candidate_output is extremely long and may hit token or processing limits
• Payment not received or x402 flow incomplete — request rejected before processing
• Contradictory evidence supplied by caller yields ambiguous repair plan
• Idempotency key collision returns cached prior result rather than re-processing