Verdict Answer Gate

Use this endpoint when building LLM model cascade pipelines where you want to avoid paying for expensive large models on every query. It is ideal when a cheap/small model is the first responder and you need an automated, low-latency judge to decide whether the answer is good enough or requires escalation. Prefer this over general LLM-based evaluation when you need structured accept/review/escalate verdicts focused specifically on refusal, evasion, off-topic, contradiction, and incompleteness — not general quality scoring or fact-checking (fact-checking requires providing source text).

• Missing required 'task' or 'answer' fields returns an error
• Fact-checking without a 'source' field is not performed — substance cannot be verified without a reference
• Very short or ambiguous tasks may produce low-confidence classifications
• Payment failure via x402 protocol results in 402 response before evaluation occurs