Grades chatbot answers against source docs for faithfulness, relevance, retrieval quality, and hallucination risk.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.