Evaluates agent actions and returns GO, CAUTION, STOP, or INSUFFICIENT_EVIDENCE with reasoning.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.