Compares expected and actual top-level response fields and types for bounded preflight analysis.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.