Validates agent step outputs against expected type, length, and content rules.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.