Checks agent outputs against length, content, format, and language constraints.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.