Checks ML training labels for consistency and accuracy, returning quality scores.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.