Compares one prompt across models and returns a third-model quality score.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.