Scores the probability of human error from task and environment data.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.