Evaluates chain-of-thought quality for steps, transitions, conclusions, and self-correction.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.