Checks batches of inputs against guardrails and returns safety validation results.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.