Scores ML datasets for missing values, duplicates, label noise, imbalance, and sample count.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.