Scores LLM outputs against reference answers using ROUGE-1 F1.
Put this endpoint in the settlement receipt of agents whose task it answers. You pay only when the recommendation converts on-chain.