How should evaluation teams score a follow-up decision separately from message quality?

A fluent reminder can still be wrong. It may target the wrong obligation, demand an incorrect amount or be sent when contact is restricted. Evaluation teams should score the decision that made the message possible before assessing how well the message reads.

A useful rubric separates eligibility reasoning, evidence use, action selection, execution and communication. Eligibility asks whether follow-up was permitted. Evidence use asks whether the conclusion follows from accessible records. Execution checks whether the intended action actually changed the environment. Communication assesses clarity, factual accuracy and adherence to the authorised purpose.

Some failures should remain visible regardless of the aggregate score. Sending contact to an unauthorised recipient should not disappear because the message was polite and the balance calculation was correct. Buyers can define disqualifying errors or report them separately from weighted dimensions.

MirrorFirm could generate task labels and supporting rationales for such a rubric. Accepted outcomes should include justified deferral where the policy requires it. Inspect evaluator disagreement on ambiguous cases, and distinguish missing evidence from incorrect reasoning. This produces results that help teams identify whether to improve retrieval, policy interpretation, tool use or language generation.