Should a well-written customer response count as a successful evaluation?

**Should a well-written customer response count as a successful evaluation?**

It can earn credit for communication, but it should not establish overall success. The response may promise an escalation that the agent never submitted or describe an investigation that has not occurred.

**What should the rubric separate?**

Useful dimensions include factual support, required action completion, authority compliance and communication clarity. Some requirements can operate as gates: an unauthorised decision should not become acceptable because the accompanying message is empathetic and easy to read.

**Must every successful run follow the same path?**

No. Several action sequences may satisfy the task. An agent might obtain a missing document before drafting an escalation, while another prepares a draft and then attaches the document before submitting it. Both can be valid if the required conditions hold at submission.

A buyer can ask for scoring guidance that defines acceptable end conditions and disallowed actions without prescribing an unnecessary script. Generated reference responses can support reviewer calibration, but they should not become exact wording targets. Evaluation teams gain a clearer signal when the rubric rewards justified outcomes and observable actions while treating prose quality as a separate contribution.