How should customers score a complaint task when more than one next action is defensible?

**Does every complaint task need one exact next action?**

No. An investigator might obtain an account note before reviewing a related service record, or reverse that order, without affecting the eventual decision. A single preferred sequence can penalise valid work if the policy does not require that sequence.

**What should replace an exact-match answer?**

A generated task can define acceptable action sets, prerequisites and prohibited outcomes. Scoring can distinguish a required condition from a suggested route. An escalation may require sufficient evidence and an identified authority gap, while allowing different ways to gather that evidence.

**How can a reseller inspect the rubric?**

Ask whether the evaluator explains why an alternative is acceptable and which observable facts would make it unacceptable. Include borderline cases in review: an extra evidence request might be harmless in one state but cause an avoidable deadline breach in another.

MirrorFirm could generate task rubrics with these distinctions. Preserve the rubric version when licensing tasks onward so customers can interpret score changes. This approach measures defensible complaint handling without assuming that every competent agent must reproduce a reference agent's exact path.