Must an agent's assessment match one reference answer exactly?

**Should every successful answer use the same wording and conclusion?**

No. Some evidence sets support more than one defensible recommendation, particularly when a supplied policy leaves room for reviewer judgement. Exact narrative matching can reward imitation while penalising a well-supported alternative.

**What should remain fixed?**

The evaluator can require accurate factual claims, recognition of material contradictions and a clear separation between established facts and judgement. It can also specify conclusions that the evidence cannot support, regardless of wording.

**How can a buyer inspect the scoring approach?**

Request an assessment rubric with acceptable reasoning paths, disallowed claims and examples of uncertainty that must be surfaced. Ask how a concise qualified answer compares with a confident answer that overlooks a limitation.

For contested cases, retain a route for human adjudication and record the reason for any scoring decision. A generated task pack could include alternative reference rationales rather than a single model paragraph. This helps developers evaluate evidence-based judgement without treating stylistic similarity as a measure of correctness.