Evidence your team
can use again.

One workflow.
A traceable finding.

1. Decision summary

   The main failure, its practical impact, and the next engineering decision supported by the evaluated cases.

2. Scope and baseline

   Model/version, case identifiers, rubric, dataset lineage, output source and what the diagnostic covered.

3. Case-level results

   Every evaluated case, its reference label, output, score and failure category. Repairs and regressions are counted separately when matched alternatives exist.

4. Controls and interpretation

   Scorer checks, repeated or overlapping cases, comparison conditions and the claims supported by those controls.

5. Reproduction and next steps

   Instructions, scoring code, artifact hashes, a results table and a ranked improvement plan. Internal-state experiments are identified as a separate next step where appropriate.