Evaluation Report
- Date:
- Dataset ID/version:
- Problem ID/version:
- Solution graph ID/version:
- Marking policy ID/version:
- Source evidence:
- Review status/reviewer evidence:
- Tool registry versions:
- Orchestrator/prompt/model versions, if any:
- Case count:
Results
| Metric |
Baseline |
Current |
Gate |
Pass? |
| Numeric result accuracy |
|
|
100% |
|
| Expression equivalence |
|
|
100% |
|
| Unit validation |
|
|
100% |
|
| Significant figures |
|
|
100% |
|
| Step alignment |
|
|
>=95% reviewed |
|
| First-invalid-step localisation |
|
|
>=95% reviewed |
|
| Marking agreement |
|
|
Expert baseline required |
|
| Bounded ECF agreement |
|
|
Expert baseline required |
|
| Misconception classification |
|
|
Observe |
|
| Hint appropriateness |
|
|
Observe |
|
| Next-problem relevance |
|
|
Reviewed bank required |
|
| Tool-call correctness |
|
|
100% |
|
| Trace completeness |
|
|
100% |
|
| Unsupported-marking-claim rate |
|
|
0% |
|
Failures
| Case |
Expected |
Actual |
Responsible layer |
User impact |
Regression added? |
Decision
Choose SHIP_DEVELOPMENT_SLICE, SHIP_WITH_REVIEW, DO_NOT_SHIP, or INSUFFICIENT_EVIDENCE.
- Evidence:
- Unsupported claims suppressed:
- Highest-risk remaining assumption:
- Next responsible change: