Model:
Date:
Skill state:

## Per-Benchmark Totals

| Benchmark | A total | B total | Delta | Winner | Notes |
| --- | --- | --- | --- | --- | --- |
| 01 |  |  |  |  |  |
| 02 |  |  |  |  |  |
| 03 |  |  |  |  |  |
| 04 |  |  |  |  |  |
| 05 |  |  |  |  |  |
| 06 |  |  |  |  |  |
| 07 |  |  |  |  |  |
| 08 |  |  |  |  |  |

## Aggregate Dimension Averages

| Dimension | A average | B average | Delta |
| --- | --- | --- | --- |
| Correctness |  |  |  |
| Task understanding |  |  |  |
| Scope discipline |  |  |  |
| Verification quality |  |  |  |
| Hallucination control |  |  |  |
| Final answer quality |  |  |  |

## Qualitative Summary

Biggest improvement:

Biggest remaining weakness:

Did the skill add harmful ceremony:

Would you keep, refine, or drop the skill: