skills/fable-eval/examples/eval-regression-run.md

# Example: Prompt Improvement Evaluation

## Intervention
Optimized task router prompt for faster intent matching.

## Evaluation Results
- Baseline Accuracy: 92% (23/25 scenarios).
- Candidate Accuracy: 100% (25/25 scenarios).
- Holdout Set: 10/10 passed.
- Verdict: ACCEPT.