Datasets:
| # Evaluation Protocol | |
| Evaluate approaches against each sample oracle. | |
| Required metrics: | |
| - decision/status match | |
| - issue-type precision/recall/F1 | |
| - clarification necessity accuracy | |
| - expected question relevance | |
| - LAT structural validity | |
| - business-rule extraction coverage | |
| - governance decision correctness | |
| - anti-pattern violation rate | |
| Report by task family, channel family, modality, difficulty and issue type. Do | |
| not create or use splits. | |