Control-level coverage prediction on OrgB policy corpus
82.31F1 ScorePROPARAG
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PROPARAG2026.05 | 82.31 | 84.91 | 80.81 | 84.84 | — | |
| PROPARAG2026.05 | 82.31 | — | — | — | 81.4 | |
| B4: Dense RAG + Labelretrieval=Dense RAG, mode=Label2026.05 | 70.96 | 72.18 | 71.64 | 72.18 | — | |
| Dense RAG + Label2026.05 | 70.96 | — | — | — | 68.9 | |
| B2: Retrieval + Labelmode=Retrieval + Label2026.05 | 67.54 | 68.73 | 69.12 | 68.73 | — | |
| Retrieval + Label2026.05 | 67.54 | — | — | — | 65.6 | |
| B3: BM25 + Labelretrieval=BM25, mode=Label2026.05 | 54.33 | 56.89 | 56.21 | 56.89 | — | |
| BM25 + Label2026.05 | 54.33 | — | — | — | 52.1 | |
| B5: Document-Levelgranularity=Document-Level2026.05 | 51.07 | 53.64 | 54.27 | 53.64 | — | |
| Document-Level2026.05 | 51.07 | — | — | — | 48.9 | |
| B1: Single-Shotmode=Single-Shot2026.05 | 49.62 | 51.88 | 52.31 | 51.88 | — | |
| Single-Shot LLM2026.05 | 49.62 | — | — | — | 47.3 |