Skill Retrieval on SkillResolve-Bench 1.0 (held-out pairs)
91.1R@20SkillResolve
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SkillResolveK=202026.06 | 91.1 | 74.7 | 0 | |
| BGE rerank top20K=202026.06 | 87.7 | 48.2 | 87.9 | |
| Hybrid lexical RRFK=202026.06 | 87.7 | 58.7 | 87.9 | |
| Attribution-listwiseK=202026.06 | 85.3 | 65.7 | 0 | |
| SkillRouterK=202026.06 | 78.4 | 57.7 | 81.4 |