Evidence Retrieval on Synthetic benchmark B=256 197 episodes, 907 queries
8.4Hit@2OSU-Mem
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OSU-MemConstruction principle=overlap2026.06 | 8.4 | 22.1 | |
| Flat + MMR2026.06 | 5.2 | 15.8 | |
| Coverage-Aware2026.06 | 4.9 | 15.7 | |
| Flat Vector2026.06 | 4.5 | 15.6 | |
| Disjoint Hierarchy2026.06 | 4.4 | 15 | |
| Tree Abstraction2026.06 | 4.4 | 15.7 | |
| OSU-Mem (disjoint)Construction principle=disjoint2026.06 | 3.4 | 13.8 | |
| Periodic Summary2026.06 | 1.3 | 9.3 | |
| Budget-Trunc Replay2026.06 | 0 | 14.9 | |
| Sliding Window2026.06 | 0 | 13.6 |