End-to-end Agent Loop Performance on MemoryAgentBench
80Pre-Verify Success RateStateFuse
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| StateFuseTasks=50, Variant=Conservative2026.07 | 80 | 100 | 50 | 20 | 0 | 2.4 | 0 | |
| StateFuseTasks=50, Variant=core2026.07 | 80 | 100 | 50 | 20 | 0 | 2.4 | 0 | |
| MV registerTasks=50, Variant=Conservative2026.07 | 80 | 100 | 50 | 20 | 0 | 2.4 | 0 | |
| Provenance surfaceTasks=50, Variant=Conservative2026.07 | 80 | 100 | 50 | 20 | 0 | 2.4 | 0 | |
| Raw logTasks=50, Variant=Conservative2026.07 | 80 | 100 | 50 | 20 | 0 | 2.4 | 0 | |
| OracleTasks=502026.07 | 80 | 100 | 50 | 0 | 0 | 2.4 | 0 | |
| StateFuseTasks=502026.07 | 60 | 80 | 50 | 40 | 20 | 2.4 | 1.2 | |
| MV registerTasks=502026.07 | 60 | 80 | 50 | 40 | 20 | 2.4 | 1.2 | |
| Raw logTasks=502026.07 | 60 | 80 | 50 | 40 | 20 | 2.4 | 1.2 | |
| CollapsedTasks=502026.07 | 40 | 60 | 50 | 60 | 40 | 2.4 | 2.2 |