Agentic Reasoning and Task Execution on Controlled-domain Knowledge-work
0.464Mean Task Rubric Score (G)Offline AW Controller
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Offline AW ControllerPolicy Variant=AW, Seeds=3, Rollouts per task=3, Number of tasks=20, Total trials (N)=1802026.07 | 0.464 | 0.014 | 0.059 | 0.002 | 0.097 | 5.6 | |
| BasePolicy Variant=Base, Seeds=3, Rollouts per task=3, Number of tasks=20, Total trials (N)=1802026.07 | 0.45 | — | — | — | — | — |