User Simulation on WildChat (In-Distribution)
8.62Turn CountPrompted Simulator
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Prompted SimulatorCondition=Scenario2026.05 | 8.62 | 68.5 | 87.4 | 17.8 | 41.2 | |
| Trajectory-Cond. SFTCondition=Scenario2026.05 | 7.37 | 81.4 | 83.2 | 28.7 | 24.5 | |
| Prompted SimulatorCondition=Cognitive2026.05 | 6.45 | 88.3 | 89.1 | 19.2 | 28.4 | |
| Cognitive Profile Conditioned SFT2026.05 | 5.19 | 95.1 | 83.8 | 29.1 | 14.1 | |
| Dynamic State Conditioned SFT2026.05 | 4.78 | 109.5 | 76.8 | 30.2 | 21.1 | |
| Prompted SimulatorCondition=Scenario, Length-Constr.2026.05 | 4.45 | 92.2 | 88 | 19.5 | 38.2 | |
| Ground Truth Data2026.05 | 4.31 | 113.3 | 85.4 | 48.4 | 11.5 | |
| Unconditioned SFT2026.05 | 3.55 | 85.6 | 91.1 | 32.3 | 11.3 |