User Simulation on ConvApparel Efficient Matchmaker Ultra-terse Fast V2 (held-out agent)
25.4Avg. Words per TurnStatic, Agnostic
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Static, AgnosticZero-shot=true, Simulator Paradigm=Static, Agnostic (Standard)2026.05 | 25.4 | 92.1 | 26.8 | 8.8 | |
| Dynamic, AgnosticZero-shot=true, Simulator Paradigm=Dynamic, Agnostic2026.05 | 14.1 | 45.2 | 18.2 | 24.8 | |
| Static, AwareZero-shot=true, Simulator Paradigm=Static, Aware2026.05 | 13.2 | 43.5 | 16.4 | 32.6 | |
| Dynamic, AwareZero-shot=true, Simulator Paradigm=Dynamic, Aware (Ours)2026.05 | 11.8 | 40.1 | 12.2 | 40.9 | |
| Ground Truth DataZero-shot=true, Simulator Paradigm=Ground Truth Data2026.05 | 11.7 | 39.7 | 11.9 | 41.3 |