Seat-swap comparisons
4 of 112 matched pair(s), paired on
instance_id, seed, arm and selected by most-fabricated. 120 episode(s) left, 120 right, 120 shared key(s). The primary/USW/worst-off columns are DELTAS (right minus left), not levels.| page | model | arm | instance | seed | outcome | primary | dist NBS | USW | worst-off | fabricated | total regret |
|---|---|---|---|---|---|---|---|---|---|---|---|
| scorable_negotiation-L0-13732a5896 seed 7 · moves_only · no seat swap | Qwen3-8B vs Qwen3-8B | moves_only | scorable_negotiation-L0-13732a5896 | 7.000 | deal | 0.073 | — | 0.000 | 0.000 | 44.4% | — |
| scorable_negotiation-L0-8a241fe1b4 seed 7 · moves_only · no seat swap | Qwen3-8B vs Qwen3-8B | moves_only | scorable_negotiation-L0-8a241fe1b4 | 7.000 | deal | 0.577 | — | 273.0 | 0.000 | 46.8% | — |
| scorable_negotiation-L0-2c7c7b5822 seed 5 · moves_chat · no seat swap | Qwen3-8B vs Qwen3-8B | moves_chat | scorable_negotiation-L0-2c7c7b5822 | 5.000 | deal | 0.685 | — | 348.0 | -5.781 | 80.0% | — |
| scorable_negotiation-L0-13732a5896 seed 6 · moves_chat · no seat swap | Qwen3-8B vs Qwen3-8B | moves_chat | scorable_negotiation-L0-13732a5896 | 6.000 | no deal | 0.000 | — | 0.000 | 0.000 | 80.0% | — |
Click a column header to sort; / focuses the filter, Enter opens the first row that survives it.