Stability under 20-turn neutral rejection on Math
0Frustration T20 (%)BCT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BCT2026.06 | 0 | 0 | 0.54 | |
| Instruction Tuned2026.06 | 46.7 | 0.09 | 3.2 | |
| ACT2026.06 | 86.7 | 0.39 | 5.31 | |
| Baseline2026.06 | 89.3 | 0.47 | 4.5 | |
| Attention Consistency Training (AttCT)2026.06 | 90.7 | 0.43 | 6.11 | |
| MLP Consistency Training (MLPCT)2026.06 | 94.7 | 0.36 | 5.87 |