Failure prediction stopping on ALFWorld
94.6Max AccuracyEndtoEndRL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EndtoEndRLPredictor=attention-based failure predictor2026.06 | 94.6 | 77.5 | 0.1151 | |
| Plug-in ThresholdPredictor=attention-based failure predictor2026.06 | 94.6 | 88.7 | 0.004 | |
| α-STOPPredictor=attention-based failure predictor2026.06 | 94.3 | 89.6 | 0.0046 | |
| LLM Judge (best)Predictor=attention-based failure predictor2026.06 | 94.3 | — | — | |
| FIRMBOUNDPredictor=attention-based failure predictor2026.06 | 93.9 | 86.9 | 0.0171 | |
| ALERT*Predictor=attention-based failure predictor2026.06 | 93.2 | 85.9 | 0.0208 |