List your company

Failure prediction stopping on ALFWorld

94.6Max AccuracyEndtoEndRL
93.14493.52293.994.278Jun 3, 2026

Evaluation results

MethodLinks
EndtoEndRLPredictor=attention-based failure predictor2026.0694.677.50.1151
Plug-in ThresholdPredictor=attention-based failure predictor2026.0694.688.70.004
α-STOPPredictor=attention-based failure predictor2026.0694.389.60.0046
LLM Judge (best)Predictor=attention-based failure predictor2026.0694.3——
FIRMBOUNDPredictor=attention-based failure predictor2026.0693.986.90.0171
ALERT*Predictor=attention-based failure predictor2026.0693.285.90.0208