List your company

Failure prediction stopping on AppWorld

78.8Max Accuracyα-STOP
70.58472.71774.8576.983Jun 3, 2026

Evaluation results

MethodLinks
α-STOPPredictor=attention-based failure predictor2026.0678.874.40.0002
Plug-in ThresholdPredictor=attention-based failure predictor2026.0676.871.20.0074
FIRMBOUNDPredictor=attention-based failure predictor2026.0675.870.10.0129
EndtoEndRLPredictor=attention-based failure predictor2026.0673.969.40.0125
ALERT*Predictor=attention-based failure predictor2026.0673.2680.0198
LLM Judge (best)Predictor=attention-based failure predictor2026.0670.9——