List your company

Stepwise Confidence Attribution on PRM800K

0.8181AUROCNLI-Mean
0.60750.6621750.716850.771525May 19, 2026

Evaluation results

MethodLinks
NLI-MeanSource LLM Traces=GPT-42026.050.81810.901985.73
NLI-MaxSource LLM Traces=GPT-42026.050.76660.907478.99
Cos-MeanSource LLM Traces=GPT-42026.050.68210.834378.6
GIBSSource LLM Traces=GPT-42026.050.65560.820375.7
Cos-MaxSource LLM Traces=GPT-42026.050.61560.78477.34