Calibration on BBH
6.12ECECAGE-CAL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CAGE-CALType=Trained2026.05 | 6.12 | 90.48 | |
| Plurality share + Scaling-bin.Type=Post-hoc plurality, Technique=Scaling-binning2026.05 | 6.86 | 65.95 | |
| Plurality share + IsotonicType=Post-hoc plurality, Technique=Isotonic regression2026.05 | 7.96 | 65.37 | |
| DiscoUQ-LLMType=Trained2026.05 | 8.54 | 65.93 | |
| Plurality share + PlattType=Post-hoc plurality, Technique=Platt scaling2026.05 | 10.2 | 66.15 | |
| Scalar + GBTType=Trained2026.05 | 10.89 | 69.68 | |
| Collab. Cal.Type=LLM-elicited2026.05 | 11.59 | 64.71 | |
| LLM-Cal (no topo)Type=LLM-elicited, Topology=None2026.05 | 13.87 | 63.92 | |
| LLM-Cal (+topo)Type=LLM-elicited, Topology=Included2026.05 | 14.4 | 62.56 | |
| GraphCalType=Trained2026.05 | 19.77 | 71.53 | |
| Plurality shareType=Post-hoc plurality2026.05 | 23.87 | 66.15 |