Calibration on Mean macro-average across benchmarks
5.56Expected Calibration Error (ECE)CAGE-CAL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CAGE-CALType=Trained2026.05 | 5.56 | 83.61 | |
| Plurality share + Scaling-bin.Type=Post-hoc plurality, Technique=Scaling-binning2026.05 | 6.98 | 72.58 | |
| DiscoUQ-LLMType=Trained2026.05 | 7.08 | 73.46 | |
| Plurality share + IsotonicType=Post-hoc plurality, Technique=Isotonic regression2026.05 | 7.72 | 71.74 | |
| Plurality share + PlattType=Post-hoc plurality, Technique=Platt scaling2026.05 | 8.16 | 72.8 | |
| Scalar + GBTType=Trained2026.05 | 10.6 | 71.16 | |
| Plurality shareType=Post-hoc plurality2026.05 | 18.04 | 72.8 | |
| GraphCalType=Trained2026.05 | 19.5 | 70.31 | |
| Collab. Cal.Type=LLM-elicited2026.05 | 22.52 | 71.7 | |
| LLM-Cal (+topo)Type=LLM-elicited, Topology=Included2026.05 | 23.59 | 71.25 | |
| LLM-Cal (no topo)Type=LLM-elicited, Topology=None2026.05 | 23.78 | 70.6 |