Sentence-Level Confidence Prediction on SciFact
0.91AUROCVERDI LR ensemble
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VERDI LR ensembleCalls=1×2026.05 | 0.91 | — | |
| KMeans-outlier2026.02 | 0.544 | 0.696 | |
| Shape Coherence2026.02 | 0.522 | 0.701 | |
| Local Variation2026.02 | 0.515 | 0.689 | |
| Semantic entropyProtocol=NLI, single-pass2026.02 | 0.512 | 0.684 | |
| Logprob confidenceCalls=1×2026.05 | 0.511 | — | |
| Mean logp2026.02 | 0.505 | 0.677 | |
| Spectral Stability2026.02 | 0.475 | 0.725 | |
| Structure-feature2026.02 | 0.46 | 0.67 | |
| SelfCheckGPT2026.02 | 0.458 | 0.65 | |
| Structural Signal2026.02 | 0.335 | 0.64 | |
| Semantic-feature2026.02 | 0.255 | 0.602 |