Claim-level specificity control on LongFact pilot
724Claims EmittedCalibrated CSS
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Calibrated CSSPolicy=Calibrated, Model=GPT-5.4, Ex.=1402026.04 | 724 | 757 | 99.31 | 94.11 | 93.55 | 92.89 | |
| Oracle CSSPolicy=Oracle, Model=GPT-5.4, Ex.=1402026.04 | 723 | 757 | 100 | 94.4 | 94.4 | 94.4 | |
| Uncalibrated CSSPolicy=Uncalibrated, Model=GPT-5.4, Ex.=1402026.04 | 721 | 757 | 99.58 | 59 | 58.76 | 58.36 |