List your company

Claim-level specificity control on HotpotQA pilot

630Claims EmittedOracle CSS
400.16459.83519.5579.17Apr 19, 2026

Evaluation results

MethodLinks
Oracle CSSPolicy=Oracle, Model=Claude Sonnet 4.6, Ex.=1402026.0463064810095.9995.9995.99
Calibrated CSSPolicy=Calibrated, Model=Claude Sonnet 4.6, Ex.=1402026.0462864899.0494.7593.9593.02
Uncalibrated CSSPolicy=Uncalibrated, Model=Claude Sonnet 4.6, Ex.=1402026.0462264899.3671.9171.4270.8
Calibrated CSSPolicy=Calibrated, Model=GPT-5.4, Ex.=1402026.0442947097.99088.0986.17
Oracle CSSPolicy=Oracle, Model=GPT-5.4, Ex.=1402026.0442647010089.9689.9689.96
Uncalibrated CSSPolicy=Uncalibrated, Model=GPT-5.4, Ex.=1402026.0440947098.2962.0961.1159.62