ResearchBenchmarksConfabulation Detection on Human-crafted confabulationsConfabulation Detection on Human-crafted confabulationsFollow100Finance AUROCΓ9597.5100102.5Jan 26, 2026Evaluation resultsMethodMethodLinksFinance AUROCOverall AUROCMedical AUROCLegal AUROCCross-domain DegradationLocal ImprovementΓCalibration Method=Glo...Calibration Method=Global μ2026.01100—91.393.53.8—ΓCalibration Method=Loc...Calibration Method=Local, k=152026.01100—93.593.5——