ResearchBenchmarksReward Weighting Adaptation on Cancer (held-out λnk ∈ {1, 2, 3, 4, 6, 7, 8, 9})Follow102.49RewardKθ − Uˆ5.530830.702955.87581.0471Mar 2, 2026Evaluation ResultsMethodMethodLinksRewardKθ − Uˆ2026.03102.49KI − Uˆ2026.0383.62KI2026.0348.72MFM2026.0341.05Direct2026.0338.35CFM2026.0336.03Kθ2026.0330.63NLOT2026.039.26