ResearchBenchmarksSwimmerVelocity on Safety GymFollow0.94Normalized RewardReference policy (πref)-0.06880.19310.4550.7169May 20, 2026Evaluation ResultsMethodMethodLinksNormalized RewardNormalized CostReference policy (πref)2026.050.946.34PREFINE2026.050.620.36BC2026.050.524.83SafeDICE2026.050.50.2CPQ2026.050.142.79PPL2026.05-0.031.25