ResearchBenchmarksReward rollout alignment on 10-task benchmark T2: Cleaning WhiteboardFollow0.667Rollout ρSARM2-0.48844-0.188470.11150.41147Jun 9, 2026Evaluation ResultsMethodMethodLinksRollout ρSARM2standard_full=truestandard_full=true2026.060.667SARM2Spectral Embedding=exc...Spectral Embedding=excluded2026.060.556SARM2Multi-Goal=excludedMulti-Goal=excluded2026.060.333RW2026.060.222RobometerFine-tuning=trueFine-tuning=true2026.060RobometerFine-tuning=falseFine-tuning=false2026.06-0.111TR2026.06-0.444