ResearchBenchmarksAvoid on Avoid 20% corruption (test)Follow0.86Success RateRoDif0.06960.27480.480.6852Jan 31, 2026Evaluation ResultsMethodMethodLinksSuccess RateMode AlignmentRoDif2026.010.860.86CPO2026.010.770.7IPO2026.010.490.47SimPO2026.010.10.09