ResearchBenchmarksPolicy Alignment on Synthetic dataset 100 independent trialsFollow66Match CountIBMDP34.842.95159.1Jan 21, 2026Evaluation ResultsMethodMethodLinksMatch CountMatch RateIBMDPAction Selection=Top 2...Action Selection=Top 2 (two most frequent across ensemble)2026.016666IBMDPAction Selection=Top 1...Action Selection=Top 1 (most frequent across ensemble)2026.014747VI-SimType=Deterministic sim...Type=Deterministic similarity planner2026.013636