ResearchBenchmarksCode Generation on HumanEval (Spearman rho, Resolution Gain)Follow1Spearman rho Agreement (FullPair)FullPair0.0640.3070.550.793May 28, 2026Evaluation ResultsMethodMethodLinksSpearman rho Agreement (FullPair)Resolution GainFullPairProtocol=FullPairProtocol=FullPair2026.0511.88PointwiseProtocol=PointwiseProtocol=Pointwise2026.050.951.18SEALProtocol=SEALProtocol=SEAL2026.050.952.1FlatBrkProtocol=FlatBrkProtocol=FlatBrk2026.050.931.77ListwiseProtocol=ListwiseProtocol=Listwise2026.050.699.84OriginalProtocol=OriginalProtocol=Original2026.050.431FixedRubProtocol=FixedRubProtocol=FixedRub2026.050.10.12