benchmarksfactneutralSignificance in disparity detection tracks each axis's gap against its own minimum detectable effect, with standardization improving correlation from 0.56 to 0.78Machine Learning02 Aug 2026http://arxiv.org/abs/2607.28608v1