Selective Inference

FDR-controlled selection and honest post-selection inference (Candès school): Benjamini–Hochberg on valid p-values, fixed-X and model-X knockoffs for variable selection without p-values, conformal p-values + BH for outlier/anomaly lists, e-value derandomization for reproducible selections, and the select-then-infer trap (winner's curse) with its remedies. Use whenever many hypotheses/features/units are screened and the SELECTED SET carries the claim — "which features matter with FDR control", "flag anomalous employees/items", "this dashboard tests 50 cuts", "are these selected effects real", "my top performers regressed to the mean" — and as the audit when someone reports…