A Reproducible Evidence-Audit Framework for Leakage, Feature Stability, and Transportability in Translational Omics Classification

Submission-aligned code, outputs, figures, supplements, and manuscript for a reproducible evidence-audit framework in high-dimensional translational omics classification. A naive global-selection workflow is contrasted descriptively with a guarded nested workflow; because the arms also differ in tuning and resampling, the contrast is not presented as an isolated causal leakage estimate. Public breast-cancer neoadjuvant-chemotherapy cohorts are used as a stress test: GSE25055 (discovery/internal validation), GSE25065 (same-platform same-study-family validation), and GSE41998 (cross-platform transportability sensitivity).