EvoXplain: attribution populations for the explanation-multiplicity analysis

Attribution populations underlying the EvoXplain explanation-multiplicity analysis. evoxplain_tcga_lr_cgrid_shap_lime_100x100.tar — 100 train–test splits × 100 elastic-net fits on a TCGA pan-cancer tumour-versus-normal task, with SHAP and LIME attribution vectors, held-out accuracies and regularisation parameters per fit (718 files, 979 MB). evoxplain_lewis_audit_faithful_individual.tar — 20 refits of a published biomarker classifier on resamples of its own cohort, with per-sample SHAP values (20 files, 650 MB). Analysis code, frozen outputs and per-file checksums: https://github.com/bensmailchama-boop/evoxplain-reproduction