Reproducibility materials for "Leakage-Audited Benchmarking Reveals Limited Evidence for Cross-Subject Auditory-Evoked EEG Vowel Perception Decoding"

Versioned evidence package for manuscript JNE-110683.R2. The archive contains complete Study 2 event-to-trial ledgers; control-condition preprocessing and retained-epoch identities; the numerically verified (1,094, 61, 308) analysis tensor; 36,102 primary leave-one-subject-out predictions across 33 complete replicas; re-executed XGBoost, Random Forest, and LightGBM outputs; five-seed records for five deep architectures; a 9,616-refit exploratory MDM training-size analysis; exact normalization, temporal-bin, and inductive MDM_EA specifications; source data and panel registries for all 25 panels across seven figures; executable scripts; portable tests; environment specifications; and SHA-256 manifests. The original OpenNeuro ds006104 dataset is cited and is not redistributed in this archive. 本存档中重新分发。