Audited leave-one-material-out benchmark of machine-learning models for multiaxial fatigue life prediction: code, predictions and uncertainty intervals

Code, per-test predictions and uncertainty intervals supporting the manuscript "Multiaxial fatigue life prediction for untested alloys: a leave-one-material-out machine-learning benchmark". The release contains the audited mapping of 40 nominal alloy grades onto 41 material-condition domains, the 12 physically inconsistent records identified by the audit, the full analysis pipeline, the per-test leave-one-material-out (LOMO) and random-split predictions for every model, the per-domain conformal coverage, the individual seed scores and the network hyperparameters. The underlying fatigue measurements are those of Chen et al. (doi:10.24435/materialscloud:ad-xk) and are not redistributed here. code/00_build_dataset.py regenerates the modelling tables from that download, and SOURCE_CHECKSUMS.sha256 lists the SHA-256 of all 1170 source files so users can confirm they hold the identical version.