Full-pipeline transportability of lumbar MRI classifiers: Matched evaluation of robustness, adaptation, and deployment policies

This reproducibility archive supports the study “Full-pipeline transportability of lumbar MRI classifiers: Matched evaluation of robustness, adaptation, and deployment policies.” It contains grouped split manifests, resolved experiment configurations, seed-level and ensemble predictions, extracted features, post-hoc adaptation artifacts and outputs, label-shift analyses, validation-fitted temperature-scaling results, crop-geometry stress outputs, observation- and study-level operating-policy analyses, study-clustered and seed-by-study uncertainty outputs, checkpoint/environment metadata available in the supplied research bundles, unit-tested analysis code, manuscript result tables, figures, and SHA-256 checksums. Raw MRI/DICOM data and trained model weights are not redistributed. Source datasets must be obtained from their original repositories under their respective licenses and access conditions.