Data and code supporting: Validation Design and Experimental Provenance Shape Apparent Machine-Learning Transferability in Legacy Mössbauer Spectroscopy

This record contains the source-verified dataset, frozen machine-learning analysis code, canonical model outputs, retrospective endpoint and coverage-audit materials, figure-generation/QC code, and reproducibility documentation supporting the manuscript “Validation Design and Experimental Provenance Shape Apparent Machine-Learning Transferability in Legacy Mössbauer Spectroscopy”. The primary analysis comprises 171 eligible states from 45 source publications organized into 37 reconstructed experimental-provenance groups. The primary endpoint is the operational high-field ferromagnetic fraction. The retrospective coverage audit is reported separately from the frozen primary analysis and did not modify the reported primary results. Raw Web of Science and Scopus database exports are not redistributed in this public release; verified record counts and SHA-256 fingerprints are retained together with a controlled re-execution protocol.