Data and code for "Stable rankings, unstable shortlists: a cross-algorithm stability audit of SHAP feature attributions in composition-based band gap prediction"

This deposit contains the featurised dataset, analysis code, and machine-readable results for the short communication "Stable rankings, unstable shortlists: a cross-algorithm stability audit of SHAP feature attributions in composition-based band gap prediction," submitted to Materials Letters. Files included:- bandgap_features.csv: featurised dataset (4604 inorganic compositions, 132 Magpie descriptors, experimental band gap target)- extend_stability.py: Python script to reproduce and extend the stability audit on other Matbench regression targets- extended_results.json: machine-readable stability metrics from 45 fitted models (3 algorithms x 3 seeds x 5 folds)- featurize_log.txt: run log confirming successful featurisation Source dataset: matbench_expt_gap, distributed through Matbench (https://matbench.materialsproject.org)