Code, data, trained models, and DFT inputs for: Conformal prediction quantifies the reliability limits of machine-learned band-gap screening in oxide double perovskites

Complete reproducibility archive for the manuscript "Conformal prediction quantifies the reliability limits of machine-learned band-gap screening in oxide double perovskites" (Mohiuddin, Kabir & Ferdousi, submitted 2026). Contains: the full ML pipeline (merge, ground-state deduplication, featurization, stacked GBDT training, CV+/Mondrian conformal calibration, screening, adaptive-conformal analyses); the merged 6,295-entry training dataset with per-row database provenance; per-sample predictions with conformal intervals; the tiered 104-candidate table with raw-database entry evidence; the pairwise label-noise table; VASP input files for the in-house HSE06 anchor calculations (POTCARs excluded per license); figure scripts; and the trained model binaries (CatBoost/XGBoost/LightGBM/ stacking/MAPIE pickles, Python 3.10). Models are exactly reproducible from the code and data with fixed seeds. Development repository: https://github.com/kabir-lab-du/sr2bbo6-conformal-screening