Data, code, and supplementary archive for a cross-database natural-carbonation ML audit for low-clinker binders

Corrected version 2 computational reproducibility archive supporting a cross-database audit of literature-mined natural-carbonation machine learning for low-clinker binders. Dimensionally corrected chemistry descriptors, coefficient-to-coefficient external validation, reference-grouped split-conformal auditing, constituent-level virtual generation, strength-prior calibration, domain abstention, kinetic sensitivity, and four-model endpoint stability are included. Processed derivative datasets are redistributed under CC BY 4.0 with source attribution and change notices. No mixture in the archive has been experimentally validated or certified for practice.