Reproducibility package for provenance-aware inverse design of ultra-high-performance concrete

This archive provides the processed dataset, master analysis code, software environment record, data dictionary, and source-aware split manifest for the study “Provenance-aware inverse design of ultra-high-performance concrete from heterogeneous literature data”. The processed dataset contains 2,065 UHPC mixture records derived from the public Mendeley Data release by Malik et al. (2025), Version 2, https://doi.org/10.17632/czb7ww5pkz.2. Records were retained when numeric 28-day compressive strength and traceable source identifiers were available. The archive also contains metadata required to reproduce the source-aware evaluation and inverse-design workflow. The parent dataset is licensed under Creative Commons Attribution 4.0 International. This archive contains processed derivative data and analysis code and does not replace the parent database.