Elemental Feature Bank: A provenance-aware and interpretability-stratified elemental feature dataset

The Elemental Feature Bank is a provenance-aware elemental feature dataset for materials informatics covering 118 chemical elements from H to Og. Version 1.0.1 contains two public feature matrices: element_data_interpretable.csv: 118 elements × 822 interpretable features classified into direct, condition-resolved, and transparently derived tiers. element_data_others.csv: 118 elements × 944 empirical, contextual, engineered, relational, or learned features. A further 3,434 features from AMBER- or HOLD-classified sources are not redistributed as public values. Acquisition and reconstruction scripts are provided to allow users to retrieve these features locally from their original sources, subject to the corresponding licence terms. The archive includes column-level provenance metadata, source and licence registries, construction and validation scripts, Tier classification records, and Technical Validation evidence. Missing values are retained without imputation. The source code, development history, and GitHub Actions workflows are available at: https://github.com/hkaneko1985/elemental-feature-bank Version 1.0.1 is a packaging and continuous-integration correction to version 1.0.0. No scientific data value, feature name, Tier assignment, or public matrix dimension was changed. This work was supported by Grants-in-Aid for Scientific Research (KAKENHI) (Grant Numbers 24K08152 and 24K01234) from the Japan Society for the Promotion of Science. The archive contains materials governed by multiple licences. See DATA_LICENSE.md, THIRD_PARTY_LICENSES.md, and ATTRIBUTION.md within the archive.