Supplementary aggregate results and verification code for "Large-scale benchmarking of open gender-inference systems in scholarly identity data"

This record contains selected aggregate results and verification code accompanying the study “Large-scale benchmarking of open gender-inference systems in scholarly identity data”. The study compares NomQuamGender, WGND-based approaches and a hierarchical WGND-based pipeline using 330,994 scholarly author records with biography-derived F/M reference labels. These labels are operational proxies, not independently verified observations of gender identity. The materials comprise two workbooks containing four main tables, the numerical values underlying four figures, and seven supplementary aggregate tables. They also include an independently runnable arithmetic and cross-table consistency checker, an aggregate validation report, file checksums, and data-free scripts for pooled evaluation of stored predictions. The public checker operates on the two released workbooks without requiring individual-level records. The other evaluators require appropriately authorised, locally held inputs that are not distributed. Their input schemas, execution instructions and limitations are documented in the accompanying text files. Author-level records, names, identifiers, biographies, token-specific results and diagnostic matrices by evaluation partition are not included. The released code does not implement or rerun the complete gender-inference pipeline. These materials support inspection of the principal comparisons and verification of the released aggregates, without claiming unrestricted or complete reproducibility. This is version 1.0, the first public release of the supplementary materials. Files are distributed individually. Download the required files into one folder without renaming them, together with LICENSES.txt, and follow the README instructions. References to v1.30 retained in the workbooks, checker or audit filename identify the companion manuscript and its verified workbook set, not a previous Zenodo release. Licensing The two Excel workbooks, accompanying text documentation, aggregate audit report and checksum manifest are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). The five Python files — verify_public_aggregates.py, evaluate_stored_predictions.py, evaluate_cultural_strategies.py, evaluate_confidence_thresholds.py and input_utils.py — are licensed under the MIT License. LICENSES.txt provides the file-by-file allocation and the complete MIT notice. These licences apply to different components of the deposit; they are not alternative licences for the entire collection. They cover only the rights held in the released materials and do not grant rights in third-party datasets or software that are not distributed in this record.Funding: These materials are part of project PID2024-156739NB-I00, funded by MCIU/AEI/10.13039/501100011033 and by ERDF, EU.