A U.S. inventor–patent linkage dataset for National Inventors Hall of Fame inductees

This record contains the public-safe release package for “A U.S. inventor–patent linkage dataset for National Inventors Hall of Fame inductees”. The dataset provides curated inventor–patent linkages for inventors inducted into the United States National Inventors Hall of Fame (NIHF). The release is organized around a curated accepted inventor–patent linkage layer rather than a broad candidate pool. The curated accepted release contains 28,130 inventor–patent links, 26,957 unique accepted patents, and covers 625 NIHF inductees. A stricter subset is provided for high-precision analyses. Review, rejected, uncertain/excluded, and benchmark-discordant records are provided separately as diagnostic records and should not be merged with the accepted release as valid inventor–patent links. The public release package is organized into six folders: data/, validation/, process/, derived_metadata/, docs/, and reports/. The data/ folder contains the core inventor and inventor–patent linkage files. The validation/ folder contains independent validation samples and summaries. The process/ folder contains public-safe evidence and cluster-audit derivatives. The derived_metadata/ folder contains minimal citation metadata for accepted patents. The docs/ and reports/ folders contain data dictionaries, release manifests, protocol notes, third-party data notes, patent identifier handling notes, and quality-control reports. The release includes an independent stratified audit of accepted and excluded records. The accepted-link precision audit includes 300 sampled accepted records, of which 287 remained accepted after manual audit, corresponding to an audit precision of 95.67% (Wilson 95% CI: 92.73%–97.45%). The excluded/diagnostic false-negative probe includes 300 sampled excluded or diagnostic records, of which 41 were judged as potentially valid links after a new round of manual verification, corresponding to a false-negative probe rate of 13.67% (Wilson 95% CI: 10.24%–18.02%). The release also includes a minimal citation edge table for accepted patents. The citation edge table contains 1,017,689 forward citation edges and can be used to compute lifetime forward-citation counts and citation coverage summaries. Because citation_year is incomplete, the citation edge table should not be used as a complete annual citation-history dataset. Restricted InCoPat raw exports are not redistributed. The released files contain public-safe identifiers, release-layer labels, derived evidence flags, reason-code summaries, validation summaries, public benchmark comparison indicators, minimal derived citation metadata, data dictionaries, protocol documentation, and quality-control reports. Researchers with licensed access to InCoPat may use the documented search and derivation protocols to reproduce the restricted-input portions of the workflow. All patent identifiers should be treated as strings, including patent_id_clean, patent_id, focal_patent_id_clean, citing_patent_id_clean, and seed patent identifiers. Some identifiers may contain leading zeros; converting them to integers may break cross-file joins.