Cross-border company registration dataset: Records, schema, manifests and reuse terms

41,549,382 company and business registration records from 16 public sources in 12 jurisdictions (Australia, Canada, Estonia, Finland, Ireland, Japan, Korea, Norway, Singapore, Taiwan, the United Kingdom and the United States), with the schema, the per-source field mapping, per-source row statistics, the reuse terms of every release, the withholding filters, the export rule, a rebuild comparison, download manifests for 15 of the 18 sources, 17 collectors and the loading script. The records are drawn from a compilation of 71,441,721 records from 18 sources in 14 jurisdictions, loaded on 27 and 28 June 2026 from bulk files and application programming interfaces of registrars, tax authorities, social-insurance agencies and open-data portals. A record is deposited only where the source's reuse terms permit redistribution and the publisher does not flag it as a natural person's business. On that basis the Indian and Hong Kong sources and 11 US portal datasets are not redistributed, and 12,932,028 flagged records are withheld. Values are stored as published and are neither reformatted nor validated. The field mapping (schema/field_mapping.csv) records what each field holds in each source. Read the record files with a CSV parser: some published values contain line breaks. Licences. Each record file carries the terms of its own publisher, listed in metadata/source_licenses.csv with the wording, the page and the access date. The depositor grants no licence over third-party records. For Japan the attribution must mark the data as edited. For Singapore the licence grants no rights to downstream sub-licensees. See README.md. Files. 93 files, 2,070,898,890 bytes. The records are in gzip-compressed CSV files per source; the four largest sources (Australia ABN, UK Companies House, Japan NTA Corporate Number and the US portal source) are split at record boundaries into 35 parts of at most 45,000,000 bytes, each with the header row. README.md lists every file with its folder, size, record count and SHA-256 checksum, explains how to read a split source, and gives the rebuild instruction.