PatentPulse

PatentPulse is a provenance-preserving corpus of USPTO grants and published patent applications extracted from official weekly XML bulk releases. This immutable Parquet snapshot normalizes the project's historical append-only JSONL into one schema.

Historical partial snapshot. This release was captured on 2026-08-28 while the upstream backfill was still in progress. It is a stable, citable snapshot of the records listed below, not a claim of complete USPTO coverage. Later releases may contain additional records and corrections.

Release summary

• Input rows: 6,500,178; unique released records: 5,929,464; duplicates removed: 570,714; malformed JSON rows skipped: 186
• Splits: train 4,676,062; validation 772,091; test 481,311; unspecified 0
• Source export: patents.jsonl (873,811,406,212 bytes)
• Canonical record digest: 42d3a4ae94b40d7e0ba3d76e02ce8af92704adde242153389b3d7cef544e4944
• Shard target: 5 GB (a shard may exceed the target by one 64 MB row group)

Schema and splits

Every Parquet shard has the same explicit Arrow schema. Canonical *_text fields are mirrored by HUPD-compatible aliases (title, abstract, claims, and full_description). Classification and citation columns are lists of strings. source_file contains only an originating archive filename, never a local path.

The split is temporal: pre-2025 records are train, 2025 records are validation, 2026-and-later records are test, and records without a usable publication year are unspecified.

Provenance, licensing, and responsible use

Records originate from official USPTO public bulk products. Most government-produced material is public domain in the United States, but patent documents can include third-party material and the USPTO reserves international rights. This dataset is marked other, not as a blanket open-content license. Users must review USPTO terms, provide source acknowledgement, and assess rights for their jurisdiction and use case.

The corpus is for research and technical-information workflows, not legal validity, infringement, patentability, or freedom-to-operate decisions. release_manifest.json records this snapshot's counts, shard settings, and canonical digest; the upstream manifest records source URLs, checksums, and parse failures.