IP Data Library: Free Intellectual Property — Patent & Innovation Data

Research-grade innovation measures computed from the complete US patent record — citations, originality & generality, firm patent panels — free, citable, reproducible, and honest about methodology. US patent data is public domain.

9,454,161US Patents

152,631,929
(152.6+ Million)Citation Links

540,995Patent Holders

1976–2025Years Covered

Download Data Read the Docs API Access AI & MCP

Counts measured on the library's data store at generation — never estimated. Vintage: PatentsView 2026-07-15 release (USPTO Open Data Portal), coverage 1976-01-06 to 2025-12-30.

What is this?

Per-patent and per-firm innovation measures for every US patent granted since 1976, built from USPTO / PatentsView bulk data (US-government public domain) and refreshed each PatentsView release.

Citations with explicit truncation flags, originality and generality (raw and bias-corrected), claims, grant lag, team size, technology fields — plus disambiguated assignee×year panels.

Snapshot-pinned Parquet bundles and a free REST API, with a producer-first citation on every download. One free ElkassabgiData key works here, on HF and on Econ.

# browse what is served — no key needed
curl https://api.ipdatalibrary.com/v1/bundles

# then load a bundle
import pandas as pd
df = pd.read_parquet("patent_measures.parquet")
df[["patent_id", "fwd_cites_5yr", "originality_adj"]].head()

What is in the library

Built & validated

Innovation measures

Per-patent and per-firm measures on all 9,454,161 granted patents (details below) — the dataset is computed, verified, and staged for launch.

Built & validated

Firm patent panels

1,485,924 assignee×year observations across 540,995 disambiguated patent holders: patent counts, citation-weighted output, mean originality/generality.

Arriving — downloads in progress

Full patent text

Claims, descriptions and abstracts for every patent since 1976 and every published application since 2001 — the substrate for claim-scope and text-similarity measures.

Arriving — downloads in progress

Ownership, value & disputes

Patent assignments (sales, mergers, collateral pledges), maintenance-fee renewals (the classic revealed-value signal), and litigation dockets.

Arriving — downloads in progress

Trademarks back to 1884

The second IP pillar: every US trademark application and registration — marks, owners, classes, assignments, and TTAB disputes.

LIVE

API & downloads

Snapshot-pinned bundles are downloadable now (below) with the free ElkassabgiData account — the same API key that works on HF and Econ. MCP tools for AI assistants follow the family rollout.

Downloads

Snapshot-pinned Parquet bundles, served with producer-first citation headers. Browse free; download with the free family API key (get one — an existing HF or Econ key works as-is).

v2026-07-16

Patent-level measures

All measures for every granted patent: windowed forward/backward citations with truncation flags, originality & generality (raw + bias-corrected), claims, grant lag, team size, tech fields. Parquet, 249 MB.

Download patent_measures.parquet

v2026-07-16

Assignee-year panels

1,485,924 assignee×year observations: patent counts, citation aggregates and tech-class shares per disambiguated holder per year. Parquet, 58 MB.

Download assignee_year.parquet

Programmatic access. The open endpoints /v1/bundles and /v1/stats list what is served. Every download response carries an X-Citation header; cite USPTO/PatentsView as the producer and this library for the computed measures.

curl -H "X-API-Key: YOUR_KEY" -o patent_measures.parquet https://api.ipdatalibrary.com/v1/bundles/v2026-07-16/patent_measures.parquet

The measures

Standard, literature-grounded, and computed transparently — every value traces to the formula and the vintage.

↗

Impact

Forward citations (total + fixed 5- and 10-year windows)
Backward citations, incl. self-citation counts by shared assignee
Truncation flags on recent grant years — right-censoring is disclosed, never hidden

◯

Breadth

Originality & Generality (Hall–Jaffe–Trajtenberg)
Reported raw and bias-corrected (Hall 2005); undefined values are null, never zero
CPC-subclass technology fields (WIPO sectors included)

▤

Patent & firm attributes

Claims count, application→grant lag, inventor team size
Government-interest (federally funded) flag
Disambiguated assignee panels by grant year

Methodology, honestly

Every measure follows the canonical literature — Hall, Jaffe & Trajtenberg (2001), Hall (2005) — and the full formulas ship with the data. Two disclosures researchers should know before comparing numbers across papers:

Technology classes. Originality and generality are computed over CPC subclasses (~679 classes), each patent assigned its primary class. The original NBER variables used ~400 USPC main classes; finer classes mechanically raise the scores, so levels are not directly comparable to the 2001 NBER files.

Truncation. Forward-citation counts are right-censored for recent patents. Fixed-window counts (5y/10y) carry explicit truncation flags whenever the window extends past the data's last grant date, and windows are measured from the citing patent's grant date — not the citation record's date field, which carries the cited patent's date.

Citation, always: United States Patent and Trademark Office — PatentsView (public domain); measures computed by the IP Data Library. Producer first, library second — the family standard.