Methodology
How the composite is built from four pillars across five patent offices, and how the US drill-down measures are counted from the USPTO's own records.
Summary
Each pillar uses one metric, scored 0 to 100 across five patent offices. Parity sits at the peer median, and the four distances are averaged at equal weight.
Overview
How it works
The index has two layers. On top is a composite of four pillars: cost, time, independent inventors, and foreign share. We measure each one the same way at the patent offices of the United States, Europe, China, Japan and South Korea, then express the result as a single signed number between −100 and +100. Below that, the independents pillar opens onto the full US drill-down, which we count from the USPTO's own records.
Each pillar rests on one quantity the offices themselves publish. One definition covers all five, with no per-office adjustment, and where an office does not publish it we leave the cell empty instead of estimating a value.
The parity line is the median score of the four peer offices, and the US distance is its own score minus that line. The composite is the plain mean of those four distances, so all four count equally. The two figures below run one pillar through that arithmetic, then show the four distances it produces.
Every cell carries a source citation and a quotation of the passage the number came from. The independents figures were collected twice from the same brief, and the result was challenged independently before publication.
The US drill-down draws on two USPTO datasets, both named in the sources table. One covers every utility patent granted since 1976. The other records how applications moved through examination, and it stops in mid 2023.
One pillar, end to end
How a published number becomes a score
Every pillar runs through the same three moves. Here is the Time pillar, which measures the average wait for a final decision, with all five offices on the ruler.
Parity 61.45
The score is the ruler renamed, not a new measurement: the fastest office is set to 100, the slowest to 0, and everyone else falls where their published figure puts them. The parity line is the median of the four peer offices, so two of them sit above it and two below. The United States scores 0 on this pillar because it is the slowest of the five, and its distance is the gap from its own score to that line.
Source: WIPO World Intellectual Property Indicators 2025, Table A62; corroborated against the IP5 Statistics Report.
Four pillars, one number
The headline is the average of four distances
Each pillar produces one distance from its own parity line, on the same −100 to +100 scale. The composite is their plain mean: no pillar counts for more than another.
Cost
5 of 5 offices
Time
5 of 5 offices
Independents
3 of 5 offices
Foreigners
5 of 5 offices
The composite
the plain mean of the four
Equal weights are a choice, not a finding. They are stated here so anyone can reweigh: because each distance is published on its own, a different weighting can be worked out from this figure without new research. The right half of the scale is empty because no pillar puts the United States above its parity line.
Indicators
The four pillars
Each pillar has one metric, scored 0 to 100 across the offices that publish it: all five for cost, time and foreign share, three for independents. The figure beside each label is the US distance from the peer-median parity line.
Cost
−21.39Official fees to obtain and hold one representative patent to month 120, in USD, priced 2026. The United States charges $5,440, against a floor of $2,549 in Japan and a ceiling of $14,319 in Europe.
Time
−61.45Average wait for a final decision, in months, 2024. The United States takes 29.5 months, the slowest of the five, against 12.9 in Japan.
Independents
−23.05Whether an independent inventor gets into the register at all: the share of an office's home-country grants that go to a person rather than a company, 2020. The United States sits at 8.2%, against 19.8% in South Korea and 3.8% in China. Only these three offices publish the figure, so this pillar compares the US with two peers instead of four.
Foreigners
−64.59Share of an office's granted patents that go to applicants based outside its territory, 2024. The United States sits at 55.50%, level with Europe at 55.67% and far above China at 10.20%.
Coverage
Where the number exists, and where it does not
One definition per pillar, applied unchanged to all five offices. When an office does not publish the figure, the cell stays empty: nothing is estimated to fill it.
United States
Europe
China
Japan
South Korea
Cost
5 of 5 offices
Time
5 of 5 offices
Independents
3 of 5 offices
Foreigners
5 of 5 offices
Three pillars compare five offices. Independents compares three, because the other two do not publish the share of grants going to a person rather than a company. That pillar still enters the composite, measured against the peers that do publish it, and the narrower base is the reason it carries the caveat below.
Indicators
US drill-down measures
Counted from the USPTO's own records, which cover every application and every grant.
Inventor held share
Share of granted patents with no company named on them, the standard stand-in for an independent inventor. The figure peaked at 22.3% in 1980 and stood at 5.1% in 2025.
Grant rate
Share of applications eventually granted, by applicant size, pooled over the 2005 to 2017 filing years: large 58.4%, small 41.7%. The gap never drops below 14 points and reached 20.6 in 2015.
Abandonment
Share of applications given up before any decision, same years, across 6.65 million applications: small 27.7%, micro 25.3%, large 18.5%. In 2022, small and micro applicants walked away from 27,751 applications without answering the examiner, which works out to one every 18.9 minutes.
Pendency to grant
Median days from filing to grant, 2015 filing year: large applicants 809 days (253,485 grants), small 762 (66,129 grants). Small applicants who reach a decision are granted slightly faster, so the barrier is surviving the process rather than examiner speed.
AI actor split
Who receives AI patents, 2025: companies 12,809, universities 615, individuals 42, government 36. An inventor employed by a company counts under the employer, so "individual" means a patent held by a person in their own name.
AI filing share
Share of published applications naming AI in the title, by filing quarter: 0.62% in 2018-Q1, 2.56% in 2024-Q2, a 4.1× rise.
The window
Why the process measures stop at applications filed in 2017
The drill-down reads two USPTO datasets that end in different places. One records every granted patent up to the end of 2025. The other is a snapshot of how applications moved through examination, taken in mid 2023.
Granted patents
9.45M utility patents
Filing years measured
grant rate · abandonment · pendency
An application filed in 2017 has had six years to reach a decision by the time the examination snapshot was taken, so almost all of that cohort has an outcome to count. A 2020 cohort would still be part-decided, and counting it would understate both the grant rate and the abandonment rate. Measures that read the grant record instead, such as inventor held share, run to 2025.
Sources: USPTO PatentsView (granted patents, 1976 to December 31, 2025) · USPTO Patent Examination Research Dataset, PatEx, 2022 vintage, snapshot mid 2023.
Sources
Data sources
Every figure comes from the office that publishes it. On July 31, 2026 we recomputed the core findings from the raw files in 111 automated checks, and every number matched.
What we tested
The headline moves under every alternative. It does not turn positive.
Two choices could have made the number: where the parity line is drawn, and which pillars count. Changing either moves the headline. Neither reverses it.
Best of the five
the strictest line
Median of the four peers
the rule the index publishes
Mean of all five
the most lenient line
Drop the weakest pillar
published rule, three pillars
The first three rows change the parity rule and nothing else. The fourth keeps the published rule and drops the pillar with the narrowest base, the one a reader is most likely to challenge; without it the headline sits further from parity, not closer. The limitations below are the ones this exercise cannot settle.
Caveats
Known limitations
- 01The drill-down records what happened to applications and says nothing about why. A gap in grant rates is no evidence of examiner bias, and an abandoned application shows only that the applicant stopped.
- 02Scores describe a position inside the five-office panel rather than an absolute standard, and the pillars do not share a unit: a point of cost does not convert into a point of time. Parity is also a median, so an office can be level with a peer and still sit far from the line. The United States at 55.50% is within rounding of Europe at 55.67%, yet the peer median falls between Japan and South Korea, leaving both well below it. The same arithmetic puts Europe at −64.97, slightly worse than the US.
- 03The five offices start their pendency clocks at different events, and none of them publishes a harmonized figure. Correcting for that would mean replacing a published number with an estimate of our own, so the gap stands as the offices report it.
- 04Cost counts official fees only, the component every office sets in law and publishes. Professional fees are excluded because no office publishes them on a comparable basis, which makes this pillar a floor on what a patent costs rather than the whole bill. It is also the pillar most sensitive to the choice of parity rule, because Europe's fee ceiling sits so far above the other four.
- 05Two pillars carry a direction the index sets rather than one the numbers settle: a register where most grants go abroad is treated as serving the domestic inventor less, and a high share of individual grants is treated as openness to one. Both conventions are declared here rather than left implicit in the scoring.
- 06Two measures need reading with their scope in mind. Inventor held share counts patents with no company named at grant, and some are assigned to a company afterwards, so the fifty-year decline is firmer than any single year. The AI figures cover one deliberately narrow class of patents, and applications filed after mid 2024 are still publishing under the eighteen-month lag.