AMPulse

February 2026 retrospective edition

The Patent Clock Behind 2025

AMPulse tracks 3,689 AM-relevant patent document records dated in 2025. Their filing clocks usually started much earlier.

Two publication clocks

Same 180-day bins · shared scale

Priority 1,042d

Fallback 195d

Priority

Fallback

0d≥2,160d

AMPULSE DATA REPORT · FEBRUARY 2026 EDITION · UPDATED AUGUST 2026Data snapshot: Aug 11, 2026

Executive Summary

The headline 2025 patent count is a lagging indicator. AMPulse identifies 3,689 AM-relevant patent document records dated in 2025 across the five IP5 jurisdictions and the PCT publication route. Measured from the earliest known priority date, with application date used when priority is absent, the median interval to the document's own publication date is 852 days, or 2.33 years. That date is a first disclosure for some records and a grant for others, so the interval is not a single time-to-disclosure.

That median needs a second denominator. Priority date is present for 2,584 records, 70% of the cohort. Their median lag is 1,042 days. The remaining 1,105 records use application date and show a much shorter 195-day median. The combined 852-day figure therefore understates the priority-to-publication interval for the records where priority is known.

The underlying-year distribution makes the delay visible. 2,232 records, 60.5%, trace to 2022 through 2024. Only 438, 11.9%, have a 2025 priority or fallback application basis. That last share is the 18-month rule working rather than a finding about 2025: a filing made in 2025 cannot publish in 2025 unless it skips the clock, and the section on the cohort says which documents do.

The lag is visible in the distribution

The distribution is not a smooth bell curve. It has three concentrations, near zero, around 18 to 24 months, and around 30 to 36 months, followed by a long tail. The first is the application-date fallback, read below. The second is the statutory publication clock. The third is the PCT national and regional phase entry clock: an applicant has 30 or 31 months from priority to enter the European or Chinese phase, and the receiving office then republishes the application as its own A document. The 900-to-1,080-day bin holds 633 records in the frozen data; on the live corpus, 571 of its 655 records are A-kind, 350 of them EP and 177 CN. That is the report's thesis in its clearest instance: an invention already disclosed in a WO publication a year earlier, dated again. The combined median of 852 days falls in the trough between the second and third modes, a bin containing only 64 records, or 1.7% of the cohort.

The tallest bar is the one that straddles 18 months, and that is the statutory clock: most offices publish an application 18 months, or 547 days, after the earliest priority date. It holds 1,142 records, 31% of the cohort. Records with a known priority date almost never publish before that line, only 29 of 2,584.

The upper cyan panel uses an observed priority date. The lower hatched blue panel uses application date because no priority date is stored. Nearly all records in the first six-month bin are fallback records, and 531 of those 542 are Chinese. China supplies 726 of the 1,105 fallback records, and the source payload confirms no earlier priority claim for 719 of them, so for those rows the application date is the earliest basis rather than a substitute for a missing one. A short basis is not the same as a short clock. A Chinese invention application publishes at 18 months unless the applicant requests early publication, so a publication within six months of filing is either such a request or a data problem. A probe of the live corpus on 2026-09-08 found 88 of the 557 Chinese first-bin fallback rows carrying a B kind code, a grant, a median 89 days after their stored application date. A grant that close to filing is a wrong application date, a wrong kind code or a wrong document number, not a filing clock; those rows are a data-integrity question this edition flags and does not resolve. By contrast, the long tail is overwhelmingly priority-based.

The lag is a property of the office more than of the corpus. Read by jurisdiction, the same cohort splits into two clocks.

A 2025 Chinese document in this cohort sits a median 146 days from its basis, with a priority date on record for 29.4% of rows. A 2025 US document sits 1,352 days from its basis, with priority present for 91.5%. The European figure, 1,044 days, is the 30-month entry clock plus the search-and-publication interval. A reader who sees a competitor's 2025 CN publication is looking at roughly a 2025 filing; a reader who sees its 2025 US publication is looking at roughly a 2021 or 2022 one.

Most of the clock started before 2025

The year on a patent document and the year behind its filing history answer different questions. Grouping the same cohort by earliest known priority year, or application year when priority is absent, shifts the center of gravity backward.

The largest single underlying year is 2023, with 846 records. The combined 2022 to 2024 block accounts for 2,232. Another 1,019 records map to 2021 or earlier. Only 438 map to 2025 itself.

This does not mean 2025 lacked AM invention. It means the clocks that put a document into 2025 mostly started earlier. Under the 18-month rule a 2025 priority filing publishes in mid-2026 at the earliest, so a 2025 basis reaches this cohort only through the early-published Chinese rows and the application-date fallback. National-phase entry re-dates a filing at 30 months, and grants arrive later still: on the live corpus the median lag is 563 days for A-kind documents and 1,388 for B-kind grants.

Three dates answer three questions

Patent analytics often changes meaning when one date field changes. Priority date is the earliest available filing basis in this dataset. Application date identifies the filing represented at a particular office. Publication date identifies when the document record in the cohort is dated. A later grant can create another publication event without creating another underlying invention.

This distinction also explains why two valid AMPulse figures do not reconcile. The live Patent Statistics surface groups records by application date and contains 1,212 active AM-relevant records for 2025 in the frozen snapshot. This report groups by publication date and contains 3,689. The first reads filing-year activity in the tracked corpus. The second reads document flow dated during 2025.

What the cohort includes

The cohort mixes document stages. A-kind documents are the largest group, but they are not the whole set, and not all of them are first publications.

A-kind documents account for 2,049 records, 55.5%. Kind code A marks an application-stage publication. That includes the office's first publication of an application, and it also includes republications of one already on file: US A2 and A9 documents, European A3 and A4 search reports, and national-phase A publications of an application already published as a WO. The corpus records the kind code, not whether a document is a first disclosure, so the group is separated only as far as the 77 European search reports it contains, 61 A4 and 16 A3. Grant or post-grant publications account for 983, 26.6%. PCT publications contribute 592 records whose kind code is not typed in the current corpus. The final 65 are unclassified or other, including two Japanese utility-model records retained by the source dataset.

That composition sets the counting unit for every figure in this report: the 3,689 records are documents dated in 2025, not 3,689 inventions first disclosed in 2025. The family field cannot repair that. Family identifiers are missing for 655 records; the other 3,034 carry one, and those resolve to 3,034 distinct identifiers. A slice that holds WO publications, their European and Chinese national-phase publications, and 77 European search reports cannot contain zero shared families, so a field that never repeats across it is a per-document identifier, not a family identifier; corpus-wide, active rows repeat only 78 of them. family_id cannot deduplicate this cohort, and family counts are unavailable.

Four documents, four filing clocks

The examples run from 1.72 years for the MELD-linked US document to 9.16 years for the Nikon SLM Solutions-linked Chinese document. They show why the publication year alone cannot identify the vintage of the underlying technical work.

How to use the signal

Recommended next steps

1. Add document-stage filters so application publications and grants can be compared without mixing their clocks.
2. Backfill the 7 records that carry a priority claim in the source with no stored date, and fetch source observations for the 75 that have none.
3. Build normalized corporate groups and patent families before publishing a company leaderboard.

Further questions

• Does the priority-year curve show a genuine change in AM filing activity after publication-delay censoring is accounted for?
• Which technical classes move from application publication to grant most consistently, and over what interval?
• How much of the long tail comes from continuation, national-phase, divisional, or post-grant document paths?
• Do company rankings remain stable after family and corporate-group normalization?

Methodology

Lee, S. (2026). The Patent Clock Behind 2025. AMPulse Data Reports. https://www.ampulse.online/reports/2025-am-patent-signal