Research
Research methodology
Every Copymark research study is a dated snapshot of the USPTO trademark corpus we maintain for the public directory. This page documents the counting rules so journalists, students, and counsel can cite the numbers with confidence.
Current published snapshot: 2026-11-08
Corpus
The source of truth is the USPTO bulk trademark case-file database ingested into our SQL Server warehouse (application serials in dbo.case_file, owners in dbo.owner, Nice classes in dbo.clasification, prosecution history in dbo.event). Public research pages never query the warehouse on each page view. A snapshot job aggregates the metrics offline and publishes JSON and CSV artefacts to object storage.
Snapshot date
Each study and its CSV share a snapshotDate (UTC calendar day). Weekly snapshots cover a completed Monday–Sunday week, monthly snapshots cover a completed calendar month, and the annual snapshot covers a completed calendar year. Those period snapshots count applications filed inside the window. A full-corpus snapshot, when published, still treats the current calendar year as year-to-date through snapshotDate.
One application, one serial
Application-level metrics count each serial_no once. Multi-class applications still count as one application for state, owner, and outcome studies.
Primary owner
When an application has multiple owner rows, we select the primary owner using the same rule as the public trademark catalogue: highest own_type_cd, then lowest own_seq. State and applicant studies use that primary owner only.
Owner-name normalization
Top-applicant rankings group names after uppercasing, stripping common punctuation, and collapsing whitespace. That merge rule can combine distinct legal entities that file under similar strings, and can split a single company that uses variant legal names. Display names shown on the page are representative strings from the group; the CSV also includes the normalized key.
Nice classes
Class studies use international class codes 001–045 from dbo.clasification. A multi-class application is counted once in each class it claims.
Office Action mailing events
Office Action rates count distinct serials with at least one matching prosecution-history event: non-final actions GNRT, CNRT, or GNRN; final refusals GNFR or CNFR. These are mailing/notification events. They are not statutory refusal grounds.
What this release does not measure
USPTO bulk records do not store the legal reason an examiner refused a mark. Statutory grounds such as §2(d) likelihood of confusion live inside Office Action letters and are not classified in this release. Outcome studies report terminal status codes (for example failure to respond) as disposition proxies, not as refusal grounds.
License and attribution
Aggregate tables and CSV downloads on these research pages are offered under Creative Commons Attribution 4.0 International (CC BY 4.0). Please credit Copymark and link to the study URL and snapshot date. Underlying USPTO trademark records remain public U.S. government data.
Snapshot definitions
- Highest own_type_cd, then lowest own_seq (same rule as the public catalogue).
- Uppercase; strip common punctuation; collapse whitespace. Merges some distinct entities and splits some legal-name variants.
- Application-level metrics count each serial once; Nice-class metrics count a multi-class application once per claimed class.
- Non-final: GNRT, CNRT, GNRN. Final refusal: GNFR, CNFR. Distinct serials, not every mailing.
- This snapshot counts applications filed from 2026-11-02 through 2026-11-08 (Week of 2–8 November 2026).
- Statutory refusal grounds (including §2(d) likelihood of confusion) are not native USPTO bulk fields and are not measured in this release.