NorthLaw.ai
U.S. Trademark Examiner Analytics
U.S. Trademark Examining Attorney Analytics · filings 1985–2022
Answers For each USPTO trademark examining attorney, the share of decided applications that registered — stabilized toward the national rate and shown with a 95% interval — and, for filings 2005–2022, their prosecution profile: how often they issue a non-final action or a final refusal, how often an application registers after an action, and the median days from filing to the first action, each beside the national figure.From USPTO Trademark Case Files bulk data, 2023 release (data to March 2024): 10.0 million applications filed 1985–2022 with an examining attorney recorded, 9.2 million decided; and the event history of the same release — 209 million prosecution events — for the 7.0 million applications filed 2005–2022. Rebuilt from the raw files by builders that reproduce every row exactly (graph re-proved 9 September 2026; the prosecution builder reproduces the graph’s own totals before it writes); every chosen number is in the panel at the foot of the page.Not Not a prediction for your application — the examining attorney is not known at filing — and not a clearance search. Outcomes are the USPTO's current status codes, so a registration later cancelled, or a case whose §8 declaration was accepted (status 701), counts as pending under this rule; examiners with fewer than 20 decisions are flagged. The event history records that an office action issued and when, not the ground of refusal.Use On your own. The inputs are facts you already have; the result is a screening estimate that states its own limits.

Outcomes

Decided Applications by Filing Year

Top Examiners by Volume

EXAMINERDECIDEDREGISTEREDABANDONEDREG. RATEACTIVE
Registration rates are empirical-Bayes stabilized toward the national average (60.3%) and shown with 95% intervals; examiners with fewer than 20 decided applications are flagged. Denominator is decided applications only (registered or abandoned) — pending applications are excluded. Descriptive history, not a prediction.
EXAMINERDECIDEDREGISTEREDABANDONEDREG. RATEACTIVEACTION RATEFINAL REFUSALREG. AFTER ACTIONDAYS TO 1ST ACTION
EXAMINERDECIDEDREG. RATEACTIVE

Application-level browsing is aggregated to examiner rows in this build; per-application detail is available in the underlying data.

Source: USPTO Trademark Case Files, 10.0M examined-era applications (filings 1985–2022), 9.2M decided. Outcomes are classified by the USPTO's current status code (registered: 700, 702, 710, 730, 748, 800, 810, 811; abandoned: 602, 604, 606, 608, 612; every other code, including a cancelled registration or a §8-accepted case, is pending) — dates were rejected as the classifier because they misclassify about 570,000 revived and cancelled cases. The full rule is in the assumptions panel above. One examining attorney per application — no name disambiguation required. Registration rate = registered ÷ (registered + abandoned); pending excluded. Rates stabilized (empirical-Bayes) with 95% intervals; examiners under 20 decided flagged. Descriptive administrative history — not legal advice or a prediction. Every number names its denominator.
Assumptions behind these numbers (9 chosen values)

Every number below was chosen rather than measured: it is a modelling judgment, a labelling threshold, or a statistical convention. Each one was tested across the range shown in the September 2026 audit, and the effect on this tool's outputs is recorded in the methodology. Measured inputs — FJC federal civil case outcomes, SSA life tables, BLS benefit rates, BEA regional price parities, work-life expectancy tables — are not listed here; they carry their own provenance.

assumptionvaluebasisrange tested
tmkg.shrink_k
prior strength, in pseudo-decisions, pulling each examining attorney's registration rate toward the national rate: rate = (registered + K·p0) / (decided + K)		25.2557stored constant in the live database; the recovered builder reproduces every row from it. Closest derivation: p0(1−p0) / variance of raw rates over examiners with ≥ 20 decisions = 25.0 (near, not exact)
F-71 (9 Sep 2026): registered from the meta table. An examiner with one decision moves up to 0.58 toward the national rate; one with 1,000 moves about 0.01.		[10, 50]
tmkg.limited_threshold
decided applications below which an examiner is flagged 'limited'		20judgment
27 of 1,606 examiners are flagged.		[10, 50]
tmkg.wilson_z
z for the 95% Wilson interval on each examiner's raw rate		1.96convention1.645 (90%) to 2.576 (99%)
tmkg.status_registered
USPTO current-status codes counted as registered		[700, 702, 710, 730, 748, 800, 810, 811]measured: recovered by exact-reconciliation gate against the live graph (29 Aug 2026), re-proved 9 Sep 2026
Status is the case's CURRENT status: a registration later cancelled (900) or a case with a §8 acceptance (701) is pending under this rule, matching the original builder. Dates were rejected as the classifier because they misclassify ~570,000 revived and cancelled cases.		n/a
tmkg.status_abandoned
USPTO current-status codes counted as abandoned; every other code is pending		[602, 604, 606, 608, 612]measured: recovered by exact-reconciliation gate (29 Aug 2026)n/a
tmkg.filing_years
filing-year window of the universe (applications with an examining attorney recorded)		[1985, 2022]judgment
Extending the window changes every published figure and the printed postcards; a product decision, not a refresh.		n/a
tmkg.prosecution_years
filing-year window for the prosecution profile (office actions, final refusals, registration after an action, days to first action) built from the event history		[2005, 2022]judgment
Electronic event codes are complete from the mid-2000s. The 2021–2022 cohorts are partly right-censored — median days to first action 259–290 against about 105 for 2008–2019 — and the by-year table shows that rather than trimming it.		n/a
tmkg.median_min_n
observations below which a median (days from filing to first action, to approval) is not shown for an examiner or a year		20judgment[10, 50]
tmkg.prosecution_limited
applications filed 2005–2022 below which an examining attorney's prosecution rates are flagged 'limited'		100judgment
Distinct from tmkg.limited_threshold (20 decided applications), which guards a registration rate that is shrunk toward the national rate; the prosecution rates are unshrunk shares, so the floor is higher.		[50, 200]

Data behind these numbers (3 sources, 3 fully documented)

Measured inputs and where they come from. documented: source, vintage and derivation recorded and verified; partial: source known, vintage or script missing; undocumented: the file is on the server but its origin could not be recovered.

sourcewhat it provides · file · vintagestatus
Trademark case files 2005-2022
USPTO Trademark Case Files Dataset		modelling corpus (6,748,762 applications; registration rate 0.613), base rate for zones
model_ready_phase1.parquet; tm_base_rate.json; models/model_v_{a,b,c}.pkl (2026-03-10); built by train_all_phases.py, retrain_vc.py — pipeline reproduced EXACTLY (F-27, dAUC 0.0000); vintage: filed 2005-2019, validated on 2020-2022		documented
USPTO Trademark Case Files (bulk)
USPTO Office of the Chief Economist		one row per trademark application: filing date, current status code, examining attorney, and 76 other fields
uspto_tm/raw/case_file.csv (3.25 GB, 79 columns, file dated 2024-03-19); vintage: USPTO Trademark Case Files Dataset, 2023 release (data to March 2024): the server file is byte-identical to case_file.csv in Zenodo record 17268251 (zip md5 fab66a545b2d8a0da15cff265b599b9e, verified 23 Oct 2026)
Universe for the examiner KG: rows with an examining attorney recorded, filings 1985–2022 (10,039,277). The builder uspto_tm/build_uspto_examiner_kg_v2.py reproduces the live graph exactly from this file (gate re-run 9 Sep 2026). Status codes are current status. Vintage settled 23 Oct 2026 by the Zenodo match (F-103).		documented
USPTO Trademark Case Files — event history
USPTO Office of the Chief Economist		one row per prosecution event: serial number, event code (741 codes) and date — assignment to an examining attorney, non-final actions, final refusals, suspensions, examiner's amendments, approval for publication, responses, abandonment
uspto_tm/raw/event.csv (6.39 GB, 209,031,161 rows) and event_description.csv, extracted from uspto_tm/zenodo/trademark_case_files_2023.zip; vintage: the same 2023 release (data to March 2024); Zenodo record 17268251, CC-BY-4.0, md5 verified 23 Oct 2026
Builder uspto_tm/build_examiner_prosecution_v2.py: event codes grouped by the USPTO's own descriptions (the code lists are stored in the output's meta table), first date per application and category, joined to case_file.csv under the graph's status rule for filings 2005–2022 (7,041,829 applications). Control: the builder reproduces the live graph's totals — 10,039,277 applications, 9,155,726 decided, 1,606 examiners, national rate 0.6029456 — exactly before it writes. Output uspto_examiner_prosecution_serve.db. The dataset records that an action issued and when, not the ground of refusal.		documented

Source of truth: assumptions.py in the NorthLaw repository (Audit C, F-56–F-59). If one of these does not fit your matter, the expert-review engagement re-runs the analysis with your figures.