Methodology · Public
Every number on this site, defended.
Every metric we publish comes from public USPTO records. Below is exactly how each is computed, what sample size is required before we display it, and where the data has known gaps. Every profile these numbers appear on is listed in the examiner directory.
The headline read
The four numbers fixed above every profile: refusal, escalation, withdrawal, difficulty. Each leads with its plain-English answer; the formula is one click away.
Initial refusal rate · over examined apps
Coverage gate · ≥10 fetched-OA apps AND ≥0.33 fetch coverage
Of the applications this examiner examined, the share that drew a substantive refusal: §2(d), descriptiveness, surname, geographic, or another substantive ground. Procedural-only actions don't count, and a docket-noise filter drops backlog re-examinations and abandoned pro se floods. The base counts approvals the examiner granted without ever writing a refusal: first-action allowances (the only record is the allowance milestone) and no-paper approvals (straight to publication, or resolved by an examiner's amendment with no office action at all). Those are the examiner's cleanest outcomes; a base that skipped them would over-state refusal posture most for exactly the most approval-prone examiners (redefined 2026-07-24: the prior OA-only base read the corps at ~47%; the examined-apps base reads ~30%). Profile headlines read this same formula over the last two years (labeled "Last 2 yrs," gated at n ≥ 30), with the full-record rate alongside. The by-year chart and its one-line read draw the last five complete years: the in-progress year is excluded (its partial-year denominator can invert a direction), and earlier years stay in the career figure rather than the chart, because per-year coverage thins with age and a long window reads that thinning as a trend.
Show the formulaHide the formula
Initial refusal rate
= apps that drew at least one substantive office action
÷ examined apps
(real-OA apps + first-action allowances
+ attributed no-paper approvals)Edge cases & limitsHide edge cases
Escalation rate (Final | merits-resolved)
Sample-size gate · n ≥ 10 merits-resolved refusals AND ≥0.33 fetch coverage
Of the refusals where this examiner revealed their posture, pushed to a Final or approved the mark, the share pushed to Final. Read the other way, how often they back down. Conditioning on a refusal keeps applicant quality out of the denominator; conditioning on a merits decision keeps applicant persistence out too. A refusal abandoned before the examiner ever decided says nothing about the examiner, so it sits out of both sides (the same rule the §2(d) escalation rate has always used: abandonment is an applicant decision, never an examiner statistic). The approved side is ordering-checked: an approval stamped before the refusal ever issued reflects an earlier examination pass, so it counts as a back-down only when the mark went on to register, and otherwise leaves the denominator. Redefined 2026-07-23: previously the denominator was all refused apps, which diluted the rate with uncontested abandonments; the merits-resolved read runs roughly twice the diluted one.
Show the formulaHide the formula
Escalation rate
= refused apps pushed to at least one Final Office Action
÷ refused apps where the examiner showed their hand
(pushed to Final OR approved the mark after the refusal)Edge cases & limitsHide edge cases
Withdrawn when contested
Sample-size gate · n ≥ 10 contested resolved refused apps · fetch-coverage ≥ 0.33 · maturity ≥ 12 mo and ≥ 30 resolved apps
Of the refusals an applicant contested, at least one response on file: the share that still reached approval. This is the base rate for the decision an attorney is making: a client drew a refusal from this examiner: the odds it comes off if they respond.
Show the formulaHide the formula
Withdrawn when contested
= contested refused apps that still reached Approved-for-Publication
÷ refused apps with at least one response filed
that reached terminal disposition (approved or abandoned)Edge cases & limitsHide edge cases
Difficulty decile
Gate · n ≥ 30 apps + a published refusal rate · §2(d) hold & post-Final axes n ≥ 25 · appeal axis ≥ 15 resolved
A 1 to 10 decile of how hard this examiner is to overcome, ranked across the active examining corps (active = an office action in the last 365 days). It scores winnability: how firmly they hold a §2(d) at Final, how rarely they withdraw after Final, whether refusals stick through a Board appeal, not raw refusal frequency, which mostly reflects the docket they were handed.
Show the formulaHide the formula
composite = 0.35 · z(holds at Final · §2(d) + general)
+ 0.25 · z(won't withdraw post-Final §2(d))
+ 0.20 · z(refusal rate, basis-adjusted)
+ 0.12 · z(sticks through TTAB appeal)
+ 0.08 · z(office-action rounds)
(weights renormalize over the pillars an examiner has;
a thin appeal or post-Final record drops out rather than
pulling the score toward the median; basis-adjusted
refusal is always present. The post-Final withdrawal and
appeal axes are empirical-Bayes shrunk toward the corps
mean so a tiny sample can't swing the score.)
appeal-resistance
= appeals where the refusal was upheld
÷ resolved appeals
(upheld + refusal withdrawn after appeal + reversed)
decile = this examiner's rank, in tenths, across the
active corps, sorted by composite ascendingEdge cases & limitsHide edge cases
The §2(d) playbook
Read in prosecution order: initial, Final, appeal. Then the per-case playbook report.
Initial
§2(d) escalation / back-down rate
Coverage gate (≥0.33 fetched) + n ≥ 10 merits-resolved §2(d)
When this examiner opens with a §2(d) and the matter reaches a merits decision, the share they push to Final; read the other way, how often they back down first. The opening-move number, before a single response is filed.
Show the formulaHide the formula
§2(d) escalation rate
= §2(d) applications the examiner pushed to a Final
÷ §2(d) applications that reached a merits decision
(made Final OR approved without a §2(d) Final)
Back-down rate = 100% − escalation rate (the same statistic, read the other way).Edge cases & limitsHide edge cases
Initial
§2(d) co-traveling grounds rate
Coverage gate (≥0.33 fetched) + n ≥ 10 §2(d) apps
Of the applications this examiner hits with a §2(d), the share where it arrives alongside another substantive ground on the same office action. A §2(d) alone, or a multi-front response.
Show the formulaHide the formula
§2(d) co-travel rate
= §2(d) apps whose §2(d) shares an office action with another substantive ground
÷ apps that drew a §2(d) from this examinerEdge cases & limitsHide edge cases
Final · the anchor
§2(d) post-Final withdrawal rate (· when answered)
Computed from n ≥ 10 · shown as a rate at n ≥ 25 answered, resolved Finals, smoothed toward the corps mean with the same K = 40 pseudo-count the difficulty score uses
Among this examiner's Final §2(d) refusals that the applicant answered, the share where the §2(d) came off and the application reached Approved-for-Publication. The single most actionable number at the answer-or-abandon decision. It answers the question you would face: whether going back at this Final clears it. Redefined 2026-07-23: roughly half of resolved §2(d) Finals are never answered at all, and a walk-away is an applicant decision, never an examiner statistic; the old unconditioned rate blended those in and understated everyone's odds roughly two-fold (corps mean ~34% → ~70% when answered). Below the display gate the profile names the sample and withholds the percentage: a thin record's confidence interval is wider than the spread it would be read against. The tooltip beside the shown figure carries the unsmoothed rate.
Show the formulaHide the formula
§2(d) post-Final withdrawal rate (· when answered)
= answered, resolved §2(d) Finals approved for publication
AFTER the Final (the §2(d) came off)
÷ answered, resolved §2(d) Finals from this examiner
Answered = the applicant came back after the Final: a request
for reconsideration, any later response, a TTAB
appeal, or a post-Final examiner's amendment
Resolved = approved or abandoned (pending apps sit out)Edge cases & limitsHide edge cases
Appeal · the lever
The appeal lever · refusal removed after a notice of appeal
Computed from n ≥ 10 · shown as a rate at n ≥ 25 resolved, smoothed toward the corps mean (K = 40, §4's display standard) · per-path median months at n ≥ 5 · counts shown at any n
After a notice of appeal, the refusal can come off without the case ever needing a Board decision. Corps-wide the refusal came off in about two of every three resolved appeals, and the dominant path is the examiner withdrawing the refusal within a few months, not a Board ruling roughly a year out. The vs-corps verdict renders only at the confidence the sample supports, on §4's graded standard.
Show the formulaHide the formula
Lever rate
= (refusal withdrawn after appeal + Board/court reversed)
÷ resolved appeals (every §15 outcome except pending)
Withdrawal path = approval reached after the notice of appeal was
filed, with no Board ruling (median months at n ≥ 5)
Reversal = Board or court reversed on the meritsEdge cases & limitsHide edge cases
Appeal
TTAB appeal outcomes
Counts shown at any n · powers the appeal lever (§20) and the receipts table
Most appeals never reach a merits decision; they close because the refusal came off and the application moved forward, or because the applicant walked away (gave up). "Reversed" is the narrow merits win; "examiner won" is merits affirmance only. No-brief dismissals (the appeal was filed but a brief never followed) used to count as examiner wins; since 2026-07-23 they resolve like every other non-merits closure, by what happened to the application (an approval after the notice of appeal reads as the refusal withdrawn, an abandonment as the applicant giving up). "Refusal withdrawn after appeal" records a sequence, not a concession. Across roughly 24,000 appeals that closed this way since 2015, we can see the applicant filing a response or amendment during the appeal window in at least 71% of them; counting every applicant-side docket event, at least 92% show the applicant moving (both are floors: docket-event coverage is incomplete, and a filing we cannot see counts against the "nobody moved" column by default). The bucket includes approvals granted on an applicant's amendment, remands for new evidence, consent agreements filed on appeal, and cases where the cited registration itself died and the refusal went moot (TBMP 1205.01(a), 1209.04, 1212). The appeal record does not say which, so neither do we. This taxonomy is the input behind the Appeal stage: the lever rate (§20), the difficulty score's appeal-resistance axis, and the per-case receipts table all read from it.
Show the formulaHide the formula
Of ex parte appeals filed against this examiner's Final refusals:
reversed Board or court reversed on the merits (examiner lost)
examiner won affirmed on the merits
applicant gave up appeal closed without a merits ruling,
application abandoned
refusal withdrawn appeal closed without a merits ruling,
after appeal application registered or approved
pending appeal still active
Denominator = all appeals filed against this examiner.Edge cases & limitsHide edge cases
Appeal · not surfaced
§2(d)-involved TTAB appeals
Rate gated at n ≥ 10 §2(d) Finals · counts shown at any n
Appeals on applications carrying a §2(d) Final from this examiner, and how often they end in reversal. §15 spans every refusal ground; this isolates the §2(d) record. Computed and documented for transparency, but no longer shown on profiles (2026-07-12), for two reasons: the per-examiner §2(d)-isolated appeal and reversal counts run too thin to read as more than noise, and the rate itself tracks career length and docket appeal exposure (a long-tenured examiner accumulates appeals a newer one cannot have) rather than anything the examiner chooses. The Appeal stage leads with the all-grounds lever (§20) instead.
Show the formulaHide the formula
§2(d) appeal rate
= appeals on applications that carried a §2(d) Final from this examiner
÷ this examiner's §2(d) Finals (ALL, deliberately unconditioned:
"how many Finals get appealed" is a population question, and an
appeal on a still-pending app would leak past a resolved-only base)
§2(d)-involved appeals and reversals are the counts behind that rate.Edge cases & limitsHide edge cases
The playbook
Cited registrations leaderboard
Suppression · ≥5 distinct regs to render
The ten registrations this examiner cites most in §2(d) refusals. Outcomes count per application; each lands in exactly one fate bucket (cleared, contested and lost, walked away, still open), with contested and held as overlays. Whether a specific barrier can be overcome, and how, is the §2(d) detail report (§14). Suppressed below five distinct regs.
Show the formulaHide the formula
The most-cited prior registrations in this examiner's §2(d) office actions (initial and Final), ranked by distinct citing office actions: 10 mark groups, with same-mark sibling registrations collapsed behind each group's disclosure. Each row splits the DISTINCT APPLICATIONS that faced the citation by where they ended up, one bucket per application: cleared reached approval contested, lost responded, then went abandoned walked away went abandoned without ever responding still open no terminal outcome yet Overlaid on those: contested (any response was filed) and held (a response was met with a maintained or Final refusal whose follow-on action still carried the §2(d); a follow-on that classified without the §2(d) means it came off, and does not count as held. A follow-on whose grounds we have not fetched stays counted as held: a coverage gap is not evidence the §2(d) came off).
Edge cases & limitsHide edge cases
The playbook
§2(d) detail report · overcome-rate cascade
Per-level gate · n ≥ 25 contested, resolved (§4's display standard: field rates smoothed toward the field's corps rate, K = 40; the vs-corps badge colors only at the confidence the sample supports) · no verdict labels · the rate, its n, and the corps clause are the read
Whether this cited registration can be overcome before this examiner, and how. The share of applications that contested a §2(d) citing that registration and still reached approval. Contested-conditioned (2026-07-23): the same principle as the withdrawal and post-Final rates: an application abandoned without ever responding is an applicant decision that says nothing about whether the barrier gives way, and a held refusal on a still-live case isn't an outcome yet. One reg rarely has enough history against one examiner, so the read is usually the examiner's rate across the application's class field, shown against the corps baseline; a reg-specific rate overlays when it clears the gate on its own.
Show the formulaHide the formula
For an (examiner, cited reg, application field) query, each cited
registration is ranked by overcome rate (· when contested):
overcome rate = applications that CONTESTED a §2(d) citing that
reg (≥1 response filed) and reached approval
÷ contested applications at terminal disposition
(approved or abandoned)
Out of both sides: walk-away abandons (never responded) and
apps where the refusal was held but the case is still live.
Resolution cascade, per reg:
reg-level when that reg clears n ≥ 25 contested corps-wide
(the rare sharp read: the corps rate leads, the examiner's
own rate rides alongside when it too clears n ≥ 10)
field-level the ~99% workhorse: the examiner's overcome rate in the
application's class field, shown vs the corps in that field
insufficient neither clears n ≥ 25 → no rate; the field fallback is namedEdge cases & limitsHide edge cases
The full record
Terminal outcomes, disposition, pace, and the comparison layer.
Outcome
Clean approval rate
Maturity gate · ≥12 mo on the bench AND ≥30 resolved apps
Of the applications that reached a terminal disposition, the share approved for publication without ever drawing a Final. The win signal that doesn't bundle in approvals dragged through a Final round.
Show the formulaHide the formula
Clean approval rate
= apps approved for publication without ever receiving a Final
÷ resolved applications (terminal status) on this examiner's docketEdge cases & limitsHide edge cases
Disposition · cooperative
Examiner's amendment rate
Published at ≥30 eligible intervention apps in the window
When a fixable problem blocks publication, an examiner can write a formal office action, or call and fix it with an examiner's amendment. This is the share of eligible intervened applications resolved the second way. An amendment on its own rarely clears a substantive refusal (a §2(d) is not phoned away), so applications where this examiner raised one sit out of both sides; only 9% of amended applications carry one in the window. A high share is the sign of cooperation: fixable problems get a phone call, not a response cycle.
Show the formulaHide the formula
Examiner's amendment rate (· when an amendment could work)
= eligible apps resolved with an examiner's amendment (last 2 yrs)
÷ eligible apps they intervened on
(formal office action ∪ examiner's amendment)
Eligible = no substantive refusal ground raised on the app
(and every office action's text on file)Edge cases & limitsHide edge cases
Docket state · not scored
Suspension tracking
No per-examiner suspension statistic is published
A suspension parks an application, usually behind an earlier-filed conflicting application, and pauses every examination clock. We track suspension letters, the six-month checks, and the release back to the examiner, and surface the state on your docket. How often a docket suspends reflects the applications on it, not the examiner deciding them, so no per-examiner suspension rate is published.
Show the formulaHide the formula
Suspension state (per application, not per examiner) letter → [6-month checks] → released to examiner captured from the USPTO event stream; admin-review suspensions tracked separately from examiner practice
Edge cases & limitsHide edge cases
Outcome · context
Publication rate
Maturity gate · ≥12 mo on the bench AND ≥30 resolved apps
Share of resolved applications approved for publication, across all bases. Shown as a context cell in the profile's fine print; clean approval(§7) stays the headline win signal, because publication rate alone can't tell a smooth ride from a three-round prosecution that got there anyway. Read against §7, this rate's own contribution is exactly that difference: the share of approvals earned through a Final rather than before one.
Show the formulaHide the formula
Publication rate
= applications approved for publication
÷ resolved applications (terminal status) on the docketPace
Response turnaround, time-to-publication & the first-OA clock
Sample-size gate · n ≥ 10
Median days from attorney response to the examiner's next action; median days from the first Office Action to publication approval; and median days from filing to this examiner's first Office Action: the day-one client question (“the examiner has the file; when will we hear?”). The clock headlines the last two years: USPTO first-action pendency peaked in 2021 through 2023, so a career median overstates today's queue for examiners whose docket spans that era; the full-record median rides alongside, labeled. Read all three as throughput, not examiner speed: time-to-publication folds in the attorney's own response time, and filing→first-OA is dominated by USPTO queue latency. That's why the clock carries no vs-corps verdict color anywhere it renders.
Show the formulaHide the formula
Turnaround = median days from attorney response
to the examiner's next action
Time to publication = median days from the application's
first Office Action to approval
Filing → first OA = median days from the application's
filing date to this examiner's
first Office Action
(headline = last 2 years; the
full-record median rides alongside)
(Time-to-pub and filing→first-OA are restricted to apps where
this examiner issued the round-1 OA; mid-stream pickups
don't count. Filing→first-OA counts real examinations only:
non-final/final actions, never NOA milestones or the inert
backlog clear-out notices excluded corps-wide. The recent
window is dated by the FIRST ACTION, not the filing; a
filing-date window would drop still-waiting applications
and understate the median.)Edge cases & limitsHide edge cases
Time-to-publication and filing→first-OA are restricted to applications this examiner owned from round 1 (i.e., they issued the application's first OA). A mid-stream pickup (where another examiner did most of prosecution and this examiner issued only the late-stage OA) would otherwise produce misleadingly small medians. Time-to-publication therefore describes the OA-bearing side of the docket only: nearly half of approved applications never drew an office action at all (their earliest docket row is the allowance itself), and those fastest approvals are outside this clock by construction. The first-OA clock additionally skips the mass backlog clear-out notices (§ 1's docket-noise filter): those fire years after filing and would otherwise dominate the median for the two affected examiners. The recent window publishes at ≥10 first actions issued in the last two years; an examiner working only later-round cases now falls back to the labeled full-record figure.
Refusal · by class
Refusal rate by class · the class × ground heatmap
Rate cells ≥25 examined and ≥10 fetched · ground shares ≥10 classified refusals · corps baseline needs ≥20 examiners
The same substantive-refusal rate as §1, computed separately for each Nice class the examiner has examined, beside the share of that class's refusals citing each ground. Shading always encodes the gap to the corps in that same class: classes differ enormously in how often they draw a refusal at all, so shading the absolute number would mostly redraw the docket's composition instead of the examiner's behavior. Every cell prints its own number, so the table reads without relying on color.
The two column families are shaded differently on purpose. The refusal rate is a severity reading: above the corps means more refusals, plainly worse for the applicant, so it runs green below the corps and red above. A ground share is a composition reading: leaning on §2(e) more than the typical examiner is a different fingerprint, not a worse one. Those columns therefore use a single neutral hue whose depth marks distance from typical, with a caret for direction: the shading says “unusual here,” and declines to say “bad here.”
Show the formulaHide the formula
Per-class refusal rate
= (apps with a substantive refusal ÷ apps with fetched text)
× (apps with a real office action ÷ examined apps);
all four counts scoped to apps listing that class
Examined apps (per class)
= real-office-action apps + first-action allowances
+ attributed no-paper approvals, in that class
Ground share (per class)
= apps citing that ground ÷ ground-classified refused
apps in the class (independent, overlapping)
Ground columns: §2(d) · §2(e) · Other. The §2(e) column
is one per-app bucket: descriptiveness §2(e)(1),
geographic §2(e)(2)–(3) and surname §2(e)(4) together;
an app citing several of them counts once.
Cell shading (both columns)
= examiner − the CORPS, in the SAME class
· refusal rate → green below / red above (severity)
· ground shares → one neutral hue, depth = distance
from typical, caret = directionEdge cases & limitsHide edge cases
The comparison layer
Corps baselines · how every “vs corps” figure is built
Documented honesty entry · no gate of its own
Every “vs corps” clause on a profile compares against an unweighted mean across active examiners with a publishable value for that stat: one examiner, one vote, regardless of docket size. This deliberately answers “how does this examiner compare to the typical examiner?” rather than “what happens to the typical application?”: a volume-weighted average would let a handful of mega-docket examiners define normal.
Show the formulaHide the formula
Corps baseline (rates)
= unweighted mean of per-examiner PUBLISHED values;
each stat's pool is the examiners passing THAT
stat's own gate, so pools differ stat to stat
Corps timing figures
= mean of per-examiner MEDIANS (each examiner's median
is computed first; the corps figure averages those)
Appeal-lever corps figure
= POOLED across all appeals corps-wide (not a mean of
examiners); sparse per-examiner appeal counts would
make a mean-of-examiners noise-dominatedEdge cases & limitsHide edge cases
Known gaps
What we cannot yet see, stated plainly. Read these three before you trust any number here.
§66(a) Madrid Protocol coverage gap
About 25% of the §66(a) Madrid Protocol office actions we fetch carry a classification, against about 39% of domestic actions: §66(a) actions reach our classified corpus less often. Where a §66(a) action is classified, our v10 cite extractor pulls citations at least as well as it does on domestic filings. When an examiner's docket is more than 10% §66(a) filings (strict greater-than), the profile's cited-registration section and the per-registration case-detail pages carry a coverage notice; treat the cited-registration counts on those examiners as a lower bound.
Pre-2007 cutoff
USPTO didn't standardise disposition codes until early 2007. We exclude all OAs before that date; including them would inflate "no-disposition" counts and distort every rate. All published metrics on this site are computed against post-2007 data only.
Insufficient data treatment
Below per-metric thresholds we name the sample and withhold the rate rather than print a percentage. A gated §2(d) band reads "Building: n answered §2(d) Finals so far," with the gate stated beneath it. Two floors are locked: 30 applications on record before any overall stat publishes, and 10 answered, resolved Finals before a §2(d) differentiator is computed at all. A computed rate is not automatically a shown one: the post-Final §2(d) withdrawal figure (§4) displays from 25, because between 10 and 25 its confidence interval is wider than the spread it would be read against. We will not show a percentage based on n=8.
Built on the public USPTO record. Coverage begins January 2007, when the USPTO standardized outcome recording.
Data current as of · the most recent stats recompute across the corps
2,223,696 substantive office actions · 925 active examiners · coverage from January 2007
Questions about a specific metric? support@crystalmark.app. We answer methodology questions in plain English.