A handful of examiners' refusal rates cratered in a single year — with no change in their behavior. Here's what was really happening.

§ 1 · Where the flood came from

Two forces, one wave of office actions that aren't refusals.

A surge in foreign pro-se filings and a 2019 rule those filings mandatorily trip arrived together. Each adds office actions to an examiner's docket. Neither adds a single substantive refusal.

One word of precision: the overwhelming majority of these filings are lawful — low-quality, often with no genuine intent to use, but not fraud. A smaller subset crossed into a scheme the USPTO formally sanctioned. In December 2021 the agency terminated more than 15,000 applications and submissions tied to Chinese filer Yusha Zhang and Shenzhen Huanyee Intellectual Property Co. for fake specimens, false domicile addresses, and unauthorized practice of U.S. law. That is the only part of this story to which the words “fraud” or “fake” attach — and they attach to that filer, never to any examiner.

§ 2 · Why a correct count produces a wrong rate

The classification is right. The denominator is the trap.

1. 01

   A requirement is not a refusal.

   CrystalMark's classifier scores substantive grounds only — §2(d) likelihood-of-confusion, descriptiveness, surname, geographic, and the rest. A counsel, specimen, or identification requirement is correctly recorded as zero substantive refusals, because that is exactly what it is. The classification is right. The problem is what happens next.

2. 02

   Counting them dilutes the denominator.

   A naïve per-office-action refusal rate divides refusals by all office actions. Pour thousands of requirement-only actions into that denominator and the rate craters — not because the examiner softened, but because the bottom of the fraction grew. Same behavior, smaller number. The harm is denominator dilution, not misclassification.

3. 03

   We exclude the non-substantive actions.

   CrystalMark pairs substantive-only classification with a denominator exclusion: requirement-only office actions — the counsel mandate, bare specimen and ID requirements — drop out of the rate entirely. What remains is the examiner's genuine prosecution posture, measured against the cases where they actually judged a mark. Filing-mill volume can't move the number.

§ 3 · We checked it the hard way

Four ways this could be our bug. Each one, ruled out.

Before blaming the world, we blamed ourselves. We pulled 150 of the diluting office actions back from USPTO TSDR, re-extracted them, and hand-checked the classifier against every substantive ground. Here is what we tested and discarded.

1. Hypothesis 01

   “The documents are empty or lost.”

   Ruled out

   We re-fetched 150 of the underlying office actions straight from USPTO TSDR and re-extracted them. They are full ~9KB documents, not blank pages or dropped text. The examiner wrote real office actions; they simply weren't substantive refusals. Nothing was lost in the pipeline.

2. Hypothesis 02

   “The classifier missed real refusals.”

   Ruled out

   A 150-document hand check across every substantive ground — §2(d), descriptiveness, surname, geographic, the rest — found zero misclassifications. The classifier read these correctly: they carry requirements, not refusals. The low refusal count is the true count, not a detection failure.

3. Hypothesis 03

   “It's a different applicant pool, not a flood.”

   Ruled out

   It is precisely the pool that gives it away. The diluting actions cluster on foreign, overwhelmingly Chinese, pro-se applications — the exact cohort the counsel requirement and the subsidy-driven surge produce. The signal isn't random noise; it tracks a documented, datable shift in who is filing.

4. Hypothesis 04

   “The examiner changed their behavior.”

   Ruled out

   Their substantive numbers didn't move. Strip the requirement-only actions back out and the cratered year snaps back in line with its neighbors. An examiner who looks lenient on the raw per-OA rate is refusing marks at the same clip as before — the flood sat on top of unchanged behavior.

§ 4 · What this means for your prosecution

Three things to take to the docket.

1. 01

   A low refusal rate isn't always leniency.

   On a raw per-office-action basis, a docket buried in counsel-requirement actions reads as a soft examiner. It isn't. When you see a refusal rate that looks too good for the art unit, ask what the denominator is made of before you bank on it.

2. 02

   CrystalMark's rates already account for it.

   Every refusal rate and difficulty score on this site is built on substantive office actions with the requirement-only flood excluded. The number you read reflects how the examiner treats marks they actually judged — not how many foreign pro-se filings happened to land on their docket that year.

3. 03

   Filing-mill volume is the noise, not the signal.

   The surge is real and it is large, but it tells you nothing about the examiner assigned to your mark. We keep it out of the behavioral read on purpose, so the profile answers the only question that matters: what will this examiner do with a real, counseled application?