The Maintenance Cliff

§ 1 · Two drivers, one effect

Neither explanation works alone. Together they're complete.

Two forces arrived at examination in the same window, and each fits a different part of the signature in the data — the statute explains the timing, the volume surge explains the persistence.

§ 2 · It's not a data artifact

Four ways this could be noise. Each one, ruled out.

The refusal rate held flat at roughly 40% across every year in the window. Examiners kept refusing at the same clip — only the share they held on response moved. That rules out a change in applicant quality and points straight at examiner behavior. Here are the artifact explanations we tested and discarded.

1. Hypothesis 01

   “It's just COVID.”

   Ruled out

   The timing rules this out on its own. If pandemic disruption drove the drop, the cliff would land in 2020–2021. Instead, examiner behavior in 2020–2021 was nearly identical to 2018–2019; the break is 2021→2022. COVID is the volume driver, but it can't carry the timing alone — the TMA boundary does.

2. Hypothesis 02

   “Missing event codes.”

   Ruled out

   We audited the full bulk-XML event-code catalog. No new concession or withdrawal code appears starting in 2022 that would mechanically inflate the measured concession rate. The signal isn't an artifact of a code we started counting.

3. Hypothesis 03

   “A pipeline bug.”

   Ruled out

   A 1,000-case TSDR sample was hand-checked against the classifier output; no maintained-vs-withdrawn misclassification clusters in 2022. Every known stats-pipeline bug was fixed and the years recomputed — the cliff survives each pass.

4. Hypothesis 04

   “The fleet changed, not the behavior.”

   Ruled out

   A same-cohort analysis on 601 examiners who were active across the whole window shows their individualmaintenance averages fell too. This isn't lenient new hires diluting a fixed fleet — the same examiners are folding more than they used to.

§ 3 · What this means for your prosecution

Three things to do differently.

1. 01

   Borderline arguments are worth making.

   A §2(d) response that would have lost in 2019 has materially better odds today. The fleet is conceding on the margin it used to hold. If the argument is reasonable, file it — don't pre-concede to an abandonment.

2. 02

   Concise responses outperform.

   Examiners have nearly twice the docket they carried in 2021. A tight, focused response that makes the concession easy beats a sprawling one that demands a close read the examiner doesn't have time for.

3. 03

   Post-publication remedies now shape the OA.

   Refusals you might have lost in 2019 are getting through, in part because the examiner knows expungement and reexamination backstop the register. Factor that shift into fight-or-abandon calculus — the examiner already has.