4% of creator codes that passed a checkout test stopped within 30 days
Among 399 creator codes that passed a checkout test, 4% had stopped by day 30, 11.8% by day 60 and 17.4% by day 90, on a short measurement history.
By Pohex Editors. Published October 29, 2026.
Stops within 30, 60 and 90 days
Pohex's software tested creator codes at store checkouts between 2026-07-28 and 2026-10-29. The cohort holds 399 codes that passed at least once. Of these, 40 stopped and 359 were still working at their last recorded test. The Kaplan-Meier estimate puts the share stopped by day 30 at 4%, with 257 codes under observation at that horizon.
By day 60 the estimate rises to 11.8%, with 186 codes at risk. By day 90 it reaches 17.4%, with 50 codes at risk. The risk set shrinks at each step because newer codes have not been observed for that long.
What counted as stopped
A code counted as stopped at its first strike or retirement recorded after a passing test. Of the 40 stopped codes, 36 stopped at a strike and 4 at a retirement. The median code was observed for 52 days.
A single failed test counts as a stop even when a later test passed. That happened for 14 of the 40 stopped codes, so some of these stops may have been temporary.
Month by month
Dividing the codes that stopped in each calendar month by the codes under observation in that month gives 0% for Jul 2026 (85 codes), 1.7% for Aug 2026 (239 codes), 6.3% for Sep 2026 (285 codes) and 5.1% for Oct 2026 (350 codes). The series describes what was measured. It does not show why the monthly share moved.
Limits of the history
The measurement history starts on 2026-06-30, and nothing earlier is on record. Codes first seen at the start may have been working for some time already, which can make codes look shorter-lived than they are. The median code has 2 recorded tests, and 560 gaps of more than 14 days between tests occur across 291 codes, the longest being 93 days. A stop date can therefore trail the real one by up to the length of the gap.
A further 394 codes were retired with no passing test on record and sit outside every figure here. The 90-day estimate rests on 50 codes and should be read with that size beside it.
Method
Question: Of the creator codes that passed a checkout test, what share stopped working within 30, 60 and 90 days?
Cohort: Every creator code in the measurement history with at least one passing checkout test (a measured, live or revived event), on a store the site lists. A code enters the cohort on the day of its first passing test.
Denominator: Codes in the cohort that were still under observation at each horizon (the Kaplan-Meier risk set), not every code ever listed. Codes retired without any passing test on record are outside the cohort.
Observation window: The full measurement history, from the first recorded test to the last. In this edition: 2026-07-28 to 2026-10-29.
What is counted: A code counts as stopped on the first strike or retirement recorded on or after its first passing test.
Censoring: A code with no strike or retirement is right-censored on the day of its last recorded test. Kaplan-Meier estimates the share stopped by day 30, 60 and 90; a monthly series divides the codes that stopped in each calendar month by the codes under observation in that month.
Sampling: Codes are tested on one cart, one variant, quantity 1, no shipping, as a new customer, on the sweep's schedule rather than daily, so a stop date is the first test that failed, not the day the store disabled the code.
Counts: 399 codes in the cohort, 40 stopped (36 at a strike, 4 at a retirement) and 359 censored at their last test.
Limits of the data
- The measurement history runs from 2026-06-30 to 2026-10-29; nothing before 2026-06-30 is on record.
- The median code has 2 recorded tests; 416 of the 746 codes in the history have two or more.
- 560 gaps of more than 14 days between tests occur across 291 codes (the longest is 93 days), so a stop date can trail the real one by up to the length of the gap.
- 46 codes were measured once and never tested again.
- 394 codes were retired with no passing test on record; the tests that failed before the retirement were not kept, so these codes are outside any survival figure.
- 376 events across 210 codes carry the date their data file was committed rather than the date of the test (day precision at best).
- 14 of the 40 codes counted as stopped passed a later test again.
- A single failed test (a strike) counts as stopped even when a later test passed again; the study reports how many stopped codes later passed.
- Codes first seen on the first day of the history may have been working for some time before it (left truncation), which makes codes look shorter-lived than they are.
- The 90-day estimate rests on few codes, because the history is only about four months long; read it together with the risk-set size printed beside it.
- Stores the site does not list (restricted categories, development hosts) are left out, so the cohort is not every code the sweep tests.
Data: download the CSV (399 rows). The same history feeds the verification feed.