The Event contract before code freeze module forced our Android and iOS leads to agree on checkout_started properties in writing. We still argue, but we argue against a sheet instead of a Slack thread that vanished.
Reviews
Not every gate ends in a ship. That still counts.
These notes come from people who sat Release Signal Lab, Crash & Retention Audit, or a Watch Floor cycle. We kept the awkward ones. Glowing blurbs without a module name do not belong here.
Crash & Retention Audit made us stop quoting crash-free percent in the ship email. Slightly less flattering. Much harder to ignore on mid-range Android.
Week four of Release Signal Lab assumed more SQL than our growth intern had. The schema still came out cleaner; just budget an extra coaching hour if the team is product-heavy rather than data-heavy.
8 / 10
Store Listing Experiment QA · “Short, slightly dry, exactly what we needed before a Thai subtitle test. Isolated listing traffic from first-open. Deducted points because homework landed on a public holiday.”
Thai Store Readiness Instrumentation caught Buddhist calendar dates serialised as strings on iOS and numbers on Android. Our warehouse join had been silently dropping a slice of northern users for two releases.
Case study · wallet, Bangkok
Double-fire on Android 14 after a WebView bump
A consumer wallet scheduled a Play release on a Thursday. Blended Mixpanel funnels looked calm. In Release Signal Lab week three they split checkout_started by OS build and found Android 14 firing the step twice after a WebView library bump.
They held the submission, patched the double-fire, and republished with a version-scoped dashboard. Activation recovered on the candidate without the “marketing miracle” someone had already drafted. The team still uses the gate memo format internally.
Limitation they reported: the lab did not have time to rebuild their identity graph. They left with a cleaner funnel, not a new user model.
Case study · travel, remote + Chiang Mai
Hold-for-sample on a bilingual onboarding rewrite
A travel app rewrote onboarding in Thai and English in the same candidate. Crash-free sessions improved. Completion of “save first trip” did not, once locale was split. Sample on the Thai path was 180 users after five days — too thin for a ship call.
Gate outcome: hold-for-sample, then a smaller staged rollout. The PM disliked the delay; the QA lead later said the hold avoided a one-star week driven by a mistranslated date picker, not by “engagement.”
