The Paywall Theory Changed After the Minute-by-Minute Log
Jack's 90-day funnel looked like a clean little funnel:
151 App Store impressions -> 15 product-page views -> 8 downloads -> 1 paid subscription.
The first explanation was attractive: the user saw enough value, then paid when the paywall appeared.
Then Jack went back to the minute-by-minute log and changed the story himself.
The payment happened during a bad day, before the prediction had been confirmed. The earlier explanation was no longer strong enough to carry the conclusion.
That correction is more useful than the funnel by itself. A small sample can show that a payment happened. It usually cannot tell you which moment caused it, especially when several events are close together.
Jack's next idea is narrower: watch what happens after a second accident and a reset card, then see whether that sequence repeats. In the comments, he described two events the build now logs around that moment: the reset card appearing after a second accident, and someone starting the reset. For now, it is a hypothesis with one observation behind it.
The part I would keep open is whether the same product moment appears again, or whether the first payment remains a one-off story that only looks coherent after the fact.
When a first payment changes your interpretation, what evidence makes you keep the old explanation, and what makes you retire it?
Event-level logs humbling a clean funnel story is so common it should be a rite of passage. Aggregate conversion rates lie when the real drop happens in a 30-second confusion moment you never instrumented. What was the biggest mismatch between your original paywall theory and what the minute-by-minute data actually showed?
The biggest mismatch was the order of events. Jack's later comment says the user backfilled five earlier accidents during the first six minutes, then subscribed at minute 50, before any prediction had been confirmed. That rules out a confirmed prediction as the trigger for this purchase. It still doesn't tell us what caused it.
That distinction is useful—the timeline narrows what the payment can explain without pretending the remaining cause is known.
minute-by-minute logging is the kind of obsessive measurement that actually changes minds - aggregate stats never do that. we're free with no paywall at all on swapfile.live (file conversion, in-browser) and honestly haven't measured closely enough what a paywall would do. what did the log reveal that surprised you most?
What surprised me most was Jack's own reaction to the sequence: "I got this wrong in the post." He could have left the first-payment story as it was. Instead, he published the detail that undercut it: the user paid before any prediction had been confirmed. That correction was the reason I wrote about his case.