The beta: first contact
Other people used it, on their own phones, for the first time. The product held up better than the instruments did.
Everything up to here was built by someone who already knew how it worked. In late July, other people used Alter for the first time, on their own phones, live, with no one narrating over their shoulder.
What it was. The first honest test of every assumption at once. Not a demo, where you drive; a session, where you watch.
What went wrong. Two different things, and the second was the more useful.
The product mostly held. The instruments did not. Reconstructing what had happened in a single afternoon took about fifteen hand-written database joins across four tables and four containers’ logs. Errors that people had actually hit existed only in container logs: no events, no alerts, nothing that would ever reach me unless I went hunting. There was no concept of a session anywhere in the system, so the most basic question: did anyone come back?; had no answer at all. I had built a product I could not observe.
The second thing was quieter. Watching someone meet the onboarding flow, I could see it asking for things it then never used (a screen that promised to take an answer seriously and did nothing with it) and taking long enough on a phone that the pause became the experience. Written down, that flow looked thorough. Performed by a stranger, it looked like an interrogation.
What I decided. Build the missing half before building anything else: a journey read model derived from the event log I already had (no new service, no analytics vendor) plus request attribution, errors promoted to first-class events with alerting, and a report that arrives by email whether or not I remember to look. Zero AI in any of it. (The full write-up.)
Then, having built the ability to see the product, act on what it showed: onboarding was rebuilt from five screens down to three (drop a ramble, read what it made of you, see your life mapped).
The larger lesson is one I’d now front-load into any project. You can’t iterate on what you can’t see, and the instruments are never a later phase. I’d treated observability as launch hygiene. It’s actually the thing that decides whether the next six months are guesswork.
Next era: The restructure