Just Train

A workout app for the self-coached athlete — the person training hard with nobody watching the numbers. Its whole job is one loop: capture the set, compare it with last time, communicate the progress back to you.

The move that made it a product

I built it feature-complete in thirteen days. Then I stopped — and instead of shipping, I used it. Real training data went in, which surfaced affordances I hadn’t designed for and bugs I couldn’t see from the outside; my own experience as the athlete living in it pushed the UX somewhere better. Then I ran the full product pass — personas, a Kano model, user flows, an event storm, bounded contexts, a shared glossary — and rebuilt on that foundation.

The interesting half of this project is the second one. The first half is what you get from momentum; the second is what that same product looks like once a real domain model is underneath it. Seven months, and most of them went here.

A few decisions worth the name

• The athlete is the atom; the trainer is a layer. I started building for trainers and their rosters. Partway through I inverted it — the athlete’s own experience is the product, and the trainer is a layer on top. It changed onboarding, the data model, and who the app is for.
• Records are derived, not stored. A personal best isn’t a thing you save; it’s a fact you can always recompute from history. Store it and the two eventually disagree. So the history is the only truth, and records fall out of it at read time.
• Offline is a contract, not a feature. People log sets in basements with no signal. Offline couldn’t be bolted on afterward — every write is built to be safe to replay, so the app is honest whether or not the network is.
• No AI, no voice, no wearables — said as positioning, not apology. It’s a quiet, fast logger that does one thing without ceremony. Deciding what to refuse was as much of the design as deciding what to build.

The record keeps its reversals

The decision log keeps the calls I got wrong next to the ones I got right — the billing model I chose in June was gone by September; the product still doesn’t have a settled name. That’s the point, not an embarrassment: the reversals are the evidence the process was real.

Where it stands: pre-launch, and being taken to the finish the same way it was rebuilt — decide, then build.