Craft Craft that ships. that ships.

Strength logging is a crowded shelf, and almost every app on it runs on the same engine: streaks, badges, a feed. So before designing anything, I sat down with four serious lifters. The same answers kept coming back. They already track their training. They judge a set by the reps they had left in the tank, not just the weight on the bar. And every app they'd tried buried that under gamification. I'm one of them, which helped me hear the thing nobody said out loud: a serious lifter already knows whether they showed up. Turning up isn't the question.

So the problem wasn't motivation, and it wasn't more data. It was proof: a lifter can't easily see whether the work is working. That gave me the line I designed everything against — strength, kept honest. Two decisions locked here and never moved. Reps-in-reserve sits at the centre of every logged set, so the app knows how hard each set felt. And strength over time is shown honestly, so an easy day is never dressed up as progress.

With the problem fixed, the open question was how to show progress without lying about it. I explored a few readings of the same training data: an honest max estimate adjusted for effort, so a lucky day doesn't read as a gain; stall detection that watches the trend, not one bad session; muscle balance and personal records by rep range. I didn't decide alone — once early users were in the private beta, a feature survey ranked what got built, and the winners became the progress screens.

The brand grew in this diamond too. A ram's horn adds a ring for every year it lives, and training adds a notch every session. That gave me The Keeper, the ram, and a notched horn for the mark, wrapped in Dark Ledger: warm near-black, monospace numerals, quiet PRs with no confetti, nothing on screen unless it means something. Then the part that proved the definition was real. I built a streak counter, liked it, and killed it, because a streak is the exact game Notch exists to end. Deleting it sent me back to check the one-line problem still held. It did.

Converging was mostly subtraction. The exercise library went from 873 to 438, keeping the lifts a serious program actually uses, and features I was excited about moved to after launch. I designed every screen and directed AI dev tools through the build, so the design lives in the code and nothing dies in a hand-off. What I'll watch after launch isn't installs. It's whether lifters keep coming back because the app told them the truth.

The homepage was built for people who already knew the brand, not the cold traffic actually landing on it from search and ads. It opened with sixty-plus challenge cards in one dense grid. No pitch, no explanation, nothing telling you where to look first. The analytics said it plainly: first-timers arrived cold, met the wall of choices, and left without picking one.

They weren't short on challenges. They were short on a path. That reframed the whole brief: stop treating the site like a catalogue and start treating it like a sequence. Lead with the pitch, feature a handful of challenges instead of all of them, put the proof near the top, and give each section room to be read at a glance.

Working with a developer and a CRO, I designed against that sequence: a homepage that opens with the pitch, a five-step how-it-works, and proof — 450k members, an 87.93% completion rate, press logos — before a short list of popular and licensed challenges. The tempting move was to cut the catalogue of challenges down. Instead we kept all of it, one click away, so nothing was lost, only reordered. Underneath it, a shadcn design system so design, dev, campaign and creative teams work from one source.

Three plans sat in three separate tables, each with its own call to action. The page competed with itself, asking people to solve a comparison puzzle at exactly the point where hesitation costs the most: the step where money changes hands.

Three tables side by side turns picking a plan into comparing three plans first. So the brief narrowed to one sentence: make the page ask one decision at a time — which plan, then one button.

Compress three tables into one plan selector with a single button underneath. Same comparison, one control instead of three. The narrowing happened live: four to five A/B rounds with a developer and a CRO, each version tested against the last.