Recomp shows you the progress the scale hides.

The scale is the one number that lies during recomp.

Body recomposition feels like a long road to nowhere. The scale bounces between the same five pounds no matter how consistent you are — because the scale can't see muscle replacing fat. The real proof lives in progress photos buried in a camera roll, measurements scattered in a notes app, and trends nobody connects.

Existing trackers are built around weight. So people quit programs that were actually working. They just couldn't see it.

I know this because it was me. During a recomposition phase I had photos in my camera roll, measurements in Notes, and weight in a separate tracker — and I still couldn't tell if anything was working. The scale said no. My body said something else. Nothing connected the two.

Nora doesn't need more discipline. She needs to see it.

Nora starts and stops diets and walking programs about once a month. Her consistency is tied directly to the scale: when the number drops she keeps going; when it stalls she's back to old habits within weeks.

Her real problem isn't effort. It's that the only instrument she's measuring with is the one that lies during recomposition.

This reframed the entire project: Recomp isn't a tracker. It's a motivation engine that surfaces the progress the scale can't. That insight drove every major decision that followed.

Every tool owns one piece. None connect all three.

Before designing anything I tore down the four tools people actually use, looking for where each fails a recomper specifically.

The fastest way to kill this project is to build everything.

I wrote a strict scope before designing and pressure-tested every feature against one question: does Nora need this to see her progress?

A decision worth highlighting: during scoping I realized my first feature list answered "how does Nora see progress in the app?" but not "what gets her to open it at all?" Those are different problems — and for someone whose failure mode is stopping, the second matters more. That's why Weekly Wins and the streak made the v1 cut. Seeing progress is worthless if she never comes back to see it.

A small number of high-leverage calls, driving everything.

Designing the empty days and the honest ones.

Making "empty" feel like a beginning, not a dead end. Day one is exactly when Nora is most likely to quit — there's no progress to show yet. So the empty states carry the motivation on tone alone. The week-one dashboard shows her baseline photo beside a placeholder previewing where the next one goes. An empty state isn't "nothing here yet" — it's a promise of what's coming.

An AI that admits what it doesn't know. With one check-in logged there isn't enough data for a real insight. A lesser version fabricates one anyway. Recomp's says: "I need one more check-in before I can spot your trend. Recomposition shows up over weeks, not days."

Admitting the limit rather than faking confidence is what makes the insight trustworthy when it does speak — and trust is the whole reason someone believes the app over their scale.

Not a set of screens. A closed loop.

Onboarding hands off to the dashboard; the dashboard starts the weekly check-in; the check-in's payoff returns to the dashboard. A new user can walk the entire product and end up back where they started, ready to do it again.

It runs on a small, deliberate design system — a warm, grounded palette (sand, warm charcoal, and a single emerald accent spent in exactly two places: the primary action, and the companion's voice), a display serif for emotional moments paired with a clean UI sans, and consistent spacing so every screen feels like one product. When a surface is emerald, the companion is genuinely speaking — including when what it has to say is "I don't have enough data yet." Reserving the accent that tightly is what makes the wins land instead of competing for attention.

11 screens, a working prototype, and a defensible thesis.

A complete, prototyped product design covering the full lifecycle — onboarding, the weekly check-in, the signature progress dashboard, and the day-one and insufficient-data states most apps ignore — wired into a clickable prototype with a narrated walkthrough.

More important than the screen count is what the design proves: a clear product thesis backed by competitive evidence, scope discipline about what not to build, and a system designed around the real problem (retention) rather than the obvious one (tracking).

What I'd measure next. Designed with specific success metrics in mind, to be validated live: first-check-in completion, week-two retention (the true test of the come-back engine), and check-in completion rate and time (the test of the under-two-minutes goal).

The obvious problem usually isn't the real one.

Recomp looks like a tracking problem. It's a retention problem. Naming that correctly changed every meaningful decision.

Scope discipline is a design skill. Deciding what not to build — and being able to defend each cut — did more for this product than any individual screen.

Designing the empty and the honest moments is where maturity shows. The day-one states and the AI's willingness to admit uncertainty aren't glamorous, but they're what make the product trustworthy and humane.