An iPhone app with one user, built by sessions that forget.

I buy most things on Shopee and Lazada, and then I forget about them. My RM 4,600 laptop, used almost every day for nearly three years, now costs less per day than the RM 1,500 racing wheel I bought this January.

So the whole app is built around one number: the price, plus any upgrades, divided by the days I’ve owned it. On an item’s page, that number is bigger than the price. It falls every day I keep using the thing, and stops when I let it go.

A paid iOS app already does this. I wanted my own version, for me alone, without borrowing its name, icons or artwork.

In the middle of that, Things Pro became 值得, “worth it”. Claude drew six icons and I picked the fifth: a box with five ticks on its side, like days on a calendar, with the last one blue.

Each Claude session starts with no memory of the last one. On a project this size that matters more than any single feature: a new session that guesses where things stand will happily redo or undo work.

So the project has one roadmap file, in Chinese, that every session reads before touching anything. It says what’s done, what’s next and what was decided, with the reason next to each decision. At the end of every stage, the session updates it.

On 27 October I signed off four rules, now at the top of that file:

The rules also kept two sessions apart. The redesign ran in its own session, and the new features waited until it was finished, so the two never edited the same files at once.

The scariest thing this app could do isn’t look bad. It’s to open after an update with nothing in it. So the roadmap has a section on changing the database, and it’s the one part that never gets skipped.

The database changed shape four times, and each change came with a test that builds the old version, fills it, upgrades it and checks nothing went missing. Old backups still import. Photos are saved by name instead of full path, after we found that a reinstall moves the folder and breaks every image.

The rename followed the same rule. The name on the home screen, the icon and the backup files changed. The app’s internal ID and data folder did not, because changing them would make the iPhone treat it as a brand-new app, with an empty database.

So did the English version, added last as a setting. Anything already stored stays in Chinese, like the four ways an item can leave, and is only translated on screen. Switching language changes nothing in the database or in a backup.

For the redesign, Claude drew a design system first: a light grey canvas, white cards, one cobalt brand colour, black kept for cards about money and a lime used only for highlights. Then seven screens built from it, which I reviewed on a shared canvas before any code changed.

Most of it I kept. Four things I changed, and the file records each one as overriding the design system:

Archiving something and marking a wish as bought happen often, and neither destroys anything: an archived item can be restored, and a bought item keeps its photo, tags and history. Both used to open a yes-or-no dialog. Now they’re a slide: deliberate enough not to happen by accident, quick enough not to feel like a form.

Deleting can’t be undone, so it stays a red button with a dialog. The archive sheet shows how long I kept the thing and what it cost per day before I commit, and the four ways to let go — sold, gifted, scrapped, lost — stayed, even though the mock-up left them out, because older records already use them.

Typing in every purchase is how apps like this get abandoned. So the add form starts with “Scan an order screenshot”: pick a Shopee or Lazada order, and a vision model reads the name, price, date and shop.

It fills in only what it recognised, says so in a blue note, and never saves by itself. A date in the future or a negative price counts as not recognised, and an order in another currency or with several items gets a warning. The API key stays on the phone and never goes into a backup.

The analysis tab is grouped into money, worth it or not, and reminders. In the middle group, you set a line, RM 1 a day by default, and every item shows how many more days of use it needs before its cost per day drops below it.

It turns an old purchase into a small goal. The Edifier speakers need another 23 days. The chair and a pair of Nike Air Force 1s are already there.