Store Listings as Code
12 October 2026
Each of our games has a page in the Apple App Store, on Google Play and in the Microsoft Store. Those pages are the first thing a parent reads, and nearly the last thing anyone on our side looks at. Once a listing is written, nobody opens it again until something is obviously wrong, and an outdated listing never looks wrong. Release notes from three versions ago look exactly like this week's. With nine games in three stores, all of it was typed into store consoles by hand, so we are moving it into code, where each release writes it.
The release notes are the part we got wrong first. For a while, a language model wrote them from our commit messages. It was fluent and confident, and it described changes that did not happen. A fix for a server timeout became "scores and leaderboards recalculated more accurately". An Android-only fix went to the Apple App Store too, where nothing had changed. The model was not the problem. Its input was, because a commit message says what changed in the code, not what a player will notice. So each issue we fix now carries one plain sentence, written by a person when the change is made. Most issues are internal work and carry none. At release, the notes are those sentences, put together for each game, with no model anywhere on the path. A release with no notes cannot go to the stores.
Screenshots come from the game itself. For each release, the build plays every game on every screen size the stores ask for, from a phone to a television, and captures what the game really looks like in that release. Apple's iPhone, iPad and Mac sizes are joining the capture that already fills Google Play and the Microsoft Store. A screenshot cannot show a screen that no longer exists.
Then there are the settings. These include the categories, the price, the review contact, and the answers to each store's age-rating questions. When we checked the games we had set up by hand, the review contact had two different phone numbers, and the age-rating answers had drifted into three different sets across six games. Now each game declares these values once, in the same registry that the apps themselves read, and the release writes them to every store on every run.
One more rule protects what a parent sees. Our testers get new builds through the stores' testing tracks, and a store wants a complete listing before it will release even a test build. A listing that already has its pictures keeps them until the production release, so the public page changes only when the version it describes ships.
Some settings cannot be written from code, because the store offers no way for a program to change them. Where we can still read such a setting, each release checks it instead. Google Play, for example, offers no way to set a game's category, so each release reads the game's public store page, compares the category with the registry, and stops if they differ, naming the value a person must set by hand. Where we cannot even read a setting, as with Apple's App Privacy answers, a person answers it in Apple's console for each new app, and again whenever what an app collects changes. We would rather say so than call the whole listing automated.
All of this reaches the stores with our next release.
