Seven days from the first commit to App Review.

A game in Cardzo is a markdown file. The frontmatter says which template and scorer it uses, how many players it allows, how it is won, and which settings it exposes, with a key, a type, a default, and a range for each. The body is the rules, and a translation is a second file with only a body.

A codegen step turns those 78 files into one TypeScript module. Adding a game means writing a file, not code, and the website computes its own totals from the same files at build time, which is how the catalog can grow from 70 games on day one to 78 without a number being edited by hand.

The engine has one pure function: score takes the game definition, the settings, the players, and the entries, and returns the sheet. Computed totals are never written anywhere. That is what makes house rules recalculate mid match: flip shoot the moon in Hearts and the same entries produce a different sheet, instantly, with nothing to migrate.

Undo and redo are pure mutations on two stacks, and changing a setting is undoable like any other move. Entries are written to SQLite synchronously, on purpose, so a score is durable the moment it is set.

The first rule was no accounts and no backend. A week after the first commit the plan narrowed to a promise that still holds: the scorepad works offline forever, and accounts are optional. Supabase stores facts for the social layer, Circles, rated matches, and live nights, and nothing else.

Ratings are replayed on the phone from stored match facts, so the server has no formula to get wrong. The project runs on the free tier in Stockholm, kept awake by a daily cron job from the website.