Starting point
A game that was fun and not fair.
Pastry Pirates began as a board game: sail between islands, buy cocoa and flour, dodge the other captains, get home and bake. People liked it, but repeat players saw one strategy win too often. So we asked which rules would make games close, quick and varied.
We took it on as a studio project, to test our methods before we use them for clients.
What we found
Simulation shows what each rule does.
Our Python simulation had strategies play each other: the hoarder, the trader, the raider, and the one who sails straight home. We got win rates for each rule variant, exact coin-flip odds, and a ruleset where no strategy wins more than its share. We cut several rules we liked. One that seemed decorative turned out to keep games close.
The lesson we still use: simulate a system first, then decide together.
Where we went
Online now, with a physical edition on the way.
The game runs in the browser, each player on their own device. A physical edition with the same rules is in production.
AI agents wrote most of the code, under a process we designed. One agent proved a defect existed before another could fix it. A "Sea Trial" tested the whole game automatically after each change. A checker relied on evidence, not reports. We packaged the process as ClaudeKit and used it to make this site.
What changed
A working game, and a working method.
People play it. We can now balance rule-based systems by simulation, and run AI teams we trust with real work.