Precedent

Precedent is freeze tag, with a referee you can argue with.

Before each level you get to argue for one new rule, in plain words. "Kids should be able to jump over pillow forts." "A kid under a table can't be hit." That kind of thing. The referee decides whether it's fair, writes it up as a real rule, and the game plays by it from then on. The catch is that every rule binds both sides: you get it to yourself for one level, and from the next level on the other kids use it too. The referee will also sometimes point out what your rule admits (if a kid can jump a pillow fort, a thrown ball can probably get over one as well), and that goes in the rulebook for everyone.

So a lot of the game is figuring out which rules you actually want. There are eight levels, from capture the flag in the backyard to king of the hill on an ice rink to the floor is lava, and you're scored like golf, against par.

The referee is an AI

The referee is a language model (Gemma, which is open-weights). The game itself is ordinary code with a fixed set of things a rule can say and a points budget for how strong a rule can be, and the model's job is to read your argument and write it up in those terms. So a rule that's too strong gets refused or cut down, however good the argument is. It does get things wrong now and then, and if it rules something that seems plainly wrong, please tell me in the comments. That's most of how I find out.

What it costs

Every argument costs me a little in tokens, roughly 30 cents for a full run. It's free to play, so there are limits: $5 of arguing per player each month, and $50 a month for everyone together. If the referee runs out, you can keep playing with the rules you've already won, and arguing comes back on the 1st. If you'd like to chip in, there's a tip jar on Ko-fi (thank you!): https://ko-fi.com/russhamer