Game design is an art, but balancing a game? That is pure math.

Recently, I decided to put my algorithm skills to the test and entered the 2025 Tabletop Games Balancing Competition. The goal was straightforward but quite difficult: write an algorithm that automatically finds the fairest, most balanced set of rules for a variety of popular board games.

I’m thrilled to announce that my algorithm took first place! Here is a high-level look at how the competition worked and how we taught AI to be the ultimate game designer.

The Challenge: Black-Box Game Design

The competition treated game balancing as a “black-box optimization” problem.

Instead of manually playtesting with my friends over pizza, I wrote an algorithm that fed new rules into the TAG (Tabletop AI Games) framework. For example, my algorithm might say, “What happens if the starting hand size is 5 instead of 4?” The framework would then take those rules, silently play the game thousands of times in the background, and spit out a “Balance Score.” My algorithm’s only job was to blindly maximize that score by constantly tweaking the rules and seeing what worked.

The “Playtesters”

To figure out if a game was actually balanced, the framework had four different AI agents play against each other to simulate a real gaming table:

  • The Elite: A highly optimized AI (MCTS) given plenty of time to “think” about its turns.

  • The Good: The exact same AI, but given half the time to think.

  • The Greedy (OSLA): A player who just takes the best immediate move without planning ahead.

  • The Random: An agent that literally just picks moves at random (we all have that one friend).

The Games

You can’t just balance one game; a true algorithm needs to adapt. My code had to find the sweet spot for four wildly different titles:

  • 7 Wonders (Drafting and civilization building)

  • Exploding Kittens (High-chaos card drawing)

  • Can’t Stop (Pure push-your-luck dice rolling)

  • Dominion (The grandfather of deck-builders)

Getting a single algorithm to understand the nuances of Dominion strategy while also managing the pure chaos of Exploding Kittens was the ultimate puzzle.

The Catch (And What’s Next)

Winning the competition did come with a very nice financial prize! However, there was a slightly funny caveat: there was not enough people actually qualified for the final standings to hand out the cash prize.

I am currently preparing two highly technical, deep-dive posts explaining the exact code and math I used to win. However… I also want to defend my title next year. So, for now, those posts are staying locked in the vault! I’ll be keeping my secret sauce safe until the 2026 competition wraps up.