Crazy Go — an experiment in turning Go into a roguelite on non-standard boards

Go Variants
1

Hello fellow designers!

I’ve recently hit a milestone on a digital project I’m developing and I am looking for playtesters to help me analyze the core game loop and progression system.

The project is called Crazy Go. My design hypothesis was: Can the deterministic, perfect-information mechanics of the ancient game of Go be used as a compelling combat system for a single-player roguelite?

The Core Concept: Instead of fighting monsters with swords and RNG dice rolls, your “battles” are matches of Go. However, to make it fit the roguelite genre, I had to introduce asymmetry:

  • The Boards: You don’t play on a standard 19x19 grid. You play on procedurally eroded maps, spirals, islands with choke points, and hexagonal graphs.
  • The Progression: As you defeat opponents (driven by a custom Neural Network AI), you earn “Komi” (permanent points advantage) and draft Spells (like playing a 2x1 domino piece, or gaining a temporary shield).

A quick note on the AI: While the custom AI currently struggles to fight effectively on the standard and non-standard topologies. I will keep trying different methods to fix the algorithm because the board variations are really fun, but for now, the MVP is fully playable so you can test the concept and see if you enjoy the mechanics!

I am specifically looking for feedback on:

  1. The Roguelite Loop: Do the spell drafts and champion passive abilities feel rewarding?
  2. The Fun Factor: Does playing on a board with choke points and missing corners feel engaging from a puzzle/strategy perspective? Even with the current AI limitations, does the concept hold promise?

The game is fully playable in the browser (no downloads required, no account needed). You can try the MVP here: https://victologo.itch.io/crazy-go

Any feedback on the design theory, the UI, or the mechanics would be incredibly helpful for the next phase of development. Thanks in advance!

3 Likes
2

Seems like a fun idea. I played some games, too short to really get a feel for it, but I enjoyed the meteor strike raining on the board.

Two things for feedback

  • UI: clicking away the “game start” overlay already places a stone on the board.
  • Bot: 9x9 beginner bot doesn’t pass early enough, it plays on for too long until, even after all its stones are gone and the board is almost filled with mine.
3

Thank you, I’ll bear these improvements in mind for future versions (I think that training the current game algorithm with more 9x9 or 13x13 games on a square board will easily and quickly improve the neural network; I might have it ready in a day, although for non-standard boards the improvements will have to wait a few more weeks).

4

I’m curious why for square boards you’re training your own neural network instead of just using KataGo.

1 Like