Montana

Why it exists

What makes a maze difficult in a way that stays interesting? A large map is easy to make. A map that makes you doubt your memory, or lets you walk past the answer without recognizing it, is a more interesting problem. Montana is a first-person game in Unreal Engine about navigation, disorientation and impossible space: a place that invites exploration, then quietly challenges your understanding of how it works.

Approach

The project has grown into two experimental tools.

MazeLab generates, searches, playtests and preserves mazes. It supports candidate selection, player ratings, run results and session archives, so I can play a maze, record how I moved through it, and compare that with what the generator expected.

RealityLab is a workbench for "reality warpers": controlled changes to the environment itself, such as a hallway that gets longer when you walk it backward. It exists so I can design and test one warper quickly without routing every small change through a coding agent.

The plan is to build a library of warpers, drop them into real maps, and combine them with selected maze seeds to play reality-warped mazes. That combined game is still in development; the labs are how I find out which ingredients deserve to go into it.

One experiment that changed the project. In a 15x15 playtest the exit sat only four grid cells from the start, with a shortest legal route of 24 transitions. I passed right beside it early, yet reaching the gateway took 990 transitions, covering 207 of 225 cells with 784 revisits. That is a single run, not an average, but it captured something a simple distance measure misses: an answer can be nearby and still be very hard to recognize or reach.

Engineering the feedback loop. Recorded sessions preserve the maze structure and ordered movement trace. One reported 100,000-maze generation benchmark improved from 243.7 seconds to 94.0 seconds, roughly 2.6x, on that specific benchmark.

What I learned

Observability is part of the creative tool. I lost a full day debugging a hallway visual that turned out to be caused by a ceiling element, which led to in-session toggles for ceiling, lighting, material and a debug overlay. I also learned to build the second warper before building more architecture around the first. The tools exist to make trying an idea, feeling it, and deciding whether it's good as fast as possible.