GameLab · Source archive

Gamelab: Elevator Madness

Originally published 2026-09-27 on Advisory Hour.

Originally published as an Advisory Hour Substack post on 2026-09-27. This first-party site copy preserves the text, localizes public supporting media, and connects the record to its project trail.

Original post

Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.

Elevator Madness began as an opensource model trying to recreate a threejs experience of the classic atari game Elevator Action. In that game, you use the elevator to exit the elevator on the ground floor. Along the way, the player must dodge or kill enemies, and plan their use of the elevator appropriately.

Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.

The first build with the opensource model was terrible. So were the following three. Eventually, to be frank, I gave up on the project. It was a pile of code to nowhere. Then when Opus 5.5 dropped, I figured… what the heck, let’s see what Opus5.5 does. And… wow.

Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.

The game is ridiculous. If I didn’t know better, the temptation would be to say that Claude was having fun pulling together all the opensource code to turn the project into something useful. There’s even a flying karate kick you can do as the player. That’s the level of ridiculous that’s at play here.

Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.

There’s even shaders applied to damaged states.

Gamelab: Elevator Madness image from the original Advisory Hour post.
Gamelab: Elevator Madness image from the original Advisory Hour post.

There’s also great wall art. Here’s a gallery of the different wall art available.

This project raised an interesting idea. Is it better to send a grunt forward to build out a project and then come back around to have a more capable model like Opus5.5 or Astra “fix” it? Or, is it better to just start at the smarter model and see what you get?

What I’ve found so far in early tests is that the models seem to do better working with a base project, even if it’s wrong, then starting from scratch.

And that is why I’ll be posting on more variants in the coming days.

Play Elevator Madness: Elevator Madness: Rival Protocol

Read more about the project: Elevator Madness: Rival Protocol | GameLab

Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.
Gamelab: Elevator Madness gallery image from the original Advisory Hour post.

Connected work