whoami

This is a working log, not a product page.

I'm building a framework for an action roguelike — the aim being that most of the game logic lives in the framework, so swapping in a different set of assets gives you a meaningfully different game. I started by pressure-testing that with store-bought asset packs, seeing what held up when the art changed underneath it. Since September the art comes another way: I stopped letting AI agents read store-bought packs (a license decision, written up in Where the art comes from now), so new models are commissioned to a written brief from a modeling agent, or come from CC0 libraries.

It's in VR, and that turns out to matter more than "it's in VR." The player's body is the input, so almost nothing can be judged from a desk. Repeatedly, something has been provably correct, passed every test, logged exactly what it should — and been wrong the moment someone wore it. A damage band that had been switched off for weeks because its ceiling sat below the average swing. Enemies that "couldn't fire" and were firing. A mechanic that taught players the opposite of what it was for. Each of those looked healthy in every number I had.

So a lot of what I write here is about measurement: how to ask a question in the shape of the thing you're actually asking about, and how to tell a real result from a plausible one. That's become as much the point as the game.

There are two projects now. Experiment 1 is the one most of these posts are about: a VR roguelike on Meta Quest, with a run loop, a hub that grows, and items worth choosing between. Experiment 2 is an action RPG meant to stand on the same framework — single-player on foundations built for multiplayer, for PC VR — and it is still at the stage of decisions rather than code.

Whether either turns into something I release is still open. It's closer than it was, but the point is still getting the systems solid enough that a more serious project could stand on them, and keeping an honest record of how that goes.

Two kinds of posts: devlog entries about what got built, and learning entries about what I got wrong or had to redo. The systems page tracks the individual pieces and how far along they are.