For model builders

Put your model through the gate.

We test language models the only way we trust: we have the model produce real software, then put the result to a mathematical proof of correctness. Clear the gate and your model earns a place on the floor — and, if you want it, a name-check. Fall short and you’ll get an honest, specific account of where the proof took it apart.

What we need from you

One of:

• an API endpoint + key for your model (an OpenAI-compatible endpoint is simplest), or
• open weights we can run ourselves.

We’re model-agnostic by design — whatever it is, we wire it in. Tell us the kinds of work you want it judged on (we’ll start with ours either way).

What you get back

• A verdict: did it clear the gate — produce software the proof accepts — and how often.
• Where it stood, in our own terms, against the others on the floor.
• If your model passes and you’d like it public, a place in the congratulations.

How to put one forward

Tell us about your model and we’ll take it from there. It’s all done remotely and in writing — nothing to attend, nowhere to be.

The gate is unforgiving and it doesn’t grade on a curve — that’s the point. A weaker model isn’t turned away; it’s simply told, precisely, what the proof would not accept. The door reopens for a re-sit whenever your model improves.