Custom environments
Environments authored against a specific weakness in a specific model, chosen from measurement rather than guesswork.
The leaderboard makes the scope conversation short. A model's failures are already broken out by vulnerability class and by failure mode, so instead of arguing about what to build, we start from where it measurably fails and produce more of that.
The generator produces candidates and the gates decide which survive. A task only exists if the app boots and passes its own tests, an exploit against it demonstrably works, a reference patch kills that exploit, and the patched app still passes. Nothing enters on a reviewer's opinion, which is why volume does not degrade quality.
Scope is usually a class, a framework, or a difficulty band: more of what your model fails, in the shape your stack actually uses.
What you get
- Task authoring targeted at named classes, frameworks or difficulty bands
- The same four executable gates, so custom tasks are graded identically to the public set
- A before-and-after measurement on your model, not just a delivery of files
- Exclusivity terms, where the tasks are not resold or published
Why this is not public
- Off-the-shelf coverage is whatever the generator happened to emit, not what your model is weak at
The method is fully inspectable either way. Browse every public task, exploit and reference patch before you talk to us.