How it works, and where it stops.

Nothing here is hidden behind a score. These pages show what gets sent, how a verdict is reached, and the cases where it will let you down.

§Pages

One check, start to finish: a real check from our trial, step by step
Two-minute start: what to type where
Why this works: models are bad at random, each in its own way
The eight questions: exactly what gets sent
How it decides: a distance, then a test for luck
Reading a result: what each line is telling you
Where it stops: the cases it will let you down
Using OpenRouter: presets, pinning a host, rate limits
Sharing and the demo: links, paper, and the pretend gateways
How the arena works: the attacks, and how a hit is counted
How unmask works: one key, a line-up of every model on file
What spec check tests: seven promises of the chat format, one small request each
How the drift log works: each endpoint against its own first record
When a run fails: the errors you may see