Test your model. Tell no one, including us.
When a scenario pool retires from active duty it goes public, and this page will let you run any model against it entirely in your browser.
How it will work: paste an API key or point at a local model, run the retired pool, and see the scores against the published corpus. Everything happens client-side. Nothing is uploaded, nothing is recorded, and your key never leaves your machine; the page will be readable source, so you can check that claim rather than trust it.
Why retired pools only: active pools stay sealed under commit-reveal so nobody can train to the test. Retirement is the reveal.
Until then, you can sit the experiment yourself with the original frozen instrument. For a ping when this opens, the Keeping List is on the Community page.