The Gauntlet
One task, two setups. Ours runs first and is sealed. You run yours, then judge.
Theme · Resume from a handoff
Six multiple-choice questions. Some rest on a false premise.
How it works
- Get a code. Early-access members get one by email.
- Run ours. Our setup answers the questions in up to 90 seconds. Its answers are sealed: signed, so they can’t change once you see the questions, and locked until you judge. The lock is on your honour: the seal proves our answers didn’t change, not that you didn’t see them.
- Run yours. Copy the prompt into the AI you already use. You can run it in the agent you built from a template. Save its reply as result.md.
- Judge. Compare blind shows both sets of answers side by side as A and B, in shuffled order, and you pick the better. Or score both against the answer key.