DebatableHow judging works
HELP TEST THE COUNCIL

Break the judge.

Can you make a judge invent a concession, follow an instruction hidden in speech, or change its verdict when only the names change?

Submit an invented case and explain the failure. We test it privately, preserve reproducible failures as regression cases, and award a discovery badge after expert human review. These cases never affect normal round ratings. A changed AI answer alone is not a verified failure.

Invented conversation. Add each person’s words in the order they were spoken.

Up to five submissions per account each day. Submitting does not run a paid model call. Keep the returned submission ID if you want to withdraw public badge consent.

Withdraw public badge consent

Verified discoveries

Loading verified discoveries…