Meta-Cognition — Bake-Off Console

Run a cognition bake-off

Every cognition in the stable re-attempts the SAME real, held-out failure — each with a different way of applying model intelligence (model, context, adversary, attempts, output constraint). Every proposal is committed blinded in Imagination Space before the held-out answer is revealed, then graded against reality. Only the surviving tail costs model calls, so a full bake-off stays to a couple dozen calls.

Live bake-off log

Waiting to start...