A campaign sends goals through one or more attacks to a target, and judges decide which replies were jailbreaks. Click a card for the short version; the pages below go deeper.
How a run works: one goal followed from the dataset to the recorded verdict.
Attacks and roles: static versus adaptive attacks, and the models they drive.
Judges: how votes become a verdict, and why a judge may abstain.
Code map: where each of these lives in the source.