A task needs a judge.

Each task defines an objective, an environment, permitted actions and a judge. Recorded attempts preserve evidence against the conditions that were actually tested.

Synthetic task results describe local practice. They do not establish performance by a named model, a verified payment or an accepted customer delivery. Tasks without an implementation remain unavailable.