Class TournamentRanking

The error a tournament throws when every candidate failed. When every candidate hit the same Predictive Studio refusal, the first one is rethrown so its type and code survive; otherwise one error lists each distinct message.

A candidate's leaderboard score: its trust verdict's headline metric, so the leaderboard and the headline can never disagree for the same model. Null when the model was not measured.

Leaderboard rows for ranked candidates; the first row is the winner only when it was measured.

Best-first: measured candidates by score (higher is better), unmeasured candidates last in plan order.