Five frontier seats. One rotating judge.

Each model reasons in parallel on the active target. Confidence scores are published with every dossier.