Competing AI agents argue to an auditable, scored winner — with receipts and a list of what didn't win. A decision, not an answer.