Try llm-consortium with --judging-method rank