(10/n) 🎯 Takeaway: Judge reliability should be treated as a first-class property of research-agent evaluation pipelines. Fine-grained protocols help, but robust judging still needs careful choices…