The PReMISE framework improves the evaluation of open-ended responses by LLM judges through better-defined rubrics, enhancing measurement accuracy and reducing exploitative scoring. This advancement addresses key issues in rubric reliability and preference alignment, potentially leading to more trustworthy AI assessments in various applications.