Tag
SESSE is a training-free framework that decomposes holistic LLM-as-judge evaluations into structured sub-questions, enabling better interpretability and diagnosis of label ambiguity while achieving competitive performance with fine-tuned models.