Margin-Adaptive Confidence Ranking for Reliable LLM Judgement
For practitioners using LLMs as judges, this work improves the reliability of automated evaluation by ensuring better monotonicity between confidence and disagreement risk, leading to more efficient human-LLM agreement testing.
The paper addresses the problem of unreliable confidence estimates in LLM-based human agreement testing by learning a dedicated confidence estimator with margin-based ranking and generalization guarantees, achieving higher success rates in satisfying target agreement levels across multiple datasets.
Jung et al. (2025) introduce a hypothesis testing framework for guaranteeing agreement between large language models (LLMs) and human judgments, relying on the assumption that the model's estimated confidence is monotonic with respect to human-disagreement risk. In practice, however, this assumption may be violated, and the generalization behavior of the confidence estimator is not explicitly analyzed. We mitigate these issues by learning a dedicated confidence estimator instead of relying on heuristic confidence signals. Our approach leverages simulated annotator diversity and a margin-based ranking formulation to explicitly model how confidently an LLM distinguishes between human-agreement and human-disagreement cases. We further derive generalization guarantees for this estimator, revealing a margin-dependent trade-off that informs the design of an adaptive estimator training procedure. When integrated into fixed-sequence testing, the learned confidence estimator yields improved ranking accuracy and empirically strengthens the monotonic relationship between confidence and disagreement risk, leading to higher success rates in satisfying target agreement levels across multiple datasets and judge models.