FairJudge : An Adaptive, Debiased, and Consistent LLM-as-a-Judge

Bo Yang, Lanfei Feng, Yunkui Chen, Xiao Xu, Yu Zhang, Shijian Li
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:144545-144563, 2026.

Abstract

Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to task and domain-specific evaluation criteria, systematic biases driven by non-semantic cues such as position, length, format, and model provenance, and evaluation inconsistency that leads to contradictory judgments across different evaluation modes (e.g., pointwise versus pairwise). To address these issues, we propose FairJudge, an adaptive, debiased, and consistent LLM-as-a-Judge. Unlike prior approaches that treat the judge as a static evaluator, FairJudge models judging behavior itself as a learnable and regularized policy. From a data-centric perspective, we construct a high-information-density judging dataset that explicitly injects supervision signals aligned with evaluation behavior. Building on this dataset, we adopt a curriculum-style SFT-DPO-GRPO training paradigm that progressively aligns rubric adherence, bias mitigation, and cross-mode consistency, while avoiding catastrophic forgetting. Experimental results on multiple internal and public benchmarks show that FairJudge improves agreement and F1 across several evaluation settings, reduces selected non-semantic biases, and achieves competitive or stronger performance than larger general-purpose LLMs on judge-oriented tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-yang26n, title = {{F}air{J}udge : An Adaptive, Debiased, and Consistent {LLM}-as-a-Judge}, author = {Yang, Bo and Feng, Lanfei and Chen, Yunkui and Xu, Xiao and Zhang, Yu and Li, Shijian}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {144545--144563}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/yang26n/yang26n.pdf}, url = {https://proceedings.mlr.press/v306/yang26n.html}, abstract = {Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to task and domain-specific evaluation criteria, systematic biases driven by non-semantic cues such as position, length, format, and model provenance, and evaluation inconsistency that leads to contradictory judgments across different evaluation modes (e.g., pointwise versus pairwise). To address these issues, we propose FairJudge, an adaptive, debiased, and consistent LLM-as-a-Judge. Unlike prior approaches that treat the judge as a static evaluator, FairJudge models judging behavior itself as a learnable and regularized policy. From a data-centric perspective, we construct a high-information-density judging dataset that explicitly injects supervision signals aligned with evaluation behavior. Building on this dataset, we adopt a curriculum-style SFT-DPO-GRPO training paradigm that progressively aligns rubric adherence, bias mitigation, and cross-mode consistency, while avoiding catastrophic forgetting. Experimental results on multiple internal and public benchmarks show that FairJudge improves agreement and F1 across several evaluation settings, reduces selected non-semantic biases, and achieves competitive or stronger performance than larger general-purpose LLMs on judge-oriented tasks.} }
Endnote
%0 Conference Paper %T FairJudge : An Adaptive, Debiased, and Consistent LLM-as-a-Judge %A Bo Yang %A Lanfei Feng %A Yunkui Chen %A Xiao Xu %A Yu Zhang %A Shijian Li %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-yang26n %I PMLR %P 144545--144563 %U https://proceedings.mlr.press/v306/yang26n.html %V 306 %X Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to task and domain-specific evaluation criteria, systematic biases driven by non-semantic cues such as position, length, format, and model provenance, and evaluation inconsistency that leads to contradictory judgments across different evaluation modes (e.g., pointwise versus pairwise). To address these issues, we propose FairJudge, an adaptive, debiased, and consistent LLM-as-a-Judge. Unlike prior approaches that treat the judge as a static evaluator, FairJudge models judging behavior itself as a learnable and regularized policy. From a data-centric perspective, we construct a high-information-density judging dataset that explicitly injects supervision signals aligned with evaluation behavior. Building on this dataset, we adopt a curriculum-style SFT-DPO-GRPO training paradigm that progressively aligns rubric adherence, bias mitigation, and cross-mode consistency, while avoiding catastrophic forgetting. Experimental results on multiple internal and public benchmarks show that FairJudge improves agreement and F1 across several evaluation settings, reduces selected non-semantic biases, and achieves competitive or stronger performance than larger general-purpose LLMs on judge-oriented tasks.
APA
Yang, B., Feng, L., Chen, Y., Xu, X., Zhang, Y. & Li, S.. (2026). FairJudge : An Adaptive, Debiased, and Consistent LLM-as-a-Judge. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:144545-144563 Available from https://proceedings.mlr.press/v306/yang26n.html.

Related Material