mlsanityData to 7 Oct 2026

Judging the Judges: A Systematic Evaluation of Bias Mitigation Strategies in LLM-as-a-Judge Pipelines

arXiv 2604.23178 ↗First posted 25 Apr 2026cs.AIcited by 7

Not on the map. Why?

Abstract (arXiv)

LLM-as-a-Judge has become the dominant paradigm for evaluating language model outputs, yet LLM judges exhibit systematic biases that compromise evaluation reliability. We present a comprehensive empirical study comparing nine debiasing strategies across five judge models from four provider families (Google, Anthropic, OpenAI, Meta), three benchmarks (MT-Bench n=400, LLMBar n=200, custom n=375), and four bias types. Our headline practical finding is that a mid-tier model with the right debiasing can outperform frontier judges at a fraction of the cost: Gemini 2.5 Flash with the Combined Budget strategy reaches the highest agreement of any configuration we tested (71.0%, kappa=0.549) at ~0.001 per evaluation, about 15x cheaper than the best frontier setup (Claude Sonnet 4, 69.5%, ~0.015). Other key findings: (1) Style bias is the dominant bias (0.10-0.76 across models, favoring markdown over plain prose), far exceeding position bias (<=0.04), yet is rarely studied. (2) Verbosity bias is heterogeneous when measured length-aware: Pro, Flash, and Llama prefer longer answers (+0.24 to +0.44), Claude prefers concise (-0.12), and GPT-4o is neutral (-0.04); on truncation controls all models correctly prefer the complete response (0.88-1.00 accuracy). (3) Debiasing helps multiple models: Claude S8 (+11.5 pp), Flash S8 (+7.5 pp), and Claude S5 (+7.3 pp) survive Holm-Bonferroni correction, with Flash S1 (+4.7 pp) and Llama S8 (+4.5 pp) also significant. We release our evaluation framework, the 375-pair controlled dataset, and per-instance cached results for all nine strategies.

Key facts. Cited by 7 papers in the map (data through 7 Oct 2026). Builds on 35 of its 43 references that are in the map. First posted on arXiv 25 Apr 2026.

Builds on

35 of 43 references.

References · 35

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2024The Llama 3 Herd of Modelscited by 14,192
  3. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  5. 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,205
  6. 2024A Survey on LLM-as-a-Judgecited by 1,384
  7. 2024Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preferencecited by 1,151
  8. 2023Large Language Models are not Fair Evaluatorscited by 1,050
  9. 2024Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluatorscited by 865
  10. 2024LLM Evaluators Recognize and Favor Their Own Generationscited by 651
  11. 2023Bias and Fairness in Large Language Models: A Surveycited by 626
  12. 2024From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipelinecited by 484
23 more
  1. 2024RewardBench: Evaluating Reward Models for Language Modelingcited by 480
  2. 2024Prometheus 2: An Open Source Language Model Specialized in Evaluating Other Language Modelscited by 431
  3. 2024Humans or LLMs as the Judge? A Study on Judgement Biasescited by 314
  4. 2023PandaLM: An Automatic Evaluation Benchmark for LLM Instruction Tuning Optimizationcited by 311
  5. 2024Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Modelscited by 301
  6. 2024JudgeBench: A Benchmark for Evaluating LLM-based Judgescited by 294
  7. 2024Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judgecited by 249
  8. 2023FLASK: Fine-grained Language Model Evaluation based on Alignment Skill Setscited by 198
  9. 2023Generative Judge for Evaluating Alignmentcited by 163
  10. 2023Benchmarking Cognitive Biases in Large Language Models as Evaluatorscited by 147
  11. 2024Aligning with Human Judgement: The Role of Pairwise Preference in Large Language Model Evaluatorscited by 146
  12. 2024Large Language Models are Inconsistent and Biased Evaluatorscited by 146
  13. 2024Who Validates the Validators? Aligning LLM-Assisted Evaluation of LLM Outputs with Human Preferencescited by 142
  14. 2024OffsetBias: Leveraging Debiased Data for Tuning Evaluatorscited by 94
  15. 2023Wider and Deeper LLM Networks are Fairer LLM Evaluatorscited by 91
  16. 2023Verbosity Bias in Preference Labeling by Large Language Modelscited by 89
  17. 2023Split and Merge: Aligning Position Biases in LLM-based Evaluatorscited by 71
  18. 2023TIGERScore: Towards Building Explainable Metric for All Text Generation Taskscited by 69
  19. 2023Style Over Substance: Evaluation Biases for Large Language Modelscited by 68
  20. 2025Any Large Language Model Can Be a Reliable Judge: Debiasing with a Reasoning-based Bias Detectorcited by 14
  21. 2025Judging with Many Minds: Do More Perspectives Mean Less Prejudice? On Bias Amplifications and Resistance in Multi-Agent Based LLM-as-Judgecited by 10
  22. 2026CARE: Confounder-Aware Aggregation for Reliable LLM Evaluationcited by 10
  23. 2026Quantifying and Mitigating Self-Preference Bias of LLM Judgescited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 7.

7 papers.

  1. 2026PaperJury: Due-Process Review for Bounded LaTeX Revisioncited by 1
  2. 2026LivingArena: Do LLMs Know What Other LLMs Don't? Peer-Probing as Scalable Evaluationcited by 1
  3. 2026Who Judges Matters: Measuring Family-Conditioned Preference in LLM-as-Judge Panelscited by 1
  4. 2026Does Capability Transfer to Subjective Behavior -- and Would Our Instruments Tell Us? A Self-Evolving, Trust-by-Construction Evaluation Paradigmcited by 0
  5. 2026Are We Evaluating Knowledge or Phrasing? Mitigating MCQA Sensitivity with ParaEvalcited by 0
  6. 2026SPLIT: Cross-Lingual Empathy and Cultural Grounding in English and Ukrainian LLM Responsescited by 0
  7. 2026Structure for Reading, Prose for Writing: Asymmetric Structural Conditioning in Multi-Agent Document Authoringcited by 0