The Coin Flip Judge? Reliability and Bias in LLM-as-a-Judge Evaluation

• URL: http://arxiv.org/abs/2606.13685v1
• Date: Thu, 23 Apr 2026 18:19:10 GMT
• Title: The Coin Flip Judge? Reliability and Bias in LLM-as-a-Judge Evaluation
• Abstract summary: We study identical evaluations on 29 tasks spanning 10 categories using two OpenAI judge models.<n>Across judges, pairwise preferences flip on average 13.6% of the time, with 28% of questions exceeding a 20% flip rate and one question reaching 56%.<n>Cross-judge agreement is only 76%, semantically equivalent prompt templates change majority outcomes in 25% of tested cases, and deterministic decoding reduces but does not eliminate inconsistency.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: LLM-as-a-Judge is now widely used to rank model outputs, train reward models, and populate public leaderboards, but its run-to-run reliability remains under-characterized. We study repeated identical evaluations on 29 tasks spanning 10 categories using two OpenAI judge models (GPT-4o-mini and GPT-4.1-mini), with 50 pairwise trials and 50 pointwise trials per question, supplemented by temperature and prompt-sensitivity ablations. Across judges, pairwise preferences flip on average 13.6% of the time, with 28% of questions exceeding a 20% flip rate and one question reaching 56%. GPT-4o-mini also exhibits a significant first-position bias (72% A-majority, p = 0.024). At the same time, mean pointwise score gaps are small (0.19--0.36 on a 10-point scale) and not statistically significant in aggregate, producing a pairwise--pointwise gap: judges frequently choose a winner even when their own scalar scores provide little evidence of a meaningful quality difference. Beyond within-judge instability, cross-judge agreement is only 76% ($κ= 0.51$), semantically equivalent prompt templates change majority outcomes in 25% of tested cases, and deterministic decoding reduces but does not eliminate inconsistency. A reliability curve analysis shows that, in our dataset, 11 repeated trials are needed for a majority vote to recover the 50-trial reference verdict with 95% probability on average, rising to 15 for high-variance questions. These findings suggest that single-trial LLM judging is often too noisy for high-stakes evaluation, and that multi-trial aggregation, position randomization, and explicit uncertainty reporting should be standard practice. Because both judges are from a single provider, cross-provider replication remains an important next step.

Related papers

• NovGauge: A Fine-Grained Benchmark for Diagnosing LLMs' Capability in Paper Novelty Assessment [54.4265627247104]
  Large language models (LLMs) are increasingly used in peer review at major AI conferences.<n>Existing benchmarks assess novelty as a single holistic score.<n>We present NovGauge, a human-anchored benchmark for fine-grained novelty assessment diagnosis.
  arXiv  Detail & Related papers  (2026-09-10T08:34:56Z)
• Can We Trust LLM Judges: A Study of Capability-Dependent Biases and Multi-Judge Ensemble for Bias Calibration [0.5294698352039444]
  We study absolute scoring tasks, which mirror more realistic use cases.<n>We show that a model's task accuracy strongly predicts its judging accuracy.<n>More capable examinee models consistently receive more lenient judgments from all judges.
  arXiv  Detail & Related papers  (2026-09-10T00:25:39Z)
• A Judge Should Know What Changed:Construct Validity for LLM-as-a-Judge Evaluation [14.867869624585886]
  We formalize construct validity for an evaluator as a two-dimensional profile.<n>We show that S and R are independent and that no scalar summary preserves all relevant comparisons.<n>These results show that high judge agreement can coexist with weak sensitivity to changes in the construct being evaluated.
  arXiv  Detail & Related papers  (2026-08-25T11:32:54Z)
• Reliability without Validity: A Systematic, Large-Scale Evaluation of LLM-as-a-Judge Models Across Agreement, Consistency, and Bias [0.0]
  Judge validation in practice relies on exact-match agreement, a metric that does not correct for chance and systematically overstates discriminative ability.<n>We present the largest systematic evaluation of LLM-as-a-Judge to date: 21 judges from nine providers across MT-Bench, JudgeBench, and RewardBench, evaluated under three protocols (agreement, consistency, bias audit) over 118 runs and approximately 541,000 individual judgments.
  arXiv  Detail & Related papers  (2026-06-17T19:37:13Z)
• Step-wise Rubric Rewards for LLM Reasoning [72.17879367869503]
  Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in large language models.<n>We find 18.2% of steps in correct responses are wrong yet positively rewarded.<n>Across six mathematical reasoning benchmarks, SRaR improves average accuracy over RaR by 3.57 points.
  arXiv  Detail & Related papers  (2026-05-17T07:08:14Z)
• Calibrate, Don't Curate: Label-Efficient Estimation from Noisy LLM Judges [3.974554103907186]
  We show that in multi-judge evaluation with labeled calibration data, do not discard weak judges by accuracy alone.<n>We explain this reversal with oracle analyses showing that the optimal calibrated risk under proper scoring rules cannot increase when additional judge signals are made available.
  arXiv  Detail & Related papers  (2026-05-10T18:49:58Z)
• When LLM Judge Scores Look Good but Best-of-N Decisions Fail [0.29465623430708904]
  Large language models are often used as judges to score candidate responses, then validated with a single global metric.<n>This can be misleading when the real deployment task is best-of-n selection within a prompt.<n>For judge-based selection, the relevant audit should report within-prompt signal, tie rates, and recovery/top-1 accuracy, not global agreement alone.
  arXiv  Detail & Related papers  (2026-03-12T23:40:03Z)
• Re-Rankers as Relevance Judges [65.37611299805856]
  We reproduce re-rankers in a re-ranker-as-relevance-judge setup.<n>We perform experiments on TREC-DL 2019 to 2023 with 8 re-rankers from 3 families, ranging from 220M to 32B, and analyse the evaluation bias exhibited by re-ranker-based judges.
  arXiv  Detail & Related papers  (2026-01-08T00:02:59Z)
• Reference-Free Rating of LLM Responses via Latent Information [53.463883683503106]
  We study the common practice of asking a judge model to assign Likert-scale scores to free-text responses.<n>We then propose and evaluate Latent Judges, which derive scalar ratings from internal model signals.<n>Across a broad suite of pairwise and single-rating benchmarks, latent methods match or surpass standard prompting.
  arXiv  Detail & Related papers  (2025-09-29T12:15:52Z)
• TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them [58.04324690859212]
  Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks.<n>We identify two fundamental types of inconsistencies: Score-Comparison Inconsistency and Pairwise Transitivity Inconsistency.<n>We propose TrustJudge, a probabilistic framework that addresses these limitations through two key innovations.
  arXiv  Detail & Related papers  (2025-09-25T13:04:29Z)
• ObjexMT: Objective Extraction and Metacognitive Calibration for LLM-as-a-Judge under Multi-Turn Jailbreaks [12.396822247035578]
  We present exMT, a benchmark for objective extraction and metacognition.<n>Given a multi-turn transcript, a model must output a one-sentence base objective and a self-reported confidence.<n> Accuracy is scored by similarity to gold objectives, then thresholded once on 300 calibration items.
  arXiv  Detail & Related papers  (2025-08-23T03:32:04Z)
• Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios [49.53589774730807]
  Multimodal large language models (MLLMs) have recently achieved state-of-the-art performance on tasks ranging from visual question answering to video understanding.<n>We reveal a response uncertainty phenomenon: twelve state-of-the-art open-source MLLMs overturn a previously correct answer in 65% of cases after receiving a single deceptive cue.
  arXiv  Detail & Related papers  (2024-11-05T01:11:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.