A wrong verdict does not mean the judge lacks the information, and a simple diagnostic shows when it is worth recovering, and on rubric tasks that score one answer at a time, reading the internals gives no advantage.
Abstract
LLM judges are widely used to evaluate model outputs, but their verdicts can be unreliable: a judge may favor the worse answer for its position, length, or other surface features. When a judge is wrong, is the information needed to judge correctly absent from the model, or present in its internal representations but not reflected in the output? We study this across 64 open-weight evaluators and 14 datasets, including causal interventions on 41 judges (editing activations mid-run to see whether the verdict changes). On LLMBar, built so the superficially better answer is the worse one, the verdicts of 50 judges agree with human labels only 0.456 of the time, even after averaging both answer orders. Yet a small probe on the same judges'activations, with no weight updates, reaches 0.846, and 0.686 once surface features such as length and position are residualized out (0.507 with shuffled labels). The gap holds across eight benchmarks and model families, but is not universal: a score of how well surface features alone predict the human label, computed before any probe is trained, predicts the size of the gain (Spearman rho = 0.90). On rubric tasks that score one answer at a time, leaving no surface cue to exploit, reading the internals gives no advantage. The interventions also show that editing activations mid-network already changes the verdict, before it can be read off directly, and locate the pathways carrying position and length bias. At the same label budget, the recovered signal lets a judge flag cases where it is likely wrong and yields better labels for preference learning. A wrong verdict, then, does not mean the judge lacks the information, and a simple diagnostic shows when it is worth recovering.
Reading an LLM judge's verdict from the logits of its first generated token is cheap, requires no generation, and is exactly what constrained decoding and likelihood-scoring evaluation harnesses produce, but it is shown that this readout distorts position bias in one direction.
Gnaneswar Villuri, Hashmath Shaik, Alex Doboli· 0 citations
JudgeProfile is introduced, a framework that dissects LLM evaluation into perception (how a judge compares two responses across specific attributes like clarity, correctness, and detail) and prioritization (how much each attribute influences the final choice) and finds a hidden consensus in perception.
Qiong-Qiong Cao, Kang-Ni Liu, Xuan Kan et al.· 0 citations
Results show that pre-verdict activations support prediction of susceptibility to candidate order and outperform the non-activation predictors evaluated here.
Hashmath Shaik, Gnaneswar Villuri, Alex Doboli· 0 citations
LLM-as-a-Judge is the standard paradigm for NLP evaluation, yet its systemic reliability remains poorly understood despite being widely treated as a deterministic ground truth. We present a comprehensive reliability audit, stresstesting six frontier models across four benchmarks, five prompt formats, two presentation o...
This work proposes VERA, a VErdict-conditioned Reliability Axis that estimates reliability from hidden activations by distinguishing correct from incorrect judgments within each predicted-verdict group within each predicted-verdict group.
Qiu-Shui Xu, S. Razak, Tao Yuan et al.· 0 citations
An open contrastive decision model is near chance as a judge on the hard public benchmarks: Contrastive-LM/CLM-v0.1-8B scores between 0.351 (best- of-four, chance 0.250) and 0.593 (pairwise, chance 0.500), is statistically indistinguishable from coin flipping on RM-Bench and JudgeBench, and answers every HaluEval item...