Will the Judge Flip? Predicting Position-Sensitive LLM Judgments from Residual Stream Activations
cs.LG, cs.AI
Submitted: 2026-10-05
Updated: 2026-10-05
Comments: It was submitted to neurips workshop (JUDGE workshop) and received a review score 6.5 combined with one strong accept and one above threshold accept
Code: https://github.com/hashmathshaik/will-the-judge-flip
License: http://creativecommons.org/licenses/by/4.0/
The gist: The order in which candidate responses are presented can change an LLM judge's verdict.
Abstract
The order in which candidate responses are presented can change an LLM judge's verdict. Detecting such a position flip ordinarily requires judging each pair in both orders, which doubles the number of judgments. We investigate whether residual stream activations recorded immediately before the initial verdict can predict a flip. We use nested grouped cross-validation to evaluate regularized linear probes on 534 JudgeBench pairs for three Qwen3 judges and Llama-3.1-8B. The linear probes achieve AUROCs of .621-.850 and outperform a combined baseline that uses verbalized confidence, verdict-label logits, response lengths, and the judge's initial choice by .062-.113 AUROC. Linear probes trained on JudgeBench and then frozen achieve AUROCs of .685-.853 on 1,802 MT-Bench comparisons without MT-Bench fitting or recalibration. These results show that pre-verdict activations support prediction of susceptibility to candidate order and outperform the non-activation predictors evaluated here.
Sources
- The Llama 3 Herd of Models
- Beyond the Surface: Enhancing LLM-as-a-Judge Alignment with Human via Internal Representations
- Opening LLM Judges: Recovering Preference Signals Beyond the Final Verdict
- Calibrating LLM Judges: Linear Probes for Fast and Reliable Uncertainty Estimation
- Qwen3 Technical Report
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks