Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models

• URL: http://arxiv.org/abs/2602.04649v1
• Date: Wed, 04 Feb 2026 15:24:52 GMT
• Title: Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models
• Authors: Binghai Wang, Yantao Liu, Yuxuan Liu, Tianyi Tang, Shenzhi Wang, Chang Gao, Chujie Zheng, Yichang Zhang, Le Yu, Shixuan Liu, Tao Gui, Qi Zhang, Xuanjing Huang, Bowen Yu, Fei Huang, Junyang Lin,
• Abstract summary: We introduce Rationale Consistency, a fine-grained metric that quantifies the alignment between the model's reasoning process and human judgment.<n>Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models.<n>We introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training.
• Score: 108.26461635308796
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Generative Reward Models (GenRMs) and LLM-as-a-Judge exhibit deceptive alignment by producing correct judgments for incorrect reasons, as they are trained and evaluated to prioritize Outcome Accuracy, which undermines their ability to generalize during RLHF. We introduce Rationale Consistency, a fine-grained metric that quantifies the alignment between the model's reasoning process and human judgment. Our evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training. Our training method achieves state-of-the-art performance on RM-Bench (87.1%) and JudgeBench (82%), surpassing outcome-only baselines by an average of 5%. Using RM during RLHF, our method effectively improves performance as demonstrated on Arena Hard v2, notably yielding a 7% improvement in creative writing tasks. Further analysis confirms that our method escapes the deceptive alignment trap, effectively reversing the decline in rationale consistency observed in outcome-only training.

Related papers

• RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models [5.733004743054914]
  Large Reasoning Models (LRMs) exhibit strong performance, yet often produce rationales that sound plausible but fail to reflect their true decision process.<n>We introduce a formal framework for reasoning faithfulness, defined by two testable conditions.<n>We present RFEval, a benchmark of 7,186 instances that probes faithfulness via controlled, output-level counterfactual interventions.
  arXiv  Detail & Related papers  (2026-02-19T03:49:37Z)
• PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering [71.15346406323827]
  We introduce PRIME, a benchmark for evaluating verifiers on Process-Outcome Alignment verification.<n>We find that current verifiers frequently fail to detect derivation flaws.<n>We propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME.
  arXiv  Detail & Related papers  (2026-02-12T04:45:01Z)
• R-Align: Enhancing Generative Reward Models through Rationale-Centric Meta-Judging [69.96389360650072]
  We show that reasoning fidelity is highly predictive of downstream RLHF outcomes, beyond standard label accuracy.<n>We propose Rationale-Centric Alignment, R-Align, which augments training with gold judgments and explicitly supervises rationale alignment.
  arXiv  Detail & Related papers  (2026-02-06T15:17:11Z)
• When Small Models Are Right for Wrong Reasons: Process Verification for Trustworthy Agents [0.0]
  We reveal a critical reliability crisis: 50-69% of correct answers from small language models contain fundamentally flawed reasoning.<n>We introduce the Reasoning Integrity Score (RIS), a process-based metric validated with substantial inter-rater agreement.<n>We show RAG succeeds by grounding calculations in external evidence, reducing errors by 7.6%, while meta-cognition amplifies confusion without sufficient model capacity.
  arXiv  Detail & Related papers  (2026-01-01T23:54:15Z)
• Mitigating LLM Hallucination via Behaviorally Calibrated Reinforcement Learning [32.32593439144886]
  Behavior-calibrated reinforcement learning allows smaller models to surpass frontier models in uncertainty quantification.<n>Our model's log-scale Accuracy-to-Hallucination Ratio gain (0.806) exceeds GPT-5's (0.207) in a challenging in-domain evaluation.
  arXiv  Detail & Related papers  (2025-12-22T22:51:48Z)
• Trustworthy Reasoning: Evaluating and Enhancing Factual Accuracy in LLM Intermediate Thought Processes [16.451488374845407]
  We present a novel framework addressing a critical vulnerability in Large Language Models (LLMs)<n>This phenomenon poses substantial risks in high-stakes domains including healthcare, legal analysis, and scientific research.
  arXiv  Detail & Related papers  (2025-07-25T10:34:51Z)
• Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty [59.97939500426759]
  This paper describes RLCR, an approach to training reasoning models that jointly improves accuracy and confidence estimation.<n>We show that across diverse datasets, RLCR substantially improves calibration with no loss in accuracy.<n>We also demonstrate that verbalized confidence can be leveraged at test time to improve accuracy and calibration.
  arXiv  Detail & Related papers  (2025-07-22T17:56:01Z)
• Bridging Internal Probability and Self-Consistency for Effective and Efficient LLM Reasoning [53.25336975467293]
  We present the first theoretical error decomposition analysis of methods such as perplexity and self-consistency.<n>Our analysis reveals a fundamental trade-off: perplexity methods suffer from substantial model error due to the absence of a proper consistency function.<n>We propose Reasoning-Pruning Perplexity Consistency (RPC), which integrates perplexity with self-consistency, and Reasoning Pruning, which eliminates low-probability reasoning paths.
  arXiv  Detail & Related papers  (2025-02-01T18:09:49Z)
• Revisiting and Advancing Adversarial Training Through A Simple Baseline [7.226961695849204]
  We introduce a simple baseline approach, termed SimpleAT, that performs competitively with recent methods and mitigates robust overfitting. We conduct extensive experiments on CIFAR-10/100 and Tiny-ImageNet, which validate the robustness of SimpleAT against state-of-the-art adversarial attackers. Our results also reveal the connections between SimpleAT and many advanced state-of-the-art adversarial defense methods.
  arXiv  Detail & Related papers  (2023-06-13T08:12:52Z)
• Leveraging Unlabeled Data to Predict Out-of-Distribution Performance [63.740181251997306]
  Real-world machine learning deployments are characterized by mismatches between the source (training) and target (test) distributions. In this work, we investigate methods for predicting the target domain accuracy using only labeled source data and unlabeled target data. We propose Average Thresholded Confidence (ATC), a practical method that learns a threshold on the model's confidence, predicting accuracy as the fraction of unlabeled examples.
  arXiv  Detail & Related papers  (2022-01-11T23:01:12Z)
• Adversarial Robustness on In- and Out-Distribution Improves Explainability [109.68938066821246]
  RATIO is a training procedure for robustness via Adversarial Training on In- and Out-distribution. RATIO achieves state-of-the-art $l$-adrial on CIFAR10 and maintains better clean accuracy.
  arXiv  Detail & Related papers  (2020-03-20T18:57:52Z)

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.