Beyond the Answer Key: Robustness Evaluation of Large Language Models for Step-Level Mathematical Verification

• URL: http://arxiv.org/abs/2608.28725v1
• Date: Fri, 28 Aug 2026 16:17:11 GMT
• Title: Beyond the Answer Key: Robustness Evaluation of Large Language Models for Step-Level Mathematical Verification
• Abstract summary: Large language models (LLMs) are increasingly used as graders, verifiers, and process auditors.<n>Most mathematical evaluations still emphasize final-answer accuracy.<n>We introduce a controlled linear-equation benchmark for evaluating LLMs in the evaluator role.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models (LLMs) are increasingly used as graders, verifiers, and process auditors, but most mathematical evaluations still emphasize final-answer accuracy. This can obscure whether a model can verify a non-canonical but valid solution trace. We introduce a controlled linear-equation benchmark for evaluating LLMs in the evaluator role. Each instance asks the model to judge final-answer correctness, step-level trace correctness, and the first incorrect step. Our evaluation of state-of-the-art open LLMs reveals a significant robustness gap: models that accurately evaluate canonical solutions often fail when presented with perturbed but logically equivalent variants. Across GPT-OSS 20B, Qwen3-14B, and Phi-4-Reasoning, base models perform well on canonical traces but degrade substantially on perturbed traces, especially for error localization. On valid perturbed traces, base-model false-rejection rates reach 75.6-85.3%, showing strong sensitivity to canonical solution form. Supervised fine-tuning, distillation, and test-time compute improve robustness in some settings, but gains are model dependent and can trade off against canonical performance. The results show that reliable process-level verification remains challenging, and evaluator robustness should be measured separately from solver accuracy, even in a simple algebraic domain with exact ground truth.

Related papers

• Correct Answers from Sound Reasoning: Verifiable Process Supervision for Language Models [94.68358825189738]
  We propose a post-training framework for verifiable domains that jointly optimize prediction accuracy and reasoning quality.<n>We evaluate VPS on chess, a controlled testbed where reasoning steps can be deterministically verified against engine signals.<n>VPS preserves accuracy while significantly improving reasoning quality, reducing win-rate error by up to 30% and restoring consistency to near saturation.
  arXiv  Detail & Related papers  (2026-04-03T15:19:46Z)
• Variational Rectification Inference for Learning with Noisy Labels [74.85528327499662]
  We propose variational rectification inference (VRI) to formulate the adaptive rectification for loss functions.<n>VRI is constructed as a hierarchical Bayes by treating the rectifying vector as a latent variable.<n>By introducing the variational term in VRI, the conditional posterior is estimated accurately and avoids collapsing to a Dirac delta function.
  arXiv  Detail & Related papers  (2026-03-18T01:25:08Z)
• On Calibration of Large Language Models: From Response To Capability [66.59139960234326]
  Large language models (LLMs) are widely deployed as general-purpose problem solvers.<n>We introduce capability calibration, which targets the model's expected accuracy on a query.<n>Our results demonstrate that capability-calibrated confidence improves pass@$k$ prediction and inference budget allocation.
  arXiv  Detail & Related papers  (2026-02-14T01:07:45Z)
• When LLMs get significantly worse: A statistical approach to detect model degradations [33.63321816712603]
  Minimizing the inference cost and latency of foundation models has become a crucial area of research.<n>We propose a statistically sound hypothesis testing framework based on McNemar's test allowing to efficiently detect model degradations.<n>We find that with our tests even empirical accuracy degradations of 0.3% can be confidently attributed to actual degradations rather than noise.
  arXiv  Detail & Related papers  (2026-02-09T10:45:13Z)
• ORGEval: Graph-Theoretic Evaluation of LLMs in Optimization Modeling [18.8099769877788]
  ORGEval is a graph-theoretic evaluation framework for assessing Large Language Models' capabilities in formulating linear and mixed-integer linear programs.<n>We show that ORGEval can successfully detect model equivalence and produce 100% consistent results across random parameter configurations.<n>Our results reveal that although optimization modeling remains challenging for all LLMs, DeepSeek-V3 and Claude-Opus-4 achieve the highest accuracies under direct prompting.
  arXiv  Detail & Related papers  (2025-10-31T16:35:52Z)
• CompassVerifier: A Unified and Robust Verifier for LLMs Evaluation and Outcome Reward [50.97588334916863]
  We develop CompassVerifier, an accurate and robust lightweight verifier model for evaluation and outcome reward.<n>It demonstrates multi-domain competency spanning math, knowledge, and diverse reasoning tasks, with the capability to process various answer types.<n>We introduce VerifierBench benchmark comprising model outputs collected from multiple data sources, augmented through manual analysis of metaerror patterns to enhance CompassVerifier.
  arXiv  Detail & Related papers  (2025-08-05T17:55:24Z)
• Examining False Positives under Inference Scaling for Mathematical Reasoning [83.97128486951999]
  We systematically examine the prevalence of false positive solutions in mathematical problem solving for language models.<n>Our experimental results reveal that: (1) false positive solutions persist across different models, datasets, and decoding methods, (2) sampling-based inference time scaling methods do not alleviate the problem, and (3) the pass@N evaluation metric is more susceptible to false positives.
  arXiv  Detail & Related papers  (2025-02-10T07:49:35Z)
• A Probabilistic Perspective on Unlearning and Alignment for Large Language Models [48.96686419141881]
  We introduce the first formal probabilistic evaluation framework for Large Language Models (LLMs)<n> Namely, we propose novel metrics with high probability guarantees concerning the output distribution of a model.<n>Our metrics are application-independent and allow practitioners to make more reliable estimates about model capabilities before deployment.
  arXiv  Detail & Related papers  (2024-10-04T15:44:23Z)
• Investigating the Impact of Hard Samples on Accuracy Reveals In-class Data Imbalance [4.291589126905706]
  In the AutoML domain, test accuracy is heralded as the quintessential metric for evaluating model efficacy. However, the reliability of test accuracy as the primary performance metric has been called into question. The distribution of hard samples between training and test sets affects the difficulty levels of those sets. We propose a benchmarking procedure for comparing hard sample identification methods.
  arXiv  Detail & Related papers  (2024-09-22T11:38:14Z)
• Evaluating Mathematical Reasoning Beyond Accuracy [50.09931172314218]
  We introduce ReasonEval, a new methodology for evaluating the quality of reasoning steps.<n>We show that ReasonEval consistently outperforms baseline methods in the meta-evaluation datasets.<n>We observe that ReasonEval can play a significant role in data selection.
  arXiv  Detail & Related papers  (2024-04-08T17:18:04Z)
• Weak Supervision Performance Evaluation via Partial Identification [46.73061437177238]
  Programmatic Weak Supervision (PWS) enables supervised model training without direct access to ground truth labels. We present a novel method to address this challenge by framing model evaluation as a partial identification problem. Our approach derives reliable bounds on key metrics without requiring labeled data, overcoming core limitations in current weak supervision evaluation techniques.
  arXiv  Detail & Related papers  (2023-12-07T07:15:11Z)
• Classifier Data Quality: A Geometric Complexity Based Method for Automated Baseline And Insights Generation [4.722075132982135]
  A major challenge is to determine when the level of incorrectness, e.g., model accuracy or F1 score for classifiers, is acceptable. We have developed complexity measures, which quantify how difficult given observations are to assign to their true class label. These measures are superior to the best practice baseline in that, for a linear computation cost, they also quantify each observation' classification complexity in an explainable form.
  arXiv  Detail & Related papers  (2021-12-22T12:17:08Z)
• Evaluating State-of-the-Art Classification Models Against Bayes Optimality [106.50867011164584]
  We show that we can compute the exact Bayes error of generative models learned using normalizing flows. We use our approach to conduct a thorough investigation of state-of-the-art classification models.
  arXiv  Detail & Related papers  (2021-06-07T06:21:20Z)

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.