Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truth

• URL: http://arxiv.org/abs/2605.25052v1
• Date: Sun, 24 May 2026 12:57:01 GMT
• Title: Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truth
• Abstract summary: Chains of thought (CoTs) have become central in interpreting and auditing behaviors of large language models.<n>We develop an automated labeling pipeline that yields ground-truth faithfulness labels at both the step and CoT level.<n>Our experiments show that most metrics perform near chance, exhibit strong prediction biases and degrade on longer CoTs.
• Score: 24.21103008618097
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Chains of thought (CoTs) have become central in interpreting and auditing behaviors of large language models. Yet growing evidence suggests that these traces often fail to faithfully represent the computations behind a model's predictions. Several faithfulness metrics have been proposed, but whether they indeed measure faithfulness remains unknown. Answering this requires ground-truth labels, which are hard to obtain since internal computations are not directly observable. Consequently, most works proposing metrics report only absolute scores or comparisons to prior metrics, and the few existing benchmarks rely on proxies like plausibility or importance, properties orthogonal to faithfulness that can mislead about whether a CoT can be trusted. We address this challenge by constructing tasks whose outputs reveal which intermediate computations must have produced them, and developing an automated labeling pipeline that yields ground-truth faithfulness labels at both the step and CoT level. Building on this methodology, we present BonaFide, a benchmark of 3,066 labeled CoTs across 13 tasks and 10 models, and use it to conduct the first systematic evaluation of prominent faithfulness metrics. Our experiments show that most metrics perform near chance, exhibit strong prediction biases and degrade on longer CoTs. The best metric reaches only 0.70 AUROC at the CoT level while another reaches 0.59 at the step level, with neither transferring across settings, while entailing prohibitively high computational cost. Our results expose fundamental gaps in current faithfulness evaluation and call for the development of more reliable and efficient metrics.

Related papers

• PerceptionRubrics: Calibrating Multimodal Evaluation to Human Perception [75.70273182182397]
  PerceptionRubrics is a rubric-based evaluation framework.<n>It addresses the gap between saturated benchmark scores and real-world brittleness.<n>It pairs 1,038 information-dense images with over 10,000 instance-specific rubrics.
  arXiv  Detail & Related papers  (2026-06-26T17:59:15Z)
• Conformal Reliability: A New Evaluation Metric for Conditional Generation [62.761166941396844]
  Conditional generative models have recently remarkable success in various applications.<n>We propose a novel evaluation metric called reliability based on prediction set at a pre-specified confidence level.<n>We introduce Conformal ReLiability (CReL), a framework that can (i) construct a prediction set and (ii) accurately optimize the reliability score within the constructed prediction set.
  arXiv  Detail & Related papers  (2026-05-29T03:52:44Z)
• Are We Overconfident in Models and Results for Semi-Supervised 3D Medical Image Segmentation? [4.120238673372104]
  Semi-supervised learning has become a dominant paradigm for reducing annotation costs.<n>We argue that the current progress is clouded by a twofold overconfidence problem.<n>We propose a tri-space principled segmentation framework founded on a dual-axis reliability assessment engine.
  arXiv  Detail & Related papers  (2026-05-25T08:16:40Z)
• FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning [49.65751420291115]
  Chain-of-Thought (CoT) prompting has improved LLM reasoning, but models often generate explanations that appear coherent while containing unfaithful intermediate steps.<n>We propose FACT-E, a causality-inspired framework for evaluating CoT quality.<n>We show that FACT-E improves reasoning-trajectory selection and yields stronger in-context learnings.
  arXiv  Detail & Related papers  (2026-04-12T15:35:08Z)
• 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)
• FamiCom: Further Demystifying Prompts for Language Models with Task-Agnostic Performance Estimation [73.454943870226]
  Language models have shown impressive in-context-learning capabilities. We propose a measure called FamiCom, providing a more comprehensive measure for task-agnostic performance estimation.
  arXiv  Detail & Related papers  (2024-06-17T06:14:55Z)
• Goodhart's Law Applies to NLP's Explanation Benchmarks [57.26445915212884]
  We critically examine two sets of metrics: the ERASER metrics (comprehensiveness and sufficiency) and the EVAL-X metrics. We show that we can inflate a model's comprehensiveness and sufficiency scores dramatically without altering its predictions or explanations on in-distribution test inputs. Our results raise doubts about the ability of current metrics to guide explainability research, underscoring the need for a broader reassessment of what precisely these metrics are intended to capture.
  arXiv  Detail & Related papers  (2023-08-28T03:03:03Z)
• TRUST-LAPSE: An Explainable and Actionable Mistrust Scoring Framework for Model Monitoring [4.262769931159288]
  We propose TRUST-LAPSE, a "mistrust" scoring framework for continuous model monitoring. We assess the trustworthiness of each input sample's model prediction using a sequence of latent-space embeddings. Our latent-space mistrust scores achieve state-of-the-art results with AUROCs of 84.1 (vision), 73.9 (audio), and 77.1 (clinical EEGs)
  arXiv  Detail & Related papers  (2022-07-22T18:32:38Z)
• TRUE: Re-evaluating Factual Consistency Evaluation [29.888885917330327]
  We introduce TRUE: a comprehensive study of factual consistency metrics on a standardized collection of existing texts from diverse tasks. Our standardization enables an example-level meta-evaluation protocol that is more actionable and interpretable than previously reported correlations. Across diverse state-of-the-art metrics and 11 datasets we find that large-scale NLI and question generation-and-answering-based approaches achieve strong and complementary results.
  arXiv  Detail & Related papers  (2022-04-11T10:14:35Z)

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.