Measuring Faithfulness Depends on How You Measure: Classifier Sensitivity in LLM Chain-of-Thought Evaluation

• URL: http://arxiv.org/abs/2603.20172v2
• Date: Mon, 23 Mar 2026 21:10:16 GMT
• Title: Measuring Faithfulness Depends on How You Measure: Classifier Sensitivity in LLM Chain-of-Thought Evaluation
• Abstract summary: Recent work on chain-of-thought faithfulness reports single aggregate numbers.<n>This paper provides evidence that faithfulness is not an objective, measurable property of a model.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent work on chain-of-thought (CoT) faithfulness reports single aggregate numbers (e.g., DeepSeek-R1 acknowledges hints 39% of the time), implying that faithfulness is an objective, measurable property of a model. This paper provides evidence that it is not. Three classifiers (a regex-only detector, a regex-plus-LLM pipeline, and a Claude Sonnet 4 judge) are applied to 10,276 influenced reasoning traces from 12 open-weight models spanning 9 families and 7B to 1T parameters. On identical data, these classifiers produce faithfulness rates of 74.4%, 82.6%, and 69.7%. Per-model gaps range from 2.6 to 30.6 percentage points; all pairwise McNemar tests are significant (p < 0.001). The disagreements are systematic: Cohen's kappa ranges from 0.06 ("slight") for sycophancy hints to 0.42 ("moderate") for grader hints, and the asymmetry is pronounced: for sycophancy, 883 cases are classified as faithful by the pipeline but unfaithful by the Sonnet judge, while only 2 go the other direction. Classifier choice can also reverse model rankings: Qwen3.5-27B ranks 1st under the pipeline but 7th under Sonnet; OLMo-3.1-32B moves from 9th to 3rd. Different classifiers operationalize faithfulness at different levels of stringency (lexical mention versus epistemic dependence), yielding divergent measurements on the same behavior. These results indicate that published faithfulness numbers cannot be meaningfully compared across studies using different classifiers, and that future evaluations should report sensitivity ranges across multiple classification methodologies.

Related papers

• Token Counts Are Not Model Lineage: A Frozen-Threshold Holdout Study of Black-Box LLM API Fingerprinting [4.093009863847175]
  Black-box model attribution is increasingly relevant when large language models are served through relay and reseller APIs.<n>A tempting low-cost signal is the prompt-token count returned by an OpenAI-compatible endpoint.<n>Yet the validity of this signal for broader emphmodel-family attribution has received little direct holdout testing.<n>We conduct a frozen-threshold study over 24 labeled endpoint pairs, split evenly into a development set and an untouched holdout set.
  arXiv  Detail & Related papers  (2026-08-30T17:58:49Z)
• Which Values Do LLMs Confuse? A Schwartz-Based Recognition Study [32.505127447635864]
  We study controlled top-1 recognition over Schwartz's ten basic values.<n>We evaluate 21 instruction-tuned LLM runs under a fixed ranked-response protocol.<n>We find that models often locate the correct motivational region while ranking close alternatives unstably.
  arXiv  Detail & Related papers  (2026-07-22T15:22:14Z)
• Strained Coherence: A Pre-Failure Signal in Coding Agent Execution Trajectories [0.0]
  LLM-based coding agents sometimes acknowledge a problem in their own reasoning and then proceed anyway.<n>We build a Claude Sonnet 4.6 judge that reads full trajectories and flags spans where the pattern occurs.<n>We evaluate it on 44 Terminal-bench-2 trajectories using a Qwen3.5-35B-A3B backbone.
  arXiv  Detail & Related papers  (2026-06-05T22:52:16Z)
• Automated Proving of Shannon-Type Entropy Inequalities via Fine-Tuned Language Models and Guided Tree Search [50.16356451328644]
  Proving Shannon-type entropy inequalities is a fundamental task in information theory.<n>We investigate whether small-scale large language models, fine-tuned on atomic proof steps, can automate this process.<n>We show that GPT-5.5 solves 1.7% samples under zero-shot prompting while Psitip solves 33.3% samples.
  arXiv  Detail & Related papers  (2026-06-04T05:43:12Z)
• FormInv: A Measurement Protocol for Semantic Invariance in Mathematical Reasoning Benchmarks [0.0]
  A paraphrase-quality audit of MathCheck detected 4 semantically incorrect paraphrases in 129 groups.<n>Removing them drops GPT-4o from rank 2 to rank 4 and elevates Claude Haiku and DeepSeek V3 above it.
  arXiv  Detail & Related papers  (2026-05-27T18:59:18Z)
• ATLAS: All-round Testing of Long-context Abilities across Scales [42.83200405769639]
  ATLAS is a benchmarking framework that redefines long-context evaluation as length-dependent capability profiling.<n> ATLAS contributes three methodological principles: (i) a layered taxonomy separating foundational operations from application workloads so failures can be attributed, (ii) length-aware AUC scoring that integrates score-length curves over a fixed 8K-1M grid, replacing single-point metrics with full degradation profiles, and (iii) ATLAScore, a harmonic-mean aggregate over taxonomy categories that penalizes imbalanced profiles.
  arXiv  Detail & Related papers  (2026-05-27T07:33:47Z)
• Distributional Energy-Based Models for Uncertainty-Aware Structured LLM Reasoning [40.342912574072024]
  Large Language Models produce structured outputs such as travel plans or code solutions.<n>Individual reasoning steps may appear correct while the output as a whole violates budgets, fails test cases, or contradicts earlier deductions.<n>We propose a learned quality scorer with deterministic analytical constraint penalties for verifying structured LLM outputs.
  arXiv  Detail & Related papers  (2026-05-15T17:08:27Z)
• Metric Unreliability in Multimodal Machine Unlearning: A Systematic Analysis and Principled Unified Score [14.579552536669217]
  Unlearning in Vision-Language Models (VLMs) is required for compliance with the General Data Protection Regulation Code.<n>We present the first study of reliability in multimodal unlearning.<n>We introduce the Unified Quality Score (UQS), a metric with weights derived from each metric's Spearman correlation with the metric distance.
  arXiv  Detail & Related papers  (2026-05-04T04:13:00Z)
• Lie to Me: How Faithful Is Chain-of-Thought Reasoning in Reasoning Models? [0.0]
  Chain-of-thought (CoT) reasoning has been proposed as a transparency mechanism for large language models in safety-critical deployments.<n>This study tests 12 open-weight reasoning models spanning 9 architectural families on 498 multiple-choice questions.<n>Across 41,832 inference runs, overall faithfulness rates range from 39.7% (Seed-1.6-Flash) to 89.9% (DeepSeek-V3.2-Speciale) across model families.
  arXiv  Detail & Related papers  (2026-03-23T21:21:37Z)
• From Black Box to Glass Box: Cross-Model ASR Disagreement to Prioto Review in Ambient AI Scribe Documentation [43.148402136307716]
  Cross-model disagreement among heterogeneous ASR systems can act as a reference-free uncertainty signal.<n>We transcribed 50 publicly available medical audio clips with eight ASR systems spanning commercial APIs and open-source engines.<n>Low-agreement regions were enriched for content disagreements, with the content fraction increasing from 53.9% to 73.9% across quintiles of high-risk mass.
  arXiv  Detail & Related papers  (2026-03-02T13:02:13Z)
• 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)
• Causal Understanding by LLMs: The Role of Uncertainty [43.87879175532034]
  Recent papers show LLMs achieve near-random accuracy in causal relation classification.<n>We investigate whether pretraining exposure to causal examples improves causal understanding.
  arXiv  Detail & Related papers  (2025-09-24T13:06:35Z)
• An Auditable Pipeline for Fuzzy Full-Text Screening in Systematic Reviews: Integrating Contrastive Semantic Highlighting and LLM Judgment [0.0]
  Full-text screening is the major bottleneck of systematic reviews.<n>We present a scalable, auditable pipeline that reframes inclusion/exclusion as a fuzzy decision problem.
  arXiv  Detail & Related papers  (2025-08-17T17:41:50Z)
• CausalDiff: Causality-Inspired Disentanglement via Diffusion Model for Adversarial Defense [61.78357530675446]
  Humans are difficult to be cheated by subtle manipulations, since we make judgments only based on essential factors.<n>Inspired by this observation, we attempt to model label generation with essential label-causative factors and incorporate label-non-causative factors to assist data generation.<n>For an adversarial example, we aim to discriminate perturbations as non-causative factors and make predictions only based on the label-causative factors.
  arXiv  Detail & Related papers  (2024-10-30T15:06:44Z)
• Chain-of-Thought Unfaithfulness as Disguised Accuracy [0.0]
  Chain-of-Thought (CoT) generations align with a large language model's (LLM) internal computations. A metric that measures a model's dependence on its CoT for producing an answer is proposed.
  arXiv  Detail & Related papers  (2024-02-22T17:23:53Z)
• Norm-Scaling for Out-of-Distribution Detection [6.309365332210523]
  Out-of-Distribution (OoD) inputs are examples that do not belong to the true underlying distribution of the dataset. In this paper, we propose norm-scaling which normalizes the logits separately for each class. We show that norm-scaling achieves 9.78% improvement in AUROC, 5.99% improvement in AUPR and 33.19% reduction in FPR95 metrics over previous state-of-the-art methods.
  arXiv  Detail & Related papers  (2022-05-06T22:31:36Z)

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.