When Derived Measurements Mislead: Quantifying and Mitigating LLM Over-Trust with Privileged-Modality Reliability Evidence

2026/07/30 by Zongheng Guo, Tao Chen, Tianli Li +6
Computer Science · #cs.AI

paper · pdf

25 pages, including references and supplementary material; 3 figures and 19 tables. Code: https://github.com/Zongheng-Guo/When-Derived-Measurements-Mislead

arxiv created 2026/07/30 · arxiv updated 2026/07/31

Abstract

Derived measurements increasingly enter large language model (LLM) pipelines as direct facts despite their instance-dependent validity. We define derived-feature over-trust (DFOT) as the failure in which a downstream LLM assigns such a measurement the epistemic status of a direct fact or uses it outside its valid scope. Using physiological sensing as a case study, D1 tests acceptance of a PPG-derived rhythm contradicted by offline ECG, whereas D2 tests rejection of an offline-confirmed reliable PPG rhythm under misleading severe history. ECG supplies training supervision and offline reference construction but is never shown to the LLM. Five estimands quantify this chain: conflict over-trust rate (COTR) and context-induced error rate (CIR) characterize D1/D2; correct repair rate (CRR) measures frozen-error repair; evidence-specific repair margin (ESRM) contrasts matched and patient-disjoint shuffled evidence; and utility harm rate (UHR) measures unnecessary verification among HIGH-reliability cases used without verification at baseline. The framework does not depend on a particular reliability generator. We demonstrate it on 50,000 paired PPG-ECG records using ECG-to-PPG privileged distillation as an illustrative baseline and PPG-only inference. On a protocol-locked 187-patient test, the baseline improves four repair and specificity endpoints by 1.82-6.69 percentage points, with all paired confidence intervals excluding zero; UHR increases by 0.67 percentage points (95% CI: -0.4 to +1.7). DFOT provides a common evaluation target for stronger mitigation methods. The code is available at https://github.com/Zongheng-Guo/When-Derived-Measurements-Mislead.

Citations

• BioX-Bridge: Model Bridging for Unsupervised Cross-Modal Knowledge Transfer across Biosignals
• OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data
• PPG-Distill: Efficient Photoplethysmography Signals Analysis via Foundation Model Distillation
• Learning Game-Playing Agents with Generative Code Optimization
• ECG-guided individual identification via PPG
• PaPaGei: Open Foundation Models for Optical Physiological Signals
• PhysioLLM: Supporting Personalized Health Insights with Wearables and Large Language Models
• ClashEval: Quantifying the tug-of-war between an LLM's internal prior and external evidence
• AudioGPT: Understanding and Generating Speech, Music, Sound, and Talking Head
• Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models
• Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation
• Large Language Models Encode Clinical Knowledge
• Large language models encode clinical knowledge
• Rich Knowledge Sources Bring Complex Knowledge Conflicts: Recalibrating Models to Reflect Conflicting Evidence
• Toward Understanding Privileged Features Distillation in Learning-to-Rank
• Language Models (Mostly) Know What They Know
• Entity-Based Knowledge Conflicts in Question Answering
• Consistent Estimators for Learning to Defer to an Expert
• Attention is not not Explanation
• Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift
• Attention is not Explanation
• SelectiveNet: A Deep Neural Network with an Integrated Reject Option
• Predict Responsibly: Improving Fairness and Accuracy by Learning to\n Defer
• On Calibration of Modern Neural Networks
• Selective Classification for Deep Neural Networks
• Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
• Cross Modal Distillation for Supervision Transfer
• Distilling the Knowledge in a Neural Network

Related

• MediEval: A Unified Medical Benchmark for Patient-Contextual and Knowledge-Grounded Reasoning in LLMs
• The Reliability of LLMs for Medical Diagnosis: An Examination of Consistency, Manipulation, and Contextual Awareness
• Failures Reveal What Metrics Miss: An Evidence-Driven Agent for Recursive Refinement of ECG Classifiers
• Judge-dependent safety gains and model-specific helpfulness costs of evidence-sufficiency prompting in clinical LLMs
• Overconfidence in LLM-as-a-Judge: Diagnosis and Confidence-Driven Solution