Navigating the Mirage: A Dual-Path Agentic Framework for Robust Misleading Chart Question Answering
Summary
This paper introduces ChartCynics, an agentic dual-path framework that decouples perception from verification to robustly answer questions about misleading charts. It achieves state-of-the-art accuracy by using a diagnostic vision path and an OCR-driven data path, with a two-stage protocol for reasoning distillation and adversarial alignment.
View Cached Full Text
Cached at: 07/16/26, 01:40 AM
Paper page - Navigating the Mirage: A Dual-Path Agentic Framework for Robust Misleading Chart Question Answering
Source: https://huggingface.co/papers/2603.28583
Abstract
DespitethesuccessofVision-LanguageModels(VLMs),misleadingchartsremainasignificantchallengeduetotheirdeceptivevisualstructuresanddistorteddatarepresentations.WepresentChartCynics,anagenticdual-pathframeworkdesignedtounmaskvisualdeceptionviaa“skeptical“reasoningparadigm.Unlikeholisticmodels,ChartCynicsdecouplesperceptionfromverification:aDiagnosticVisionPathcapturesstructuralanomalies(e.g.,invertedaxes)throughstrategicROIcropping,whileanOCR-DrivenDataPathensuresnumericalgrounding.Toresolvecross-modalconflicts,weintroduceanAgenticSummarizeroptimizedviaatwo-stageprotocol:Oracle-InformedSFTforreasoningdistillationandDeception-AwareGRPOforadversarialalignment.Thispipelineeffectivelypenalizesvisualtrapsandenforceslogicalconsistency.EvaluationsontwobenchmarksshowthatChartCynicsachieves74.43%and64.55%accuracy,providinganabsoluteperformanceboostof~29%overtheQwen3-VL-8Bbackbone,outperformingstate-of-the-artproprietarymodels.Ourresultsdemonstratethatspecializedagenticworkflowscangrantsmalleropen-sourcemodelssuperiorrobustness,establishinganewfoundationfortrustworthychartinterpretation.
View arXiv pageView PDFAdd to collection
Get this paper in your agent:
hf papers read 2603\.28583
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2603.28583 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2603.28583 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2603.28583 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Chart Deception in Vision-Language Models: From Vulnerability to Mitigation
This paper introduces VisDeception, a benchmark for evaluating Vision-Language Model robustness to deceptive chart designs, and proposes a multi-agent mitigation framework that reduces deception-induced errors by grounding reasoning in structured chart metadata.
CURV: Enhancing Chart Understanding Through Curriculum Visual Grounded Reasoning
CURV proposes a curriculum learning framework that enhances chart question answering by developing intrinsic visual grounded reasoning in multimodal LLMs, achieving significant improvements on real-world benchmarks.
Is this chart lying to me? Automating the detection of misleading visualizations
This paper introduces Misviz, a benchmark dataset of 2,604 real-world visualizations and 57,665 synthetic ones annotated with 12 types of misleading design violations, enabling automated detection of deceptive charts. The work evaluates state-of-the-art multimodal LLMs and rule-based systems on this challenging task, addressing the gap in resources for training AI models to combat data visualization misinformation.
MIRAGE: Multi-Perspective Creative Language Model Reasoning with Reinforcement Learning Guidance
The paper introduces MIRAGE, an inference-time framework for enhancing LLM reasoning by dynamically switching perspectives using reinforcement learning guidance, outperforming existing prompting methods on various benchmarks.
VisAudit: Evaluating Multimodal Agents for Visual Diagnosis and Repair
VisAudit is a new benchmark for evaluating multimodal agents on autonomously diagnosing, repairing, and verifying flawed data visualizations, spanning 1,900 flawed charts across 21 chart types and 10 flaw categories. Experiments show the strongest multimodal model fully recovers only 47.4% of flawed charts in the autonomous-repair setting, revealing a substantial gap toward human-like visualization review.