[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273694-105":59,"doc-detail-273694-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","when-text-and-numbers-disagree-evidence-arbitration-in-large-language-models-preprint-abstract-and-study-summary","WHEN TEXT AND NUMBERS DISAGREE: EVIDENCE ARBITRATION IN LARGE LANGUAGE MODELS - Preprint abstract and study summary","","Large language models increasingly operate in decision settings where narrative summaries, numerical observations, and external tool outputs may conflict. This work studies how LLMs arbitrate when modalities support opposite decisions, introducing a controlled synthetic benchmark that generates both numerical time series and natural language summaries from latent risk trajectories with known alignment to ground truth. The benchmark enables independent manipulation of modality, temporal recency, source reliability, and evidence provenance. Experiments across open-weight instruction-tuned models show systematic arbitration: distinct text-versus-number preferences, stronger effects of temporal recency than explicit reliability cues, and over-reliance on external forecasts even when they contradict contextual evidence. The findings indicate heuristic integration strategies and a failure mode for tool-augmented decision systems.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/when-text-and-numbers-disagree-evidence-arbitration-in-large-language-models-preprint-abstract-and-study-summary/273694/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/when-text-and-numbers-disagree-evidence-arbitration-in-large-language-models-preprint-abstract-and-study-summary/273694.png","ImageObject",300,407,{"name":92,"@type":93},"Sage","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-20","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",12,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does the preprint address about large language models?","Question",{"text":113,"@type":114},"It examines how LLMs choose between conflicting evidence sources—such as textual summaries versus numerical observations and external tool outputs—when they support opposite decisions.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How does the study create controlled conflicts between text and numbers?",{"text":118,"@type":114},"It introduces a synthetic benchmark where latent risk trajectories generate both numerical time series and natural language summaries, with known ground-truth alignment so that exactly one evidence source matches the correct label.",{"name":120,"@type":111,"acceptedAnswer":121},"What arbitration behaviors do the authors observe in open-weight instruction-tuned models?",{"text":122,"@type":114},"Models show systematic, not random, arbitration: they exhibit text-versus-number preferences, respond more consistently to temporal recency than to explicit reliability cues, and may over-rely on external forecasts even when they conflict with contextual evidence.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},273694,1789472517,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":145,"preview_page_count":4},687197207057,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","arXiv : 2608 .20 1 16v 1 [cs .CL] 20 Aug 2026  \nWHEN TEXT AND NUMBERS DISAGREE: EVIDENCE ARBITRATION IN LARGE LANGUAGE MODELS  \nA PREPRINT  \nMattia Carletti 1 , Edward Phillips 1 , Fredrik K. Gustafsson 1 , Patitapaban Palo 1 , Lei Clifton2 , Danielle Belgrave3 , Xiao Gu 1 , David A. Clifton 1 ,4  \n1Department of Engineering Science, University of Oxford, Oxford, UK  \n2Nuffield Department of Primary Care Health Sciences, University of Oxford, Oxford, UK  \n3 GlaxoSmithKline, London, UK  \n4 Oxford Suzhou Centre for Advanced Research, University of Oxford, Suzhou, Jiangsu, China  \nABSTRACT  \nLarge language models (LLMs) are increasingly used in settings where textual summaries, numerical observations, and external tool outputs may provide conflicting evidence. We study how LLMs arbitrate between such sources when they support opposing decisions. To do so, we introduce a controlled synthetic benchmark in which latent risk trajectories generate both numerical time series and natural language summaries, allowing us to construct conflicts where exactly one evidence source is aligned with the ground-truth label. This design lets us independently manipulate modality, temporal recency, source reliability, and evidence provenance. Across open-weight instruction-tuned models, we find that arbitration behaviour is systematic rather than random: models exhibit distinct text-versus-number preferences, follow temporal recency more consistently than explicit reliability cues, and can over-rely on external forecasts even when they conflict with direct contextual evidence.  \nThese results suggest that current LLMs often rely on heuristic arbitration strategies when integrating heterogeneous evidence, highlighting a failure mode for tool-augmented decision systems.  \nKeywords Large Language Models · Multimodal Reasoning · Evidence Integration · Conflict Resolution · Numerical Reasoning  \n1 Introduction  \nIn many real-world decision-making scenarios, different sources of evidence may support conflicting conclusions. In healthcare, for example, a clinician’s assessment may describe a patient as stable while vital signs indicate ongoing deterioration. Similarly, in manufacturing, sensor readings may suggest normal machine operation even as maintenance reports point to impending failure. Such inconsistencies can arise for several reasons, including temporal mismatches between data sources (Figure 1), differences in source reliability, or the integration of observational evidence withinaccurate predictions generated by external tools.  \nLarge Language Models (LLMs) are increasingly being explored in high-stakes domains such as healthcare, mental health, and finance [Burton et al., 2024, Johri et al., 2025, Goh et al., 2025, Heinz et al., 2025, Wang et al., 2026, Xie et al., 2024, Hu and Zhao, 2026], while also being incorporated into decision-making and agentic systems that integrate external tools and heterogeneous data sources [Qin et al., 2024, Li, 2025, Yao et al., 2022] . Understanding how these models behave when confronted with conflicting evidence is therefore increasingly important.  \nIn this work, we use the term arbitration to refer to how models prioritize or reconcile competing signals when different sources support incompatible conclusions. Failures of arbitration may lead models to privilege stale, unreliable, or incorrect tool-generated evidence over more relevant observations, producing unreliable decisions even when the correct signal is present in the prompt.  \nOlder numeric evidence conflicts with newer text  \n| Prompt question\u003Cbr>Based on the evidence above, will the system's risk at time 14:25 be high (>0.5) or low (\u003C0.5)?\u003Cbr>A) high (>0.5) B) low (\u003C0.5) |\n| --- |\n| \u003Cbr>LLM\u003Cbr>|\n\nFigure 1: Evidence arbitration under conflict. A numerical time series and a textual summary, drawn from different windows of a shared latent risk trajectory, support conflicting predictions at the target time T+k: HIGH for the older numerical source, LO","cbCaicxSKAfc679h","https://ap.wps.com/l/cbCaicxSKAfc679h","pdf",970229,22,"English","# 1 Introduction\n## Conflicting evidence across modalities and domains\n## Definition of arbitration and implications\n# Evidence arbitration under conflict\n## Older numeric evidence vs newer text","[{\"question\":\"What problem does the preprint address about large language models?\",\"answer\":\"It examines how LLMs choose between conflicting evidence sources—such as textual summaries versus numerical observations and external tool outputs—when they support opposite decisions.\"},{\"question\":\"How does the study create controlled conflicts between text and numbers?\",\"answer\":\"It introduces a synthetic benchmark where latent risk trajectories generate both numerical time series and natural language summaries, with known ground-truth alignment so that exactly one evidence source matches the correct label.\"},{\"question\":\"What arbitration behaviors do the authors observe in open-weight instruction-tuned models?\",\"answer\":\"Models show systematic, not random, arbitration: they exhibit text-versus-number preferences, respond more consistently to temporal recency than to explicit reliability cues, and may over-rely on external forecasts even when they conflict with contextual evidence.\"}]","WHEN TEXT AND NUMBERS DISAGREE: EVIDENCE ARBITRATION IN LARGE LANGUAGE MODELS - Preprint abstract and study summary | PDF",55]