[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-273467-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-273467-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","hallutracer-hallucination-detection-via-depth-averaging-truth-signals","HalluTracer - Hallucination Detection via Depth-Averaging Truth Signals","","Hallucination threatens reliability in high-stakes deployments because even well-aligned large language models can generate confident yet factually incorrect text. The work shows that internal representations still contain linearly separable truthfulness signals. Existing white-box detectors discard distributed evidence by selecting isolated components or a single depth. HalluTracer aggregates truthfulness evidence across every Transformer layer before any answer token is produced.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/hallutracer-hallucination-detection-via-depth-averaging-truth-signals/273467/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/hallutracer-hallucination-detection-via-depth-averaging-truth-signals/273467.png","ImageObject",300,407,{"name":42,"@type":43},"Ava Thompson","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-17","2026-09-15",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",9,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What problem does HalluTracer address?","Question",{"text":63,"@type":64},"HalluTracer targets hallucination detection for large language models, where models can generate fluent but factually incorrect outputs in high-stakes settings.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does HalluTracer use truthfulness signals?",{"text":68,"@type":64},"It applies a layer-specific linear probe to read out a truth logit at the answer onset, then stacks probe outputs across all layers to form a logit trajectory.",{"name":70,"@type":61,"acceptedAnswer":71},"Why does depth averaging work in HalluTracer?",{"text":72,"@type":64},"A geometric analysis indicates per-layer truthfulness signals are weakly correlated, so simple depth averaging suppresses layer-specific noise while preserving most linearly accessible information.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},273467,1790620207,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":144,"read_time":145,"preview_page_count":4},1649267921044,"https://us-avatar.wpscdn.com/avatar/1800007509477c92dfb?_k=1786009248482753345","arXiv :2608 . 16353v1 [ cs .CL] 17 Aug 2026  \nHalluTracer: Hallucination Detection via Depth-Averaging Truth Signals  \nZhihao Guo 1,2 ∗ Zonghan Wu 1 ∗ Huan Huo2 DaYong Ye3 Junwei Zhang4 Weiran Yao5 Zhiwei Liu6 Qingsong Wen 1,7† Yilei Shao 1†  \n1East China Normal University, China 2University of Technology Sydney, Australia  \n3 City University of Macau, Macau 4Meta, USA 5 actAVA AI, USA  \n6Microsoft AI, USA 7 Squirrel Ai Learning, USA  \n[zhihao.guo-1@student.uts.edu.au](zhihao.guo-1@student.uts.edu.au) , [zhwu@sem.ecnu.edu.cn](zhwu@sem.ecnu.edu.cn) , [Huan.Huo@uts.edu.au](Huan.Huo@uts.edu.au)[dyye@cityu.edu.mo](dyye@cityu.edu.mo) , [Junweizhang23@gmail.com](Junweizhang23@gmail.com) , [weiran@actava.ai](weiran@actava.ai)  \n[zhiweiliu@microsoft.com](zhiweiliu@microsoft.com) , [qingsongedu@gmail.com](qingsongedu@gmail.com) , [yileishao@sem.ecnu.edu.cn](yileishao@sem.ecnu.edu.cn)  \nAbstract  \nEven well-aligned large language models confidently generate factually incorrect text, making hallucination a persistent reliability risk in high-stakes deployments.  \nThese models nonetheless carry linearly separable truthfulness signals in their internal representations. Existing white-box detectors, however, collapse this evidence to isolated components or a single depth, discarding discriminative information distributed across the full forward pass. We introduce HalluTracer, a detection framework that reads and aggregates truthfulness evidence across every layer of the forward pass before the model emits any answer token. A geometric analysis reveals that the per-layer signals are weakly correlated, so that simple depth averaging suppresses layer-specific noise and captures nearly all linearly accessible information. Across six open-source language models and five hallucination benchmarks, HalluTracer consistently outperforms matched white-box baselines, with gains ranging from one to fourteen points. Collectively, our work recasts hallucination detection from a layer-selection problem into a depth-aggregation problem governed by the geometric sparsity of the truthfulness signal.  \n1 Introduction  \nLarge language models (LLMs) produce fluent but unfaithful text, a failure mode termed hallucination, which poses concrete risks in safety-critical deployments [18, 33, 17] and persists despite alignment training [29, 3] . Mechanistic studies, however, reveal that these models internally encode truthrelevant structure that is not fully exploited during generation: linear probes recover separable truthfulness directions from hidden activations [26], and models can exhibit calibrated self-evaluation when explicitly prompted [20] . This gap between internal knowledge and output fidelity suggests a detection opportunity for token cost saving—if the model’s representations already distinguish fact from fabrication, that distinction can in principle be read out before any text is generated. Reliably doing so remains an open problem.  \nIn practice, existing methods capture only a fraction of this internal evidence. For example, ContrastConsistent Search [6] discovers latent truth directions without supervision but requires paired statements and operates at a single depth.  \n∗ Equal contribution.  \n†Corresponding authors.  \nPreprint.  \n(a) Layer-wise Activation Extraction (b) Truthfulness Projection (c) Trajectory Aggregation and Detection  \nFigure 1: Overview of HalluTracer. (a) At each Transformer layer, a readout extractor selects the hidden state at the answer-onset position. (b) A per-layer learned probe projects the readout onto a scalar truth logit. (c) The truth logits across all layers form a logit trajectory whose depth-averaged mean is fed to a logistic classifier for hallucination detection.  \nInference-Time Intervention [23] steers model behaviour using truthfulness directions extracted from a few attention heads, ignoring how factual commitment continuously develops across depth. Fact-Probe [16] demonstrates that a simple linear classifier","cbCaicAviTXhEm75","https://ap.wps.com/l/cbCaicAviTXhEm75","pdf",1036647,39,"English","# Introduction\n## Related Work\n## Method Overview\n## Inference and Geometric Rationale","[{\"question\":\"What problem does HalluTracer address?\",\"answer\":\"HalluTracer targets hallucination detection for large language models, where models can generate fluent but factually incorrect outputs in high-stakes settings.\"},{\"question\":\"How does HalluTracer use truthfulness signals?\",\"answer\":\"It applies a layer-specific linear probe to read out a truth logit at the answer onset, then stacks probe outputs across all layers to form a logit trajectory.\"},{\"question\":\"Why does depth averaging work in HalluTracer?\",\"answer\":\"A geometric analysis indicates per-layer truthfulness signals are weakly correlated, so simple depth averaging suppresses layer-specific noise while preserving most linearly accessible information.\"}]","HalluTracer - Hallucination Detection via Depth-Averaging Truth Signals | PDF",1789471559,98]