[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273393-105":59,"doc-detail-273393-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","know2guess-a-contamination-aware-multi-zone-benchmark-for-knowledge-boundary-evaluation-in-large-language-models","Know2Guess - A Contamination-Aware Multi-Zone Benchmark for Knowledge-Boundary Evaluation in Large Language Models","","Large language models should answer questions supported by their knowledge and abstain when a unique answer is not supported. Existing factuality benchmarks seldom evaluate both decisions while also separating epistemic abstention from policy refusal and recording contamination risk. The work introduces Know2Guess, a 1,200-item multi-zone benchmark spanning five domains, including synthetic unknowns in a dedicated abstention zone. Experiments across nine open-weight models quantify answering, abstention, refusal, prompt sensitivity, calibration, and contamination-related behavior.",{"@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/know2guess-a-contamination-aware-multi-zone-benchmark-for-knowledge-boundary-evaluation-in-large-language-models/273393/",{"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/know2guess-a-contamination-aware-multi-zone-benchmark-for-knowledge-boundary-evaluation-in-large-language-models/273393.png","ImageObject",300,407,{"name":92,"@type":93},"Anda","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-06","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",15,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does Know2Guess address in evaluating large language models?","Question",{"text":113,"@type":114},"It jointly evaluates whether models answer supported questions and abstain when answers are unsupported, while separating epistemic abstention from policy refusal and tracking contamination risk.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How is the benchmark organized into zones?",{"text":118,"@type":114},"Zones A–C require answers and vary in salience or transformation difficulty, while Zone D contains synthetic unknown questions with no unique supported gold answer and expects epistemic abstention.",{"name":120,"@type":111,"acceptedAnswer":121},"What evaluation protocol does Know2Guess use to distinguish decision types?",{"text":122,"@type":114},"A structured output protocol separates ANSWER, ABSTAIN_DONT_KNOW, and REFUSE_POLICY, and computes reliability by crediting correct answers in A–C and correct epistemic abstentions in D.","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},273393,1789471238,{"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},962075006959,"https://ap-avatar.wpscdn.com/avatar/e0002397efbe92a78e?_k=1776741047341049297","arXiv : 2606 .26 10 1v2 [cs .CL] 2 1 Aug 2026  \nKnow2Guess: A Contamination-Aware Multi-Zone Benchmark for Knowledge-Boundary Evaluation in Large  \nLanguage Models  \nRenwei Meng 1, ∗  \nBowen Zhang 1, ∗ Xuanyan Qiu 1  \nJian Wang2 Xican Wang 1 Shengan Yang 1  \nHaoyi Wu 1  \n1Anhui University, Hefei, China  \n2 Guangxi University, Nanning, China  \n[R32314095@stu.ahu.edu.cn](R32314095@stu.ahu.edu.cn), [wd2324041@stu.ahu.edu.cn](wd2324041@stu.ahu.edu.cn)  \n∗ Equal contribution  \nAbstract  \nLarge language models should answer questions supported by their knowledge and abstain when a unique answer is not supported. Existing factuality benchmarks seldom evaluate both decisions while also separating epistemic abstention from policy refusal and recording contamination risk. We present Know2Guess, a 1,200-item benchmark organized into three answer-expected zones and one abstentionexpected zone across five domains. Zone D contains 240 programmatically constructed syntheticunknown items with no hidden gold answers: single-entity instantiations, unsupported relation recombinations, and underdetermined questions. Template-family-aware splitting prevents construction families from crossing development and test. We evaluate nine open-weight models from 0.25B to 24B parameters in 18 deterministic runs, including answer-only and prompt-form controls, strict and normalized parsing, 10,000-resample bootstrap intervals, matched contamination analyses, and cost-sensitive scoring. Qwen3.5-9B provides the strongest observed balance, with reliability 0.5704, answer-expected accuracy 0.4833, and productive abstention 0.9249 . Mistral-Small-3.2-24B ranks second at 0.5398 . Qwen3-8B abstains on every Zone-D item but unnecessarily abstains on 35.99% of answer-expected items, illustrating why unknown-only accuracy is insufficient. Prompt wording changes reliability by as much as 0.1620 for Llama-3-8B. A character-and word-TF–IDF probe distinguishes Zone D from A–C with AUROC 0.9825 and AUPRC 0.9488 on held-out questions, exposing residual construction artifacts. Know2Guess is therefore an auditable diagnostic of answering, abstention, refusal, prompt sensitivity, calibration, and contamination-related behavior, with explicit limits on the construct validity of synthetic unknowns.  \nKeywords: large language models; knowledge boundaries; hallucination; abstention; refusal; data contamination; benchmark auditing  \n1 Introduction  \nLarge language models (LLMs) routinely produce fluent answers beyond the boundary of supported knowledge. This ability is useful when the answer is correct, but harmful when an unsupported guess is expressed with confidence. Accuracy alone gives an incomplete account of such behavior. A deployment-facing evaluation should also ask whether a model abstains when the requested fact is unavailable or underdetermined, whether it avoids abstaining on answerable questions, and whether a non-answer reflects epistemic caution rather than a policy refusal.  \nExisting evaluations cover parts of this problem. Truthfulness and hallucination benchmarks measure false or unverifiable generations [13, 9]; broad suites report multiple quality dimensions [12]; selectiveprediction work studies when a system should decline to answer [4, 21]; and temporally refreshed evaluations address staleness and some forms of leakage [10, 22] . Prompt-sensitive studies further show that  \napparent knowledge is partly an elicitation property [26, 8] . These strands rarely place answerable items, abstention-expected items, policy refusal, prompt variation, calibration, and contamination metadata inside one frozen decision protocol.  \nKnow2Guess addresses this gap with four build-time zones. Zones A–C require an answer and vary in public salience or transformation difficulty. Zone D contains synthetic questions for which the benchmark supplies no unique supported answer and expects epistemic abstention. A structured output protocol distinguishes ANSWER, ABSTAIN_DONT_KNOW,","cbCaimbtg71xMs8s","https://ap.wps.com/l/cbCaimbtg71xMs8s","pdf",407005,29,"English","# Introduction\n## Four build-time zones\n## Empirical questions and contributions\n# Abstract","[{\"question\":\"What problem does Know2Guess address in evaluating large language models?\",\"answer\":\"It jointly evaluates whether models answer supported questions and abstain when answers are unsupported, while separating epistemic abstention from policy refusal and tracking contamination risk.\"},{\"question\":\"How is the benchmark organized into zones?\",\"answer\":\"Zones A–C require answers and vary in salience or transformation difficulty, while Zone D contains synthetic unknown questions with no unique supported gold answer and expects epistemic abstention.\"},{\"question\":\"What evaluation protocol does Know2Guess use to distinguish decision types?\",\"answer\":\"A structured output protocol separates ANSWER, ABSTAIN_DONT_KNOW, and REFUSE_POLICY, and computes reliability by crediting correct answers in A–C and correct epistemic abstentions in D.\"}]","Know2Guess - A Contamination-Aware Multi-Zone Benchmark for Knowledge-Boundary Evaluation in Large Language Models | PDF",73]