[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83829-en":3,"doc-seo-83829-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},83829,2336464648322,"Aria","https://ap-avatar.wpscdn.com/avatar/2200025388227c56fec?_k=1778556882303663488",8,"Research & Report","Progressive Disclosure for LLM-Maintained Wiki Knowledge Bases a Preregistered Ablation","LLM agents answer questions using structured knowledge bases they help maintain, and “progressive disclosure” is often assumed to reduce cost by replacing monolithic indexes with a compact catalog plus one-line per-page summaries. This work tests that assumption on a 709-page LLM-maintained markdown wiki by retrofitting it for progressive disclosure. A preregistered ablation varies only the access structure while page bodies remain byte-identical. Results show quality is non-inferior, while cost decreases across access regimes with more targeted retrieval behavior.","arXiv :2607 .04576v 1 [ cs .CL] 6 Jul 2026  \nProgressive Disclosure for LLM-Maintained Wiki Knowledge Bases: a Preregistered Ablation  \nTheodore O. Cochran  \nAI for Altruism (A4A)  \n[theo@ai4altruism.org](theo@ai4altruism.org)  \nAbstract  \nLLM agents increasingly answer questions against structured knowledge bases that they themselves help maintain. A common efficiency intuition says that progressive disclosure should make this cheaper: keep a compact catalog and a one-line summary for each page, and the agent loads only what it needs instead of a large monolithic index. We test that intuition on areal 709-page markdown knowledge base maintained by an LLM. We retrofit it for progressive disclosure and run a preregistered ablation in which four versions of the corpus differ only in how the agent reaches the content, never in the content itself. The page bodies are byte-identical across arms, frozen as immutable git tags, so any difference we measure is due to access structure alone. We cross those arms with three access conditions, spanning a protocol-constrained agent, a free self-routing agent, and a catalog-preload regime, and grade the answers blind against verified gold references using a judge from a different model family.  \nA preparatory pilot upended the original premise. A capable tool-using agent never loads the monolithic index in the first place: it infers a page’s location from the question and reads it directly. The specific saving the retrofit was built to capture therefore does not materialize for such an agent, so we made answer quality the primary outcome and treated cost as secondary and condition-dependent. The confirmatory result is a measured-versus-nominal story. Quality is non-inferior: the retrieval arm matches the index baseline within the pre-set margin. Cost, however, falls in every access regime, from about a third for a self-routing agent to well over half under catalog-preload, with every confidence interval excluding zero. The saving does not come from avoiding the index load, which a capable agent sidesteps anyway, but is associated with more targeted access: the retrieval arm cites fewer pages and takes fewer tool turns per answer. The study doubles as a case study in evaluation validity, applying the same threat-to-validity discipline to the tooling that produced the result.  \n1 Introduction  \nRetrieval over a knowledge base is a core primitive for LLM agents, and a widespread design instinct is that less context is cheaper and no worse: if an agent can be handed a compact catalog and short per-item summaries, it should be able to route to the few relevant items and answer without ingesting a large index or many full documents. This “progressive disclosure” intuition, a term borrowed from user-interface design [Nielsen, 2006], motivates a range of practical choices (table-of-contents prompts, per-document abstracts, and lightweight retrieval tools), and it underlies a common expectation that restructuring a corpus for leaner access yields large token savings. That expectation is what we put to the test.  \nWe subject that intuition to a controlled test on a real artifact: a 709-page markdown wiki knowledge base, its pages cross-linked into a navigable graph, that has been continuously maintained by an LLM.  \nWe retrofit the knowledge base for progressive disclosure and ask two questions. First (RQ1), does the restructuring change the quality of an agent’s answers relative to a conventional index-catalog baseline? Second (RQ2), does it change the cost of answering, and does that effect depend on how the agent is allowed to access the corpus?  \nTwo features make the study unusually clean. (i) The corpus is frozen as immutable git tags, anda content-parity gate verifies that page bodies are byte-identical across arms, so any measured difference is attributable to access structure rather than content. (ii) The design, hypotheses, replicate count, and analysis plan were preregistered before the con","cbCailiVCbkYFTym","https://ap.wps.com/l/cbCailiVCbkYFTym","pdf",380689,5,1,14,"English","en",105,"# Abstract\n# Introduction\n## Progressive disclosure intuition\n## Study design and research questions\n## Content parity and preregistration\n## Pilot outcome and outcome reframing","[{\"question\":\"What does the paper test regarding progressive disclosure for LLM agents?\",\"answer\":\"It evaluates whether restructuring an LLM-maintained wiki into progressive disclosure affects answer quality compared with an index-catalog baseline, and whether it changes answering cost depending on access conditions.\"},{\"question\":\"How does the study ensure that measured differences come from access structure rather than content?\",\"answer\":\"The knowledge base page bodies are frozen as immutable git tags, and a content-parity gate verifies byte-identical page content across experimental arms.\"},{\"question\":\"What mechanism explains the observed cost savings despite the original loading-cost intuition?\",\"answer\":\"A capable tool-using agent already avoids loading the monolithic index by inferring page locations from the question; savings instead arise from more targeted access—fewer pages cited and fewer tool turns per answer.\"}]",1784190751,35,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"progressive-disclosure-for-llm-maintained-wiki-knowledge-bases-a-preregistered-ablation","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/progressive-disclosure-for-llm-maintained-wiki-knowledge-bases-a-preregistered-ablation/83829/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What does the paper test regarding progressive disclosure for LLM agents?","Question",{"text":76,"@type":77},"It evaluates whether restructuring an LLM-maintained wiki into progressive disclosure affects answer quality compared with an index-catalog baseline, and whether it changes answering cost depending on access conditions.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does the study ensure that measured differences come from access structure rather than content?",{"text":81,"@type":77},"The knowledge base page bodies are frozen as immutable git tags, and a content-parity gate verifies byte-identical page content across experimental arms.",{"name":83,"@type":74,"acceptedAnswer":84},"What mechanism explains the observed cost savings despite the original loading-cost intuition?",{"text":85,"@type":77},"A capable tool-using agent already avoids loading the monolithic index by inferring page locations from the question; savings instead arise from more targeted access—fewer pages cited and fewer tool turns per answer.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":20,"slug":138},19,"General","general"]