[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84229-en":3,"doc-seo-84229-105":29,"detail-sidebar-cat-0-en-105":90},{"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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84229,962075114101,"Seraphina","https://ap-avatar.wpscdn.com/avatar/e000253a75eb197efd?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780044092746381165",8,"Research & Report","The Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents","A self-evolving agent retires poor skills by observing them fail, but skill retirement relies on an honest reward signal—an assumption violated by LLM judges used for reference-free tasks. The work shows that judge bias does more than add noise: it silences the curator by disrupting the evidence flow required for contribution-based retirement. Through corrupted-reward analysis and controlled defect-injection, it identifies a sharp false-pass bias threshold beyond which retirement cannot be recovered, with downstream harm depending on whether synthesis is simultaneously starved.","arXiv :2607 .07436v 1 [ cs .AI] 8 Jul 2026  \nThe Blind Curator: How a Biased Judge Silently Disables Skill Retirement in Self-Evolving Agents  \nXing Zhang1, Yanwei Cui1, Guanghui Wang1, Ziyuan Li2, Wei Qiu2, Bing Zhu2, Peiyang He1 ∗  \n1AWS Generative AI Innovation Center  \n2HSBC Holdings Plc., HSBC Technology Center, China  \nAbstract  \nA self-evolving agent retires its bad skills by watching them fail, so what happens when the judge cannot see the failures? Skill retirement is the structural constraint that keeps a growing library from drifting below the no-skill baseline, but its guarantee assumes an unbiased reward, which is false for the LLM judges that reference-free tasks force upon us. We show that a biased judge does not merely add noise; it silently switches off the curator. We make this precise with a corrupted-reward analysis and, isolating the causal channel by injecting corruption on top of a deterministic reward, a behavioral study on a reference-free report-writing testbed with a code-generation cross-check. Symmetric noise leaves retirement intact, but false-pass bias (failures slipping through as passes) disables contributionbased retirement past a sharp threshold that no amount of data can cross.  \nSeparating genuine retirement from cap-eviction churn shows this mechanism failure is universal, holding across domains and failure rates and sparing only near-zero-false-pass, verifier-like graders. The downstream outcome, though, is regime-dependent: eval quality degrades only where the same corruption also starves skill synthesis, and otherwise holds steady, so the disabled curator is silent, surfacing in no aggregate metric. The contribution is a behavioral safety result, not a performance one. A cheap defect-injection audit then tells an operator, before deployment, which side of the threshold their judge occupies.  \n1 Introduction  \nA self-evolving agent that accumulates skills without governance degrades, as stale and redundant entries crowd retrieval, a failure mode recently named library drift (Zhang et al., 2026a) . Agents already learn by synthesizing new skills from their own failures (Wanget al., 2023; Zhao et al., 2024); the missing piece is governance, retiring skills that stop helping under a bounded cap. Ratchet (Zhang et al., 2026b) makes this precise: retirement keeps a growing library from drifting more than a fixed margin below the no-skill baseline. But that guarantee rests on one quiet assumption, that the signal telling the agent which skills failed is honest (the per-skill contribution estimator is unbiased) . Coding and QA satisfy it with unit tests and exact-match graders; the tasks agents increasingly face (research synthesis, long-form reporting, analysis) do not: with no golden answer the only scalable grader is an LLM judge (Zheng et al., 2023), and its error is not white noise. Judges are systematically biased rather than merely inconsistent (Wang et al., 2024a; Stureborg et al., 2024), tending to wave through certain failure classes (a confident misquote; a flipped conclusion that stays fluent), so their error is asymmetric: failures get reported as passes. The consequence is a blind curator (Fig. 1): the component that should retire bad skills stops seeing the evidence it retires on. Where library drift was the disease governance was built to cure, curator blindness is what befalls the cure itself when the reward is fallible.  \n∗ Corresponding author: [peiyan@amazon.com](peiyan@amazon.com)  \nThe blind-curator failure mode  \nFigure 1: The blind curator failure mode. The same failure-driven loop (solve, judge, retire) under an honest reward (left) and a false-pass judge (right): false passes break the Judge→Curator evidence, so retirement quietly stops while aggregate outcomes can look normal. The gap opens at a sharp threshold.  \nWe treat that asymmetry as the object of study. Our thesis turns on two separable knobs of the reward channel: a symmetric noise rate ρ (true label flipped ","cbCaib7ETfglRCu7","https://ap.wps.com/l/cbCaib7ETfglRCu7","pdf",484175,1,16,"English","en",105,"# Abstract\n# Introduction\n# Related work\n# The blind-curator failure mode\n# Contributions","[{\"question\":\"How does the proposed “blind curator” failure mode occur?\",\"answer\":\"The mechanism relies on the Judge→Curator evidence loop: the judge must correctly report failures so the curator can retire bad skills. A biased judge that turns failures into passes breaks this evidence, causing retirement to silently stop while overall aggregate outcomes can remain normal.\"},{\"question\":\"What factors define the reward corruption studied in the paper?\",\"answer\":\"The analysis separates symmetric noise rate ρ from a false-pass rate ρF→P (fraction of true failures reported as passes). These knobs affect the agent’s retirement behavior differently, with false-pass bias driving a critical failure threshold.\"},{\"question\":\"What is the key result about skill retirement under false-pass bias?\",\"answer\":\"Beyond a sharp threshold where ρF→P exceeds (1 − τ)/2, no amount of data can rescue contribution-based retirement. Symmetric noise can be tolerated, but false-pass bias disables curator behavior past the cliff.\"}]",1784194183,40,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"the-blind-curator-how-a-biased-judge-silently-disables-skill-retirement-in-self-evolving-agents","",{"@graph":35,"@context":84},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/the-blind-curator-how-a-biased-judge-silently-disables-skill-retirement-in-self-evolving-agents/84229/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"How does the proposed “blind curator” failure mode occur?","Question",{"text":74,"@type":75},"The mechanism relies on the Judge→Curator evidence loop: the judge must correctly report failures so the curator can retire bad skills. A biased judge that turns failures into passes breaks this evidence, causing retirement to silently stop while overall aggregate outcomes can remain normal.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"What factors define the reward corruption studied in the paper?",{"text":79,"@type":75},"The analysis separates symmetric noise rate ρ from a false-pass rate ρF→P (fraction of true failures reported as passes). These knobs affect the agent’s retirement behavior differently, with false-pass bias driving a critical failure threshold.",{"name":81,"@type":72,"acceptedAnswer":82},"What is the key result about skill retirement under false-pass bias?",{"text":83,"@type":75},"Beyond a sharp threshold where ρF→P exceeds (1 − τ)/2, no amount of data can rescue contribution-based retirement. Symmetric noise can be tolerated, but false-pass bias disables curator behavior past the cliff.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,118,121,126,129,133],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":28,"slug":117},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":45,"category_name":123,"show_sort_weight":124,"slug":125},9,"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":124,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]