[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273543-105":59,"doc-detail-273543-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","grading-needs-a-rubric-not-intelligence","Grading Needs a Rubric, Not Intelligence","","Small language models can grade open-ended examination answers as reliably as substantially more expensive models when the grading process uses an explicit rubric. The paper tests this design principle behind ANY-TO-BENCH: a frontier model ingests source documents once to extract questions and their rubrics, then cheaper models reuse these rubrics for repeated grading. Experiments across six model configurations quantify how much score variance comes from answer identity, judge identity, and reasoning effort on both writing and judging sides, including ablations that isolate the rubric’s role.",{"@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/grading-needs-a-rubric-not-intelligence/273543/",{"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/grading-needs-a-rubric-not-intelligence/273543.png","ImageObject",300,407,{"name":92,"@type":93},"Maeve","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-30","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 is the core idea behind ANY-TO-BENCH?","Question",{"text":113,"@type":114},"A frontier model ingests the exam source documents once to extract each question and its complete rubric, then cheaper models reuse the rubric to perform all grading runs with minimal judge intelligence.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How does the paper measure reliability in grading?",{"text":118,"@type":114},"It evaluates multiple model configurations over 24 open-ended questions, generating thousands of per-question grades by grading each answer sheet three times and analyzing agreement and score variance.",{"name":120,"@type":111,"acceptedAnswer":121},"Why does an explicit rubric matter more than judge intelligence?",{"text":122,"@type":114},"Ablations show that removing rubric criteria and levels has little measurable impact, while removing the official answer collapses reliability and makes judge reasoning effort matter again—indicating the rubric, especially the official answer, carries the load.","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},273543,1789471982,{"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},5909877438554,"https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272","arXiv : 2608 . 17938v 1 [cs .CL] 18 Aug 2026  \nGrading Needs a Rubric, Not Intelligence  \nJhen-Ke Lin  \nNational Yang Ming Chiao Tung University  \n[jacob.cs14@nycu.edu.tw](jacob.cs14@nycu.edu.tw)  \nAbstract  \nSmall language models can grade open-ended examination answers as reliably as substantially more expensive models when they grade against an explicit rubric.  \nWe test this claim as the design principle behind ANY-TO-BENCH: a frontier model reads source documents once, at ingestion, to extract each question and its rubric; lower-cost models then perform all repeated grading work. We evaluate six cost-efficient model configurations from two model families at three reasoningeffort levels. Each configuration answers 24 open-ended examination questions, and each also grades every answer sheet three times, yielding 3,456 per-question grades. Scores depend overwhelmingly on the answer being graded: answer identity explains 95.6 % of score variance, whereas judge identity explains only 0.2 % . Raising a writer’s reasoning effort moves earned scores by as much as 0.143 of full marks, while raising a judge’s reasoning effort moves assigned scores by at most 0.006 . Six frontier-tier judges, added as a check, reproduce these scores and are no more reliable as a panel. Two ablations then decompose the rubric on the same questions and answers. Removing its criteria and levels while keeping the official answer changes nothing measurable. Removing the official answer as well collapses reliability (ICC 0.888 to 0.628), inflates scores, and makes judge reasoning effort matter again. The rubric is what decouples grading from judge intelligence, and within the rubric the official answer does nearly all the work. We find no evidence of length preference or same-family preference under rubric-anchored grading.  \n1 Introduction  \nTurning an existing examination into a machine-runnable benchmark is mostly a solved problem for multiple-choice items. It is an expensive problem for everything else. Open-ended questions, such as proofs, essays, translations, and drawings, need a grader, and the grader has historically been the costly, unscalable part. Using a language model as the judge [1, 2] moved that cost from humans to models. But the standard recipe still spends a frontier-model call on every grade, and it carries documented judge biases: a preference for longer answers [5], for answers in certain positions [4], and for the judge’s own model family [6] .  \nANY-TO-BENCH takes a different approach, built on an asymmetry between its two model-driven steps: ingestion and grading. Ingestion is where the intelligence goes. A frontier configuration reads the source documents and extracts questions, answer formats, and scoring rubrics, and it does this once per exam. Throughout this paper, a question’s rubric is its complete extracted scoring standard: the official reference answer, plus explicit criteria with defined levels where the source publishes them. Grading happens on every benchmark run, and the tool bets that it needs almost no intelligence at all: with the rubric in hand, grading reduces to applying it. That is work small models should handle cheaply, repeatedly, and interchangeably.  \nThat is a testable claim, and we test it. Six configurations served both as answer writers and as judges over 24 open-ended questions from three Taiwanese national examinations. They comprise the small  \nPreprint.  \ntier of one model family and the mid tier of another, each at three reasoning-effort settings. Because the same configurations act both as writers and as judges, the experiment turns the reasoning-effort dial once on each side of the grading relation. The writer side shows what happens when capability matters; the judge side tests whether it matters for grading. We find:  \n• Judge intelligence does not matter (§5) . Reasoning effort moves what a writer earns by up to 0.143 of full marks, so capability registers when it sits on the answering side","cbCailNcZLtszT6P","https://ap.wps.com/l/cbCailNcZLtszT6P","pdf",362203,11,"English","# Introduction\n## Related work","[{\"question\":\"What is the core idea behind ANY-TO-BENCH?\",\"answer\":\"A frontier model ingests the exam source documents once to extract each question and its complete rubric, then cheaper models reuse the rubric to perform all grading runs with minimal judge intelligence.\"},{\"question\":\"How does the paper measure reliability in grading?\",\"answer\":\"It evaluates multiple model configurations over 24 open-ended questions, generating thousands of per-question grades by grading each answer sheet three times and analyzing agreement and score variance.\"},{\"question\":\"Why does an explicit rubric matter more than judge intelligence?\",\"answer\":\"Ablations show that removing rubric criteria and levels has little measurable impact, while removing the official answer collapses reliability and makes judge reasoning effort matter again—indicating the rubric, especially the official answer, carries the load.\"}]","Grading Needs a Rubric, Not Intelligence | PDF",28]