[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81877-en":3,"doc-seo-81877-105":31,"detail-sidebar-cat-0-en-105":93},{"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":28,"seo_description":14,"update_tm":29,"read_time":30},81877,2336464648322,"Aria","https://ap-avatar.wpscdn.com/avatar/2200025388227c56fec?_k=1778556882303663488",8,"Research & Report","The Powerless Noise: How Experimental Settings Shape the Reported Power of Noise","The paper investigates the “Power of Noise” in retrieval-augmented generation (RAG), where adding irrelevant documents to the input can improve question-answering performance. The authors reproduce Cuconasu et al.’s results and test robustness under expanded experimental settings, including different model choices, instruction prompting, and relaxed output length constraints. They find the effect is highly sensitive to inference configuration and can appear, weaken, or disappear. Error analysis attributes variance to truncation and malformed generations.","The Powerless Noise: How Experimental Settings Shape the  \nReported Power of Noise  \nMichał Mazuryk  \nUniversity of Amsterdam Amsterdam, The Netherlands [michal.mazuryk@student.uva.nl](michal.mazuryk@student.uva.nl)  \nFleur Dolmans  \nUniversity of Amsterdam Amsterdam, The Netherlands [fleur.dolmans@student.uva.nl](fleur.dolmans@student.uva.nl)  \nLouis Gehringer  \nUniversity of Amsterdam Amsterdam, The Netherlands [louis.gehringer@student.uva.nl](louis.gehringer@student.uva.nl)  \nIna Klaric  \nUniversity of Amsterdam Amsterdam, The Netherlands [ina.klaric@student.uva.nl](ina.klaric@student.uva.nl)  \nJia-Huei Ju  \nUniversity of Amsterdam Amsterdam, The Netherlands [j.ju@uva.nl](j.ju@uva.nl)  \nMohammad Aliannejadi  \nUniversity of Amsterdam Amsterdam, The Netherlands [m.aliannejadi@uva.nl](m.aliannejadi@uva.nl)  \narXiv :2607 .036 15v2 [ cs .IR] 10 Jul 2026  \nAbstract  \nRecent work has suggested that adding irrelevant documents to the input of retrieval-augmented generation (RAG) systems can improve question-answering performance, a phenomenon referred to as the “Power of Noise.” This motivated investigations into the role of noise in information retrieval. In this paper, we reproduce the main findings of Cuconasu et al. [6] and evaluate the robustness of the effect under extended experimental settings. We first confirm that the phenomenon holds under the original setup, which uses earlier-generation LLMs, restrictive prompting and constrained decoding settings. We subsequently introduce a series of extensions to investigate the underlying causes of the noise effect, examining the authors’ original design choices including the use of different models, instruction prompting, and relaxed output length constraints. Across these ablations, the Power-of-Noise pattern proves highly sensitive to inference configuration: it can appear, weaken, or disappear under small changes to prompt formulation and decoding limits. Combined with our error analysis, which shows substantial contributions from truncation and malformed generations, this variance indicates that the original effect cannot be robustly confirmed as a general benefit of noisy retrieval under these experimental conditions. More broadly, our work highlights the importance of carefully scrutinizing inference design in retrieval-augmented generation systems. Our code is available at [https://github.com/ina0105/The-Power-of-Noise-Reproduction](https://github.com/ina0105/The-Power-of-Noise-Reproduction).  \nCCS Concepts  \n• Information systems → Novelty in information retrieval.  \nKeywords  \nRetrieval-Augmented Generation; Noise in RAG; Prompt Composition; Retrieval Strategy  \nACM Reference Format:  \nMichał Mazuryk, Fleur Dolmans, Louis Gehringer, Ina Klaric, Jia-Huei Ju, and Mohammad Aliannejadi. 2026. The Powerless Noise: How Experimental  \nThis work is licensed under a Creative Commons Attribution 4 .0 International License. SIGIR ’26, Melbourne, VIC, Australia.  \n© 2026 Copyright held by the owner/author(s) .  \nACM ISBN 979-8-4007-2599-9/2026/07  \n[https://doi.org/10.1145/3805712.3808556](https://doi.org/10.1145/3805712.3808556)  \nSettings Shape the Reported Power of Noise. In Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’26), July 20–24, 2026, Melbourne, VIC, Australia. ACM, New York, NY, USA, 10 pages. [https://doi.org/10.1145/3805712.3808556](https://doi.org/10.1145/3805712.3808556)  \n1 Introduction  \nRetrieval-Augmented Generation (RAG) [13] has become a central method for connecting large language models (LLMs) with current, domain-specific information [5, 24] . By adding retrieved documents to the model prompts, RAG incorporates external knowledge thereby mitigating common LLM limitations such as hallucinations and outdated information [3, 5, 13, 24] . RAG combines retrieval and generation, but its effectiveness depends critically on retrieval quality: irrelevant or incorrect retrieved content can degrad","cbCaij3v8KoFvIdT","https://ap.wps.com/l/cbCaij3v8KoFvIdT","pdf",915971,12,1,10,"English","en",105,"# Introduction\n## Retrieval-Augmented Generation and the Role of Retrieval Quality\n## The Power of Noise and Its Reported Benefits\n# Experimental Reproduction and Extended Settings\n## Original Setup Verification\n## Extensions and Ablation Studies\n# Error Analysis and Robustness Findings\n## Truncation Effects\n## Malformed Generations","[{\"question\":\"What is the “Power of Noise” in retrieval-augmented generation (RAG)?\",\"answer\":\"It is the observation that adding highly unrelated (random) documents to RAG inputs can improve question-answering performance, sometimes outperforming inputs that include only the relevant document.\"},{\"question\":\"How does this paper evaluate whether the effect is robust?\",\"answer\":\"It reproduces prior findings under the original experimental setup and then introduces extensions that vary model choices, instruction prompting, and decoding/output constraints to test sensitivity across settings.\"},{\"question\":\"Why does the “Power of Noise” sometimes weaken or disappear?\",\"answer\":\"The paper shows the pattern is highly sensitive to inference configuration, and error analysis links the variance to truncation and malformed generations that can change model behavior.\"}]","The Powerless Noise: How Experimental Settings Shape the Reported Power of Noise | PDF",1784176809,25,{"code":4,"msg":32,"data":33},"ok",{"site_id":25,"language":24,"slug":34,"title":13,"keywords":35,"description":14,"schema_data":36,"social_meta":88,"head_meta":90,"extra_data":92,"updated_unix":29},"the-powerless-noise-how-experimental-settings-shape-the-reported-power-of-noise","",{"@graph":37,"@context":87},[38,55,70],{"@type":39,"itemListElement":40},"BreadcrumbList",[41,45,49,52],{"item":42,"name":43,"@type":44,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":46,"name":47,"@type":44,"position":48},"https://docshare.wps.com/document/","Document",2,{"item":50,"name":12,"@type":44,"position":51},"https://docshare.wps.com/document/research-report/",3,{"item":53,"name":13,"@type":44,"position":54},"https://docshare.wps.com/document/the-powerless-noise-how-experimental-settings-shape-the-reported-power-of-noise/81877/",4,{"url":53,"name":13,"@type":56,"author":57,"headline":13,"publisher":59,"fileFormat":62,"inLanguage":24,"description":14,"dateModified":63,"datePublished":64,"encodingFormat":62,"isAccessibleForFree":65,"interactionStatistic":66},"DigitalDocument",{"name":9,"@type":58},"Person",{"url":42,"name":60,"@type":61},"DocShare","Organization","application/pdf","2026-08-01","2026-07-16",true,{"@type":67,"interactionType":68,"userInteractionCount":20},"InteractionCounter",{"@type":69},"ViewAction",{"@type":71,"mainEntity":72},"FAQPage",[73,79,83],{"name":74,"@type":75,"acceptedAnswer":76},"What is the “Power of Noise” in retrieval-augmented generation (RAG)?","Question",{"text":77,"@type":78},"It is the observation that adding highly unrelated (random) documents to RAG inputs can improve question-answering performance, sometimes outperforming inputs that include only the relevant document.","Answer",{"name":80,"@type":75,"acceptedAnswer":81},"How does this paper evaluate whether the effect is robust?",{"text":82,"@type":78},"It reproduces prior findings under the original experimental setup and then introduces extensions that vary model choices, instruction prompting, and decoding/output constraints to test sensitivity across settings.",{"name":84,"@type":75,"acceptedAnswer":85},"Why does the “Power of Noise” sometimes weaken or disappear?",{"text":86,"@type":78},"The paper shows the pattern is highly sensitive to inference configuration, and error analysis links the variance to truncation and malformed generations that can change model behavior.","https://schema.org",{"og:url":53,"og:type":89,"og:title":13,"og:site_name":60,"og:description":14},"article",{"robots":91,"canonical":53},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":94},[95,99,103,107,112,117,122,125,130,133,136],{"id":21,"doc_module":4,"doc_module_name":47,"category_name":96,"show_sort_weight":97,"slug":98},"Story & Novel",90,"story-novel",{"id":48,"doc_module":4,"doc_module_name":47,"category_name":100,"show_sort_weight":101,"slug":102},"Literature",80,"literature",{"id":54,"doc_module":4,"doc_module_name":47,"category_name":104,"show_sort_weight":105,"slug":106},"Exam",70,"exam",{"id":108,"doc_module":4,"doc_module_name":47,"category_name":109,"show_sort_weight":110,"slug":111},5,"Comic",60,"comic",{"id":113,"doc_module":4,"doc_module_name":47,"category_name":114,"show_sort_weight":115,"slug":116},6,"Technology",50,"technology",{"id":118,"doc_module":4,"doc_module_name":47,"category_name":119,"show_sort_weight":120,"slug":121},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":47,"category_name":12,"show_sort_weight":123,"slug":124},30,"research-report",{"id":126,"doc_module":4,"doc_module_name":47,"category_name":127,"show_sort_weight":128,"slug":129},9,"Religion & Spirituality",20,"religion-spirituality",{"id":128,"doc_module":4,"doc_module_name":47,"category_name":131,"show_sort_weight":128,"slug":132},"World Cup","world-cup",{"id":22,"doc_module":4,"doc_module_name":47,"category_name":134,"show_sort_weight":22,"slug":135},"Lifestyle","lifestyle",{"id":137,"doc_module":4,"doc_module_name":47,"category_name":138,"show_sort_weight":108,"slug":139},19,"General","general"]