Systematic Ablation Reveals Hidden Failures in Multi-Agent AI for Science

Systematic Ablation Reveals Hidden Failures in Multi-Agent AI for Science | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Systematic Ablation Reveals Hidden Failures in Multi-Agent AI for Science VALERIO BIANCHI, Dirkjan Schokker This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9366108/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Retrieval-augmented generation (RAG) systems increasingly ground large language models (LLMs) in scientific literature, yet their design choices are typically made by convention rather than evidence. Here, we introduce a systematic ablation methodology for RAG pipeline design, validated through triple triangulation of deterministic ground-truth metrics, LLM judge scoring, and natural language inference. Across more than 36,000 evaluations spanning 200 scientific papers and 250 expert-curated questions, we find that self-correction without retrieval grounding degrades 79% of correct answers. Iterative retrieval saturates logistically with depth, capturing 80% of the improvement within seven turns. LLM judge faithfulness scoring also exhibits a construct validity failure, where answers rated perfectly faithful have only 21% of their claims actually grounded in sources. These findings, replicated across 3,300 code generation evaluations, demonstrate that reliable AI for science demands not just better models, but disciplined evaluation methodology that exposes failure modes invisible to any single metric. Physical sciences/Mathematics and computing/Computer science Physical sciences/Mathematics and computing/Scientific data Physical sciences/Mathematics and computing/Computational science Full Text Additional Declarations There is NO Competing Interest. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9366108","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":620357171,"identity":"4a91d1ae-e0ea-4643-9c46-49b79d877cf2","order_by":0,"name":"VALERIO BIANCHI","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA40lEQVRIiWNgGAWjYFAC5gYGBgMgYgayGRsk5CCiBfi0MKJqMYaIGhDSAlMDZCcicbEDg+MHGx9XFNgwmLMzP3vwc4dF+oYbuQ+YC/BpOZPYbHjGII3BspnN3LD3jETuhhvpBswz8GiRbEhsk2wwOMwARGbSjG0SudtupDEw8+DT0v+w/SdEC/s3kJZ0M0Ja+CUS2xghWnjAtiQQoeVhM9BhaTyWzTxlkr1tEob7zzxjOIxPCxt/8sGPDX9s5Mz5j2+T+NlWJy/Znsb4mKcCtxYY4EHhHSCsYRSMglEwCkYBPgAAGYFHErondcgAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0001-9025-7923","institution":"Wageningen Bioveterinary Research","correspondingAuthor":true,"prefix":"","firstName":"VALERIO","middleName":"","lastName":"BIANCHI","suffix":""},{"id":620357172,"identity":"ca867a6a-52f0-4c87-b58d-e24de417350f","order_by":1,"name":"Dirkjan Schokker","email":"","orcid":"","institution":"Wageningen Bioveterinary Research","correspondingAuthor":false,"prefix":"","firstName":"Dirkjan","middleName":"","lastName":"Schokker","suffix":""}],"badges":[],"createdAt":"2026-04-09 09:19:53","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9366108/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9366108/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":109205202,"identity":"c3cda300-3dc7-43a2-9409-b5dd437bc8e0","added_by":"auto","created_at":"2026-05-13 15:03:45","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3951248,"visible":true,"origin":"","legend":"Article File","description":"","filename":"paperdraftNMI042026VBcombined.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9366108/v1_covered_4b2d06c5-bace-44d9-86e3-05a013aea3b2.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Systematic Ablation Reveals Hidden Failures in Multi-Agent AI for Science","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9366108/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9366108/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Retrieval-augmented generation (RAG) systems increasingly ground large language models (LLMs) in scientific literature, yet their design choices are typically made by convention rather than evidence. Here, we introduce a systematic ablation methodology for RAG pipeline design, validated through triple triangulation of deterministic ground-truth metrics, LLM judge scoring, and natural language inference. Across more than 36,000 evaluations spanning 200 scientific papers and 250 expert-curated questions, we find that self-correction without retrieval grounding degrades 79% of correct answers. Iterative retrieval saturates logistically with depth, capturing 80% of the improvement within seven turns. LLM judge faithfulness scoring also exhibits a construct validity failure, where answers rated perfectly faithful have only 21% of their claims actually grounded in sources. These findings, replicated across 3,300 code generation evaluations, demonstrate that reliable AI for science demands not just better models, but disciplined evaluation methodology that exposes failure modes invisible to any single metric.","manuscriptTitle":"Systematic Ablation Reveals Hidden Failures in Multi-Agent AI for Science","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-12 13:47:59","doi":"10.21203/rs.3.rs-9366108/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ce9b7048-bdc0-467d-9e0b-c0681194b467","owner":[],"postedDate":"May 12th, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Reject before peer review","date":"2026-05-13T07:46:49+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":66019428,"name":"Physical sciences/Mathematics and computing/Computer science"},{"id":66019429,"name":"Physical sciences/Mathematics and computing/Scientific data"},{"id":66019430,"name":"Physical sciences/Mathematics and computing/Computational science"}],"tags":[],"updatedAt":"2026-05-13T07:50:58+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-12 13:47:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9366108","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9366108","identity":"rs-9366108","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}