Who Broke the Pipeline? Traceability andAccountability in Multi-Agent LLM SoftwareEngineering Pipelines

Who Broke the Pipeline? Traceability andAccountability in Multi-Agent LLM SoftwareEngineering Pipelines | 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 Research Article Who Broke the Pipeline? Traceability andAccountability in Multi-Agent LLM SoftwareEngineering Pipelines Jackson Beem, Amine Barrak, Emna Ksontini This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9350097/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Large language models are increasingly embedded across the DevOps pipeline, from planning and code generation to testing and deployment, yet multi-agent LLM pipelines remain opaque: errors propagate silently across stages with no systematic method to attribute failures to specific agents. We present a two-part study of traceable, accountable multi-agent pipelines with structured handoffs and role-level blame attribution. Part I evaluates a Planner, Executor, and Critic pipeline on three multiple-choice benchmarks using eight configurations of three frontier LLMs. Structured accountability recovers accuracy dramatically over unstructured pipelines (e.g., from 61.42% to 97.64%), and blame attribution reveals complementary role-specific aptitudes that enable data-driven model to role casting. Part II extends the pipeline to execution-grounded code generation on HumanEval, MBPP, and BigCodeBench, introducing a fourth Verifier agent. We find that LLM-based test evaluation is not a reliable substitute for actual test execution: false-fail rates reach 93%, triggering unnecessary re-generation that overwrites correct solutions. Replacing the LLM Critic with an execution-based verification harness eliminates verification harm entirely, achieves the highest accuracy on complex tasks, and reduces token cost by up to 57%. Test execution remains an irreplaceable component of the DevOps pipeline that cannot be delegated to LLM judgment alone. We further find that the Planner adds 1.8 to 3.0 times more tokens to the pipeline without meaningful accuracy gains in code generation, as the Executor produces comparable output with or without the Planner's formal specification. Multi-agent LLM pipelines Role specialization Accountability Traceability Code generation Execution-based verification Pipeline design Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 16 Apr, 2026 Reviewers invited by journal 14 Apr, 2026 Editor assigned by journal 08 Apr, 2026 Submission checks completed at journal 08 Apr, 2026 First submitted to journal 07 Apr, 2026 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-9350097","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":624086245,"identity":"e8eacbb4-f3a5-409a-8d08-0988b5d14642","order_by":0,"name":"Jackson Beem","email":"","orcid":"","institution":"Oakland University","correspondingAuthor":false,"prefix":"","firstName":"Jackson","middleName":"","lastName":"Beem","suffix":""},{"id":624086246,"identity":"7dc54434-77c8-4799-85f8-f4533c0bb9cc","order_by":1,"name":"Amine Barrak","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/ElEQVRIiWNgGAWjYFACxgYQAgI2xgcMDAkyQJYBkEOcFmag0gQeIrRAdIG0sEkQpYV/2uG2Bx932OWbtx9Lq/hQk8bDP7t5A8OHssM4tUjcTmw3nHkm2XLOmbRjN2ccy+GRuHOsgHHGOdxaGG4ntknztjEbSDCkt93mYavgYbiRY8DM24ZbizxIy9+2egMJ/udtxTz/KnjkQVr+4tFiANLC2HbYQEIi7RjQ8BweA5AWRjxaDIFaJHvPHAdqeZYsObMvjcfwRlrBwZ5z6Ti1yN1Ofybxc0c10GFphh8+fEuWk7uRvPHBjzJr3N7HCg6QqH4UjIJRMApGARoAANdBVz8i4pthAAAAAElFTkSuQmCC","orcid":"","institution":"Oakland University","correspondingAuthor":true,"prefix":"","firstName":"Amine","middleName":"","lastName":"Barrak","suffix":""},{"id":624086247,"identity":"d74dcfae-7455-4d55-9d4b-a6c7911275a1","order_by":2,"name":"Emna Ksontini","email":"","orcid":"","institution":"University of North Carolina Wilmington","correspondingAuthor":false,"prefix":"","firstName":"Emna","middleName":"","lastName":"Ksontini","suffix":""}],"badges":[],"createdAt":"2026-04-08 01:23:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9350097/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9350097/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107489527,"identity":"9c88245f-a501-484b-b131-e0b147d837b7","added_by":"auto","created_at":"2026-04-22 02:47:58","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1945678,"visible":true,"origin":"","legend":"","description":"","filename":"ASEJournalExtension.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9350097/v1_covered_b0a9fc3f-e309-4ef9-ac12-ecc2f34547ca.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Who Broke the Pipeline? Traceability andAccountability in Multi-Agent LLM SoftwareEngineering Pipelines","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"automated-software-engineering","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ause","sideBox":"Learn more about [Automated Software Engineering](http://link.springer.com/journal/10515)","snPcode":"10515","submissionUrl":"https://submission.nature.com/new-submission/10515/3","title":"Automated Software Engineering","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Multi-agent LLM pipelines, Role specialization, Accountability, Traceability, Code generation, Execution-based verification, Pipeline design","lastPublishedDoi":"10.21203/rs.3.rs-9350097/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9350097/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Large language models are increasingly embedded across the DevOps pipeline, from planning and code generation to testing and deployment, yet multi-agent LLM pipelines remain opaque: errors propagate silently across stages with no systematic method to attribute failures to specific agents. We present a two-part study of traceable, accountable multi-agent pipelines with structured handoffs and role-level blame attribution. Part I evaluates a Planner, Executor, and Critic pipeline on three multiple-choice benchmarks using eight configurations of three frontier LLMs. Structured accountability recovers accuracy dramatically over unstructured pipelines (e.g., from 61.42% to 97.64%), and blame attribution reveals complementary role-specific aptitudes that enable data-driven model to role casting. Part II extends the pipeline to execution-grounded code generation on HumanEval, MBPP, and BigCodeBench, introducing a fourth Verifier agent. We find that LLM-based test evaluation is not a reliable substitute for actual test execution: false-fail rates reach 93%, triggering unnecessary re-generation that overwrites correct solutions. Replacing the LLM Critic with an execution-based verification harness eliminates verification harm entirely, achieves the highest accuracy on complex tasks, and reduces token cost by up to 57%. Test execution remains an irreplaceable component of the DevOps pipeline that cannot be delegated to LLM judgment alone. We further find that the Planner adds 1.8 to 3.0 times more tokens to the pipeline without meaningful accuracy gains in code generation, as the Executor produces comparable output with or without the Planner's formal specification.","manuscriptTitle":"Who Broke the Pipeline? Traceability andAccountability in Multi-Agent LLM SoftwareEngineering Pipelines","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-21 16:30:24","doi":"10.21203/rs.3.rs-9350097/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"159197768027771789413173612368187144111","date":"2026-04-16T06:40:13+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-14T06:37:31+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-08T18:30:08+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-08T16:14:21+00:00","index":"","fulltext":""},{"type":"submitted","content":"Automated Software Engineering","date":"2026-04-08T01:08:41+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"automated-software-engineering","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ause","sideBox":"Learn more about [Automated Software Engineering](http://link.springer.com/journal/10515)","snPcode":"10515","submissionUrl":"https://submission.nature.com/new-submission/10515/3","title":"Automated Software Engineering","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"b15c2893-f1b6-435d-8d93-eda246e1ab0b","owner":[],"postedDate":"April 21st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-21T16:30:24+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-21 16:30:24","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9350097","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9350097","identity":"rs-9350097","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}