{"id":1190632,"date":"2026-10-21T10:26:00","date_gmt":"2026-10-21T17:26:00","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/adaptive-gepa-make-your-harness-fit-heterogeneous-requests\/"},"modified":"2026-10-22T08:40:48","modified_gmt":"2026-10-22T15:40:48","slug":"adaptive-gepa-make-your-harness-fit-heterogeneous-requests","status":"publish","type":"msr-research-item","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/publication\/adaptive-gepa-make-your-harness-fit-heterogeneous-requests\/","title":{"rendered":"Adaptive-GEPA: Make Your Harness Fit Heterogeneous Requests"},"content":{"rendered":"\n\n\n<p class=\"wp-block-paragraph\">Reflective optimizers such as GEPA improve language model prompts from execution traces and evaluator feedback; full-program extensions can also rewrite tools and control flow. In practice, a user hands the same endpoint heterogeneous requests whose effective solutions require different tools, reasoning modes, and control flow. Optimizing one shared program leaves this division of work implicit in source-code search, while optimizing a separate program per request family fixes it beforehand. We introduce Adaptive-GEPA, which learns both how to divide requests and how to solve them. It evolves a router and a library of specialist programs under one search budget. The router&#8217;s instructions, each specialist&#8217;s description, and its program code are plain, human-readable text, edited from feedback. To combine branches, it aligns specialists by the requests they handle and inherits descriptions together with programs. On a fixed mixture of four task families, the reported Qwen3-8B run evolves four experts without supplying family labels to the router or reflection model; its routing matches the task partition on all 651 test requests. Its family-mean test score (x100) rises from 52.6 to 70.6, compared with 62.5 for GEPA&#8217;s full-program adapter and 54.0 for GRPO at a nominal budget of 18,000 scored calls. These counts do not equate total compute. Figure 1 summarizes the learning curves, final test scores, and routing agreement.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Reflective optimizers such as GEPA improve language model prompts from execution traces and evaluator feedback; full-program extensions can also rewrite tools and control flow. In practice, a user hands the same endpoint heterogeneous requests whose effective solutions require different tools, reasoning modes, and control flow. Optimizing one shared program leaves this division of work implicit [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":[{"type":"text","value":"Tianyu Chen","user_id":0},{"type":"text","value":"Yasi Zhang","user_id":0},{"type":"text","value":"Ruiyi Wang","user_id":0},{"type":"text","value":"Xinran Zhao","user_id":0},{"type":"text","value":"Taoran Li","user_id":0},{"type":"text","value":"Mingyuan Zhou","user_id":0}],"msr_publishername":"","msr_publisher_other":"","msr_booktitle":"","msr_chapter":"","msr_edition":"","msr_editors":"","msr_how_published":"arXiv","msr_isbn":"","msr_issue":"","msr_journal":"","msr_number":"","msr_organization":"","msr_pages_string":"","msr_page_range_start":"","msr_page_range_end":"","msr_series":"","msr_volume":"","msr_copyright":"","msr_conference_name":"","msr_doi":"","msr_arxiv_id":"2609.38762","msr_mag_id":"","msr_other_authors":"","msr_other_contributors":"","msr_speaker":"","msr_award":"","msr_affiliation":"","msr_institution":"","msr_host":"","msr_version":"","msr_duration":"","msr_release_tracker_id":"","msr_highlight_type":"","msr_date_display_format":"","msr_main_download_label":"","msr_external_link_label":"","msr_doi_label":"","msr_published_date":"2026-09-30","msr_startdate":"","msr_presentation_date":"","msr_highlight_text":"","msr_notes":"","msr_longbiography":"","msr_publicationurl":"","msr_external_url":"","msr_secondary_video_url":"","msr_conference_url":"","msr_journal_url":"","msr_year":2026,"msr_month":9,"msr_day":30,"msr_microsoftintellectualproperty":false,"msr_pub_id":"fa6f37a97ae12af343d3867bb6735d2d9d59cf7c","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2609.38762","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_original_fields_of_study":[],"msr_s2_paper_id":"fa6f37a97ae12af343d3867bb6735d2d9d59cf7c","msr_s2_pdf_url":"","msr_citation_count_updated":"","msr_citation_count":0,"msr_influential_citations":0,"msr_reference_count":21,"msr_s2_open_access":false,"msr_s2_author_ids":[],"msr_pub_ids":[{"provider":"s2","id":"fa6f37a97ae12af343d3867bb6735d2d9d59cf7c"},{"provider":"arxiv","id":"2609.38762"},{"provider":"corpusid","id":"292477109"}],"msr_hide_image_in_river":0,"footnotes":""},"msr-research-highlight":[],"research-area":[13556,13560],"msr-publication-type":[270373],"msr-publisher":[],"msr-publication-cta":[],"msr-focus-area":[],"msr-locale":[268875],"msr-post-option":[],"msr-field-of-study":[246691,265728],"msr-conference":[],"msr-journal":[],"msr-impact-theme":[],"msr-pillar":[],"class_list":["post-1190632","msr-research-item","type-msr-research-item","status-publish","hentry","msr-research-area-artificial-intelligence","msr-research-area-programming-languages-software-engineering","msr-locale-en_us","msr-field-of-study-computer-science","msr-field-of-study-software-engineering-15"],"msr_publishername":"","msr_edition":"","msr_affiliation":"","msr_published_date":"2026-09-30","msr_host":"","msr_duration":"","msr_version":"","msr_speaker":"","msr_other_contributors":"","msr_booktitle":"","msr_pages_string":"","msr_chapter":"","msr_isbn":"","msr_journal":"","msr_volume":"","msr_number":"","msr_editors":"","msr_series":"","msr_issue":"","msr_organization":"","msr_how_published":"arXiv","msr_notes":"","msr_highlight_text":"","msr_release_tracker_id":"","msr_original_fields_of_study":"","msr_download_urls":"","msr_external_url":"","msr_secondary_video_url":"","msr_longbiography":"","msr_microsoftintellectualproperty":0,"msr_main_download":"","msr_publicationurl":"","msr_doi":"","msr_publication_uploader":[{"type":"url","title":"https:\/\/arxiv.org\/abs\/2609.38762","label_id":243109,"id":false,"viewUrl":false}],"msr_related_uploader":[],"msr_citation_count":0,"msr_citation_count_updated":"","msr_s2_paper_id":"fa6f37a97ae12af343d3867bb6735d2d9d59cf7c","msr_influential_citations":0,"msr_reference_count":21,"msr_arxiv_id":"2609.38762","msr_s2_author_ids":[],"msr_s2_open_access":false,"msr_s2_pdf_url":null,"msr_attachments":[],"msr-author-ordering":[{"type":"text","value":"Tianyu Chen","user_id":0,"rest_url":false},{"type":"text","value":"Yasi Zhang","user_id":0,"rest_url":false},{"type":"text","value":"Ruiyi Wang","user_id":0,"rest_url":false},{"type":"text","value":"Xinran Zhao","user_id":0,"rest_url":false},{"type":"text","value":"Taoran Li","user_id":0,"rest_url":false},{"type":"text","value":"Mingyuan Zhou","user_id":0,"rest_url":false}],"msr_impact_theme":[],"msr_research_lab":[],"msr_event":[],"msr_group":[],"msr_project":[],"publication":[],"video":[],"msr-tool":[],"msr_publication_type":"misc","related_content":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190632","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-research-item"}],"version-history":[{"count":4,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190632\/revisions"}],"predecessor-version":[{"id":1190803,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-item\/1190632\/revisions\/1190803"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1190632"}],"wp:term":[{"taxonomy":"msr-research-highlight","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-research-highlight?post=1190632"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1190632"},{"taxonomy":"msr-publication-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-type?post=1190632"},{"taxonomy":"msr-publisher","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publisher?post=1190632"},{"taxonomy":"msr-publication-cta","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-publication-cta?post=1190632"},{"taxonomy":"msr-focus-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-focus-area?post=1190632"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1190632"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1190632"},{"taxonomy":"msr-field-of-study","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-field-of-study?post=1190632"},{"taxonomy":"msr-conference","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-conference?post=1190632"},{"taxonomy":"msr-journal","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-journal?post=1190632"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1190632"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1190632"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}