{"id":"https://openalex.org/W7204752422","doi":"https://doi.org/10.48550/arxiv.2608.28011","title":"Coverage, Not Credit: Failure-Credit Routing of Zeroth-Order Perturbation Budgets Does Not Improve On-Pool Sample Efficiency for LLM Agents","display_name":"Coverage, Not Credit: Failure-Credit Routing of Zeroth-Order Perturbation Budgets Does Not Improve On-Pool Sample Efficiency for LLM Agents","publication_year":2026,"publication_date":"2026-08-28","ids":{"openalex":"https://openalex.org/W7204752422","doi":"https://doi.org/10.48550/arxiv.2608.28011"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.28011","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.28011","pdf_url":"https://arxiv.org/pdf/2608.28011","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.28011","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5123818285","display_name":"Yuxu Ge","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I52099693","display_name":"University of York","ror":"https://ror.org/04m01e293","country_code":"GB","type":"education","lineage":["https://openalex.org/I52099693"]}],"countries":["GB"],"is_corresponding":true,"raw_author_name":"Ge, Yuxu","raw_affiliation_strings":["University of York"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of York","institution_ids":["https://openalex.org/I52099693"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5123818285"],"corresponding_institution_ids":["https://openalex.org/I52099693"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.14499999582767487,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.14499999582767487,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10100","display_name":"Metaheuristic Optimization Algorithms Research","score":0.10029999911785126,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.0869000032544136,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/derivative-free-optimization","display_name":"derivative-free optimization","score":0.9010000228881836},{"id":"https://openalex.org/keywords/evolution-strategies","display_name":"evolution strategies","score":0.9559999704360962},{"id":"https://openalex.org/keywords/llm-based-agents","display_name":"LLM agents","score":0.746999979019165},{"id":"https://openalex.org/keywords/tool-use","display_name":"tool use","score":0.7910000085830688},{"id":"https://openalex.org/keywords/sample-efficiency","display_name":"sample efficiency","score":0.625},{"id":"https://openalex.org/keywords/optimization-efficiency","display_name":"optimization efficiency","score":0.34299999475479126}],"concepts":[{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.6894999742507935},{"id":"https://openalex.org/C2779458634","wikidata":"https://www.wikidata.org/wiki/Q24963715","display_name":"Debiasing","level":2,"score":0.5218999981880188},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.49549999833106995},{"id":"https://openalex.org/C177918212","wikidata":"https://www.wikidata.org/wiki/Q803623","display_name":"Perturbation (astronomy)","level":2,"score":0.49160000681877136},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.4180999994277954},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.4020000100135803},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.38429999351501465},{"id":"https://openalex.org/C8505890","wikidata":"https://www.wikidata.org/wiki/Q605095","display_name":"Budget constraint","level":2,"score":0.3813999891281128},{"id":"https://openalex.org/C61445026","wikidata":"https://www.wikidata.org/wiki/Q217608","display_name":"Fixed point","level":2,"score":0.3801000118255615},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.3637000024318695},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.3628000020980835},{"id":"https://openalex.org/C2776175482","wikidata":"https://www.wikidata.org/wiki/Q1195816","display_name":"Transfer (computing)","level":2,"score":0.352400004863739},{"id":"https://openalex.org/C2777363581","wikidata":"https://www.wikidata.org/wiki/Q15098235","display_name":"Harm","level":2,"score":0.31439998745918274},{"id":"https://openalex.org/C74172769","wikidata":"https://www.wikidata.org/wiki/Q1446839","display_name":"Routing (electronic design automation)","level":2,"score":0.31299999356269836},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.30730000138282776},{"id":"https://openalex.org/C85847156","wikidata":"https://www.wikidata.org/wiki/Q59015987","display_name":"Verifiable secret sharing","level":3,"score":0.2994000017642975},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.2935999929904938},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.29319998621940613},{"id":"https://openalex.org/C206729178","wikidata":"https://www.wikidata.org/wiki/Q2271896","display_name":"Scheduling (production processes)","level":2,"score":0.2736000120639801},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.26260000467300415},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.26109999418258667},{"id":"https://openalex.org/C164680029","wikidata":"https://www.wikidata.org/wiki/Q1055293","display_name":"Multiple","level":2,"score":0.2572999894618988},{"id":"https://openalex.org/C104247578","wikidata":"https://www.wikidata.org/wiki/Q4779368","display_name":"Aperiodic graph","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.25279998779296875}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.28011","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.28011","pdf_url":"https://arxiv.org/pdf/2608.28011","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.28011","doi":"https://doi.org/10.48550/arxiv.2608.28011","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.28011","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.28011","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.28011","pdf_url":"https://arxiv.org/pdf/2608.28011","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204752422.pdf","grobid_xml":"https://content.openalex.org/works/W7204752422.grobid-xml"},"referenced_works_count":14,"referenced_works":["https://openalex.org/W2596367596","https://openalex.org/W4392223846","https://openalex.org/W4403443827","https://openalex.org/W4407091248","https://openalex.org/W4414806754","https://openalex.org/W4415336592","https://openalex.org/W4415346771","https://openalex.org/W4416550811","https://openalex.org/W7133241261","https://openalex.org/W7155081671","https://openalex.org/W7160905763","https://openalex.org/W7163665632","https://openalex.org/W7172437080","https://openalex.org/W7202198707"],"related_works":["https://openalex.org/W7203909526","https://openalex.org/W7214007481","https://openalex.org/W7162879191","https://openalex.org/W7171538005","https://openalex.org/W7134812561","https://openalex.org/W7154801379","https://openalex.org/W7164907210","https://openalex.org/W7213898442","https://openalex.org/W7204131022","https://openalex.org/W7164065013"],"abstract_inverted_index":{"Trajectory-level":[0],"credit":[1,23,102],"assignment":[2],"can":[3,255],"localize":[4],"which":[5],"module":[6,111],"of":[7,72],"a":[8,26,33,48,87,158,193,237],"tool-using":[9],"LLM":[10],"agent":[11],"causes":[12],"failures":[13],"using":[14],"only":[15],"verifiable":[16],"signals.":[17],"We":[18,244],"ask":[19],"whether":[20],"such":[21],"failure":[22,252],"should":[24],"route":[25],"fixed":[27,194],"zeroth-order/evolution-strategies":[28],"(ZO/ES)":[29],"perturbation":[30],"budget.":[31],"Across":[32,142],"synthetic":[34],"environment":[35],"and":[36,39,53,94,116,127,135,157,168,249],"frozen":[37,261],"Qwen2.5-1.5B/3B":[38],"SmolLM2-1.7B":[40],"agents,":[41],"three":[42,251],"task":[43,195],"families,":[44],"six":[45,143],"allocation":[46,65],"schemes,":[47],"credit-noise":[49],"sweep,":[50],"paired":[51],"seeds,":[52],"exact":[54],"sign-flip":[55],"tests,":[56],"we":[57],"find":[58],"no":[59],"statistically":[60],"detectable":[61],"improvement":[62],"over":[63],"uniform":[64,85,210],"in":[66,149],"any":[67],"on-pool":[68,234],"comparison":[69],"(no":[70],"gain":[71],"at":[73],"least":[74],"2":[75],"percentage":[76],"points).":[77],"The":[78],"joint":[79],"soft-plus-sigma":[80],"scheme":[81],"is":[82,104,112,147,189,205,226],"equivalent":[83,106],"to":[84,131,241],"within":[86],"+/-":[88],"0.02":[89],"AUC":[90,133],"margin":[91],"on":[92,100,107,119,138,192,211,260],"1.5B":[93],"3B;":[95],"concentrating":[96],"the":[97,101,113,139,201],"full":[98],"budget":[99],"argmax":[103],"marginally":[105],"1.5B,":[108],"where":[109],"that":[110,206,227,254],"verified":[114],"bottleneck,":[115],"significantly":[117],"worse":[118],"3B.":[120],"Inverse-propensity":[121],"debiasing":[122],"does":[123],"not":[124],"rescue":[125],"routing,":[126],"misrouting":[128],"costs":[129],"up":[130],"-0.074":[132],"in-house":[134],"-0.118":[136],"end-to-end":[137],"BFCL-derived":[140],"family.":[141],"fixed-step":[144],"schedules,":[145],"loss":[146],"linear":[148],"bottleneck":[150],"starvation":[151],"rate":[152],"(R^2":[153],"=":[154,216,219],"0.94,":[155],"descriptive),":[156],"preregistered":[159],"credit-free":[160],"coverage":[161],"floor":[162],"removes":[163],"detected":[164],"harm.":[165],"Matched-budget":[166],"burst":[167],"step-compensating":[169],"catch-up":[170],"schedules":[171],"are":[172],"consistent":[173],"with":[174],"harm":[175],"arising":[176],"from":[177],"insufficient":[178],"cumulative":[179],"parameter":[180],"movement":[181],"rather":[182],"than":[183],"update":[184],"frequency.":[185],"Our":[186],"primary":[187],"estimand":[188],"optimization":[190],"efficiency":[191],"pool.":[196],"On":[197],"unseen":[198],"BFCL":[199],"functions,":[200],"study's":[202],"one":[203],"exception":[204,247],"soft":[207],"routing":[208],"exceeds":[209],"held-out":[212],"endpoints":[213],"(+0.047,":[214],"p":[215],"0.031,":[217],"n":[218],"6).":[220],"A":[221],"plausible":[222],"but":[223],"untested":[224],"reading":[225],"routing-favored":[228],"caller":[229],"improvements":[230],"transfer":[231],"while":[232],"uniform's":[233],"gains":[235],"reflect":[236],"synthesizer":[238],"behavior":[239],"specific":[240],"our":[242],"harness.":[243],"report":[245],"this":[246],"explicitly":[248],"document":[250],"modes":[253],"silently":[256],"invalidate":[257],"ZO/ES":[258],"experiments":[259],"LLMs.":[262]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-01T00:00:00"}