{"id":"https://openalex.org/W7204223279","doi":"https://doi.org/10.20944/preprints202608.2051.v1","title":"The Winner\u2019s Curse in LLM Pipeline Selection: How Correlated Candidate Pools Break Best-of-N Reporting, and How to Correct It","display_name":"The Winner\u2019s Curse in LLM Pipeline Selection: How Correlated Candidate Pools Break Best-of-N Reporting, and How to Correct It","publication_year":2026,"publication_date":"2026-08-28","ids":{"openalex":"https://openalex.org/W7204223279","doi":"https://doi.org/10.20944/preprints202608.2051.v1"},"language":"en","primary_location":{"id":"doi:10.20944/preprints202608.2051.v1","doi":"https://doi.org/10.20944/preprints202608.2051.v1","is_oa":true,"landing_page_url":"https://doi.org/10.20944/preprints202608.2051.v1","pdf_url":null,"source":{"id":"https://openalex.org/S6309402219","display_name":"Preprints.org","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"},"volume":null,"type":"preprint","indexed_in":["crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.20944/preprints202608.2051.v1","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5146560673","display_name":"Junior Charlie","orcid":"https://orcid.org/0009-0009-5252-3202","observed_orcids":["https://orcid.org/0009-0009-5252-3202"]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Junior Charlie","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0009-0009-5252-3202","affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5146560673"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"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/T10028","display_name":"Topic Modeling","score":0.7764000296592712,"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/T10028","display_name":"Topic Modeling","score":0.7764000296592712,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.05270000174641609,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.04179999977350235,"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/winner-s-curse","display_name":"winner's curse","score":0.9869999885559082},{"id":"https://openalex.org/keywords/cross-fitting","display_name":"cross-fitting","score":0.7990000247955322},{"id":"https://openalex.org/keywords/budget-allocation","display_name":"budget allocation","score":0.5730000138282776},{"id":"https://openalex.org/keywords/extreme-value-theory","display_name":"extreme value theory","score":0.3659999966621399}],"concepts":[{"id":"https://openalex.org/C2777894999","wikidata":"https://www.wikidata.org/wiki/Q4781758","display_name":"Approx","level":2,"score":0.7705000042915344},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.560699999332428},{"id":"https://openalex.org/C77553402","wikidata":"https://www.wikidata.org/wiki/Q13222579","display_name":"Upper and lower bounds","level":2,"score":0.5485000014305115},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.4984000027179718},{"id":"https://openalex.org/C2780273121","wikidata":"https://www.wikidata.org/wiki/Q109411","display_name":"Curse","level":2,"score":0.4909999966621399},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.4853000044822693},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.48100000619888306},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.46480000019073486},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.44110000133514404},{"id":"https://openalex.org/C2778067643","wikidata":"https://www.wikidata.org/wiki/Q166507","display_name":"Interval (graph theory)","level":2,"score":0.43619999289512634},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4052000045776367},{"id":"https://openalex.org/C111030470","wikidata":"https://www.wikidata.org/wiki/Q1430460","display_name":"Curse of dimensionality","level":2,"score":0.37400001287460327},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3707999885082245},{"id":"https://openalex.org/C139945424","wikidata":"https://www.wikidata.org/wiki/Q1940696","display_name":"Mean squared error","level":2,"score":0.3547999858856201},{"id":"https://openalex.org/C1893757","wikidata":"https://www.wikidata.org/wiki/Q3653001","display_name":"Inversion (geology)","level":3,"score":0.3546000123023987},{"id":"https://openalex.org/C114614502","wikidata":"https://www.wikidata.org/wiki/Q76592","display_name":"Combinatorics","level":1,"score":0.33820000290870667},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.32409998774528503},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.3068999946117401},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2646999955177307},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.2630999982357025},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.26260000467300415}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.20944/preprints202608.2051.v1","doi":"https://doi.org/10.20944/preprints202608.2051.v1","is_oa":true,"landing_page_url":"https://doi.org/10.20944/preprints202608.2051.v1","pdf_url":null,"source":{"id":"https://openalex.org/S6309402219","display_name":"Preprints.org","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"},{"id":"doi:10.5281/zenodo.22110444","doi":"https://doi.org/10.5281/zenodo.22110444","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.22110444","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"},{"id":"doi:10.5281/zenodo.22110445","doi":"https://doi.org/10.5281/zenodo.22110445","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.22110445","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.20944/preprints202608.2051.v1","doi":"https://doi.org/10.20944/preprints202608.2051.v1","is_oa":true,"landing_page_url":"https://doi.org/10.20944/preprints202608.2051.v1","pdf_url":null,"source":{"id":"https://openalex.org/S6309402219","display_name":"Preprints.org","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Best-of-N":[0],"pipeline":[1],"selection,":[2],"running":[3],"N":[4,81,95,106,341],"candidate":[5,119,163],"prompts":[6,189],"on":[7,45,53,64,112,233,250,260,299,344,359,361],"an":[8,22],"m-item":[9],"dev":[10,16,31,47,216],"set":[11],"and":[12,26,49,75,86,96,136,229,238,253],"reporting":[13,276],"the":[14,19,35,46,125,141,246,265,281,285,351,356,379],"winner's":[15,61],"score,":[17],"is":[18,33,56],"default":[20],"way":[21],"LLM":[23],"practitioner":[24],"picks":[25],"reports":[27],"a":[28,132,137,147,167,209,214,230,290,304,314,322,335,362],"prompt.":[29],"That":[30],"score":[32],"optimistic:":[34],"winner":[36,211],"was":[37],"chosen":[38],"in":[39,308,378],"part":[40],"because":[41],"it":[42,368],"got":[43],"lucky":[44],"items,":[48],"its":[50,296],"true":[51],"accuracy":[52,103,349],"fresh":[54],"data":[55],"lower.":[57],"We":[58,114,203],"measure":[59],"this":[60,122],"curse":[62],"directly":[63],"four":[65],"correctness":[66,316],"matrices":[67,283],"spanning":[68],"three":[69,201,301],"tasks":[70,345],"(SST-2,":[71],"Subj,":[72],"AG":[73],"News)":[74],"two":[76,129,218],"models":[77],"(Qwen2.5-1.5B-Instruct,":[78],"Llama-3.2-3B-Instruct),":[79],"sweeping":[80],"from":[82,88],"1":[83],"to":[84,90,154,207,223,226,272,371],"100":[85,111],"m":[87,109],"20":[89],"200.":[91],"Bias":[92],"grows":[93],"with":[94,364],"shrinks":[97],"roughly":[98,152],"as":[99,321,375],"1/sqrt(m),":[100],"reaching":[101],"0.088":[102],"points":[104],"at":[105,278,284],"=":[107,110,340],"100,":[108],"Subj.":[113],"then":[115],"characterize":[116],"why":[117],"real":[118,365],"pools":[120],"inflate":[121],"bias":[123,298],"beyond":[124],"independent-candidates":[126,247],"textbook":[127],"case:":[128],"effective-candidate-count":[130],"estimators,":[131],"spectral":[133],"participation":[134],"ratio":[135],"moment-matching":[138],"inversion":[139],"of":[140,289],"observed":[142],"selection":[143],"gap,":[144],"agree":[145],"that":[146,159],"105-to-111-candidate":[148],"pool":[149],"behaves":[150],"like":[151],"1.5":[153],"3.5":[155],"independent":[156],"draws.":[157],"Breaking":[158],"count":[160],"down":[161],"by":[162],"provenance":[164],"turns":[165],"up":[166],"result":[168],"opposite":[169],"our":[170],"own":[171],"working":[172],"hypothesis:":[173],"LLM-generated":[174],"(APE-style)":[175],"candidates":[176,192],"carry":[177],"far":[178],"less":[179],"effective":[180],"diversity":[181],"(N_eff/K":[182,193],"~":[183,194],"0.017-0.041)":[184],"than":[185,275],"either":[186],"hand-written":[187],"seed":[188],"or":[190,225,255],"paraphrase/mutation":[191],"0.14-0.24":[195],"for":[196,325,334],"both),":[197],"consistently":[198],"across":[199],"all":[200,300],"tasks.":[202],"compare":[204],"six":[205],"ways":[206],"report":[208],"best-of-N":[210],"(no":[212],"correction,":[213],"naive":[215,266],"split,":[217],"extreme-value":[219],"union-bound":[220],"corrections":[221],"keyed":[222],"K":[224],"N_eff,":[227],"cross-fitting,":[228],"bootstrap":[231],"plug-in)":[232],"bias,":[234],"mean":[235,261],"squared":[236,262],"error,":[237,263],"90%":[239,305],"interval":[240],"coverage;":[241],"correcting":[242],"against":[243],"N_eff":[244],"dominates":[245],"union":[248],"bound":[249],"every":[251,257,309],"axis":[252],"matches":[254],"beats":[256],"other":[258],"method":[259],"while":[264],"K-based":[267],"correction":[268],"overshoots":[269],"badly":[270],"enough":[271],"be":[273],"worse":[274],"nothing":[277],"all.":[279],"Resampling":[280],"stored":[282],"exact":[286],"(N,":[287],"m)":[288],"real,":[291],"live":[292,327],"EvoPromptGA":[293],"trajectory":[294],"reproduces":[295],"measured":[297],"tasks,":[302],"inside":[303],"predictive":[306],"band":[307],"case,":[310],"which":[311],"supports":[312],"treating":[313],"one-time":[315],"matrix":[317],"plus":[318],"free":[319],"resampling":[320],"faithful":[323],"stand-in":[324],"repeated":[326],"search.":[328],"Finally,":[329],"we":[330],"derive":[331],"budget-allocation":[332],"curves":[333],"fixed":[336],"evaluation":[337],"budget":[338,358],"B":[339],"*":[342],"m:":[343],"already":[346],"near":[347],"their":[348],"ceiling,":[350],"optimal":[352],"split":[353],"spends":[354],"almost":[355],"whole":[357],"m;":[360],"task":[363],"headroom":[366],"left,":[367],"keeps":[369],"paying":[370],"search":[372],"wider":[373],"even":[374],"N_eff/K":[376],"stays":[377],"low":[380],"single":[381],"digits.":[382]},"counts_by_year":[],"updated_date":"2026-10-08T09:06:37.294493","created_date":"2026-08-27T00:00:00"}