{"id":"https://openalex.org/W7202347769","doi":"https://doi.org/10.48550/arxiv.2608.11560","title":"When Offline Evaluation Misleads: A Diagnostic Protocol for Reward and Policy Selection in Delayed-Feedback Contextual Bandits","display_name":"When Offline Evaluation Misleads: A Diagnostic Protocol for Reward and Policy Selection in Delayed-Feedback Contextual Bandits","publication_year":2026,"publication_date":"2026-08-12","ids":{"openalex":"https://openalex.org/W7202347769","doi":"https://doi.org/10.48550/arxiv.2608.11560"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.11560","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.11560","pdf_url":"https://arxiv.org/pdf/2608.11560","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.11560","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101664628","display_name":"Sang Su Lee","orcid":"https://orcid.org/0000-0001-9076-3850","observed_orcids":["https://orcid.org/0000-0001-9076-3850"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lee, Sang Su","raw_affiliation_strings":["Thumbtack, Inc. San Francisco , CA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Thumbtack, Inc. San Francisco , CA , USA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075536713","display_name":"V. Loganathan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Loganathan, Vineeth","raw_affiliation_strings":["Thumbtack, Inc. San Francisco , CA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Thumbtack, Inc. San Francisco , CA , USA","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5147082978","display_name":"Shishir Dash","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dash, Shishir","raw_affiliation_strings":["Thumbtack, Inc. San Francisco , CA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Thumbtack, Inc. San Francisco , CA , USA","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5102663286","display_name":"Raghavan Vijay","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raghavan, Vijay","raw_affiliation_strings":["Thumbtack, Inc. San Francisco , CA , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Thumbtack, Inc. San Francisco , CA , USA","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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/T12101","display_name":"Advanced Bandit Algorithms Research","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T12101","display_name":"Advanced Bandit Algorithms Research","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.0,"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/T10203","display_name":"Recommender Systems and Techniques","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/contextual-bandits","display_name":"contextual bandits","score":0.972000002861023},{"id":"https://openalex.org/keywords/delayed-feedback","display_name":"delayed feedback","score":0.984000027179718},{"id":"https://openalex.org/keywords/offline-evaluation","display_name":"offline evaluation","score":0.8640000224113464},{"id":"https://openalex.org/keywords/off-policy-evaluation","display_name":"off-policy evaluation","score":0.7039999961853027},{"id":"https://openalex.org/keywords/personalization","display_name":"personalization","score":0.5649999976158142},{"id":"https://openalex.org/keywords/reward-alignment","display_name":"reward alignment","score":0.375},{"id":"https://openalex.org/keywords/learnability","display_name":"learnability","score":0.8330000042915344},{"id":"https://openalex.org/keywords/push-notifications","display_name":"push notifications","score":0.4259999990463257}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.730400025844574},{"id":"https://openalex.org/C183003079","wikidata":"https://www.wikidata.org/wiki/Q1000371","display_name":"Personalization","level":2,"score":0.557699978351593},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.5311999917030334},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.487199991941452},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.46939998865127563},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.45989999175071716},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4426000118255615},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.41119998693466187},{"id":"https://openalex.org/C73602740","wikidata":"https://www.wikidata.org/wiki/Q7795822","display_name":"Thompson sampling","level":3,"score":0.3856000006198883},{"id":"https://openalex.org/C2778738651","wikidata":"https://www.wikidata.org/wiki/Q16546687","display_name":"Novelty","level":2,"score":0.3774999976158142},{"id":"https://openalex.org/C123197309","wikidata":"https://www.wikidata.org/wiki/Q2882343","display_name":"Multi-armed bandit","level":3,"score":0.36000001430511475},{"id":"https://openalex.org/C2780992000","wikidata":"https://www.wikidata.org/wiki/Q17016113","display_name":"Generator (circuit theory)","level":3,"score":0.3427000045776367},{"id":"https://openalex.org/C2780102126","wikidata":"https://www.wikidata.org/wiki/Q10928179","display_name":"Online and offline","level":2,"score":0.3319999873638153},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.32440000772476196},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.3001999855041504},{"id":"https://openalex.org/C2778067643","wikidata":"https://www.wikidata.org/wiki/Q166507","display_name":"Interval (graph theory)","level":2,"score":0.2928999960422516},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.28859999775886536},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.2838999927043915},{"id":"https://openalex.org/C2780490138","wikidata":"https://www.wikidata.org/wiki/Q7079636","display_name":"Offline learning","level":3,"score":0.2800000011920929},{"id":"https://openalex.org/C97399411","wikidata":"https://www.wikidata.org/wiki/Q825367","display_name":"Coin flipping","level":2,"score":0.26669999957084656},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.25119999051094055}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.11560","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.11560","pdf_url":"https://arxiv.org/pdf/2608.11560","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.2608.11560","doi":"https://doi.org/10.48550/arxiv.2608.11560","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.11560","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:2608.11560","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.11560","pdf_url":"https://arxiv.org/pdf/2608.11560","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/W7202347769.pdf","grobid_xml":"https://content.openalex.org/works/W7202347769.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2810510551","https://openalex.org/W3173925177","https://openalex.org/W2186512312","https://openalex.org/W20256240","https://openalex.org/W6884980100","https://openalex.org/W3010924289","https://openalex.org/W4396778388","https://openalex.org/W2985297696","https://openalex.org/W3190837579","https://openalex.org/W2900706408"],"abstract_inverted_index":{"Personalizing":[0],"marketing":[1],"messages":[2],"with":[3,64,153],"contextual":[4,50],"multi-armed":[5],"bandits":[6],"(CMABs)":[7],"drives":[8],"real":[9],"business":[10],"value,":[11],"yet":[12],"the":[13,37,65,105,108,113,116,128,159,182,223,249,252,258],"objective":[14],"that":[15,94,190,227],"ultimately":[16],"matters":[17],"-":[18,22,69,81,132,142,226],"a":[19,40,49,70,74,78,96,133,138,147,177,194,215,236,265],"downstream":[20],"conversion":[21],"is":[23,52,130,161,213,232,239,244],"observed":[24],"only":[25],"weeks":[26],"later,":[27],"too":[28],"late":[29],"to":[30,185,264],"drive":[31],"online":[32],"learning.":[33],"Teams":[34],"therefore":[35],"train":[36],"bandit":[38,51,114,183],"on":[39,99,146,222],"fast":[41],"proxy":[42],"reward,":[43],"and":[44,110,137,143,156,257],"separately":[45],"must":[46],"judge":[47],"whether":[48],"worth":[53],"its":[54],"complexity":[55],"over":[56],"sending":[57],"one":[58,157,225],"best":[59],"message.":[60],"Settling":[61],"both":[62],"decisions":[63],"usual":[66],"offline":[67,172],"checks":[68],"batch":[71],"off-policy":[72],"estimate,":[73],"marginal":[75],"arm-discrimination":[76],"test,":[77],"confidence":[79],"interval":[80],"can":[82,174],"mislead":[83],"systematically":[84],"under":[85],"delayed":[86],"feedback.":[87],"We":[88,124],"give":[89],"an":[90],"ordered":[91,250],"diagnostic":[92],"protocol":[93],"screens":[95],"reward-and-policy":[97],"candidate":[98],"two":[100,253],"axes,":[101],"alignment":[102],"(does":[103],"optimizing":[104],"reward":[106,179],"move":[107],"north-star?)":[109],"learnability":[111],"(can":[112],"identify":[115],"reward-optimal":[117],"policy?),":[118],"before":[119],"trusting":[120],"any":[121],"reported":[122],"lift.":[123],"validate":[125],"it":[126,145,255,263],"where":[127],"truth":[129],"known":[131],"public":[134],"off-policy-evaluation":[135],"benchmark":[136],"controllable":[139],"synthetic":[140],"generator":[141],"illustrate":[144],"deployed":[148],"large-marketplace":[149],"push":[150],"system":[151],"(where,":[152],"five":[154],"arms":[155],"split,":[158],"evidence":[160],"directional":[162],"rather":[163,246],"than":[164,247],"powered).":[165],"Two":[166],"lessons":[167,254],"recur.":[168],"(N1)":[169],"A":[170],"single":[171,211],"number":[173],"mis-rank":[175],"rewards:":[176],"denser":[178],"signal":[180],"gives":[181],"more":[184],"learn":[186],"from,":[187],"so":[188,235],"rewards":[189],"look":[191],"tied":[192],"in":[193,208],"static":[195],"estimate":[196],"pull":[197],"apart":[198],"once":[199],"learning":[200],"happens":[201],"online.":[202],"(N2)":[203],"If":[204],"you":[205],"cannot":[206],"tell":[207],"advance":[209],"which":[210],"message":[212],"best,":[214],"per-user":[216],"policy":[217],"partly":[218],"just":[219],"avoids":[220],"betting":[221],"wrong":[224],"looks":[228],"like":[229],"personalization":[230],"but":[231],"really":[233],"robustness,":[234],"\"personalization":[237],"premium\"":[238],"easily":[240],"overstated.":[241],"Our":[242],"contribution":[243],"methodological":[245],"algorithmic:":[248],"protocol,":[251],"surfaces,":[256],"end-to-end":[259],"experience":[260],"of":[261],"applying":[262],"delayed-feedback":[266],"CMAB.":[267]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-08-14T00:00:00"}