{"id":"https://openalex.org/W7126079967","doi":"https://doi.org/10.48550/arxiv.2601.20276","title":"Beyond the Needle's Illusion: Decoupled Evaluation of Evidence Access and Use under Semantic Interference at 326M-Token Scale","display_name":"Beyond the Needle's Illusion: Decoupled Evaluation of Evidence Access and Use under Semantic Interference at 326M-Token Scale","publication_year":2026,"publication_date":"2026-01-28","ids":{"openalex":"https://openalex.org/W7126079967","doi":"https://doi.org/10.48550/arxiv.2601.20276"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2601.20276","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.20276","pdf_url":"https://arxiv.org/pdf/2601.20276","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/2601.20276","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124265056","display_name":"Tianwei Lin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lin, Tianwei","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101190696","display_name":"Zuyi Zhou","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Zuyi","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124157348","display_name":"Xinda Zhao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Xinda","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124245280","display_name":"Chenke Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Chenke","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102948452","display_name":"Xiaohong Li","orcid":"https://orcid.org/0000-0003-4525-1754","observed_orcids":["https://orcid.org/0000-0003-4525-1754"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Xiaohong","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124283094","display_name":"Yu Yi Chen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yu","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5063098561","display_name":"Chuanrui Hu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Chuanrui","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102940347","display_name":"Jian Pei","orcid":"https://orcid.org/0000-0002-4412-4609","observed_orcids":["https://orcid.org/0000-0002-4412-4609","https://orcid.org/0000-0002-2200-8711"]},"institutions":[{"id":"https://openalex.org/I170897317","display_name":"Duke University","ror":"https://ror.org/00py81415","country_code":"US","type":"education","lineage":["https://openalex.org/I170897317"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Pei, Jian","raw_affiliation_strings":["Duke University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Duke University","institution_ids":["https://openalex.org/I170897317"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5111170444","display_name":"Yafeng Deng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deng, Yafeng","raw_affiliation_strings":["Shanda Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanda Group","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"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/T10028","display_name":"Topic Modeling","score":0.8661999702453613,"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":0.8661999702453613,"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/T10286","display_name":"Information Retrieval and Search Behavior","score":0.11550000309944153,"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"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.005200000014156103,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/semantic-interference","display_name":"semantic interference","score":0.8460000157356262},{"id":"https://openalex.org/keywords/adversarial-benchmark","display_name":"adversarial benchmark","score":0.6439999938011169},{"id":"https://openalex.org/keywords/hard-negative-examples","display_name":"hard negative examples","score":0.4399999976158142},{"id":"https://openalex.org/keywords/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":0.8009999990463257},{"id":"https://openalex.org/keywords/memory-bank","display_name":"memory bank","score":0.40700000524520874},{"id":"https://openalex.org/keywords/reality-gap","display_name":"reality gap","score":0.5070000290870667},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.4090000092983246}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8037999868392944},{"id":"https://openalex.org/C13424479","wikidata":"https://www.wikidata.org/wiki/Q5687237","display_name":"Haystack","level":2,"score":0.7544999718666077},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6632000207901001},{"id":"https://openalex.org/C2780513914","wikidata":"https://www.wikidata.org/wiki/Q18210350","display_name":"Bottleneck","level":2,"score":0.5921000242233276},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.492900013923645},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.4643000066280365},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.436599999666214},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4339999854564667},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.41670000553131104},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.4068000018596649},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3878999948501587},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.37779998779296875},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3675999939441681},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3393999934196472},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.33390000462532043},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3260999917984009},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.321399986743927},{"id":"https://openalex.org/C47487241","wikidata":"https://www.wikidata.org/wiki/Q5227230","display_name":"Data access","level":2,"score":0.3183000087738037},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3172000050544739},{"id":"https://openalex.org/C81669768","wikidata":"https://www.wikidata.org/wiki/Q2359161","display_name":"Precision and recall","level":2,"score":0.28859999775886536},{"id":"https://openalex.org/C160145156","wikidata":"https://www.wikidata.org/wiki/Q778586","display_name":"Executable","level":2,"score":0.2874000072479248},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2833999991416931},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.2827000021934509},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.2775999903678894},{"id":"https://openalex.org/C2779458634","wikidata":"https://www.wikidata.org/wiki/Q24963715","display_name":"Debiasing","level":2,"score":0.26910001039505005},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.26899999380111694},{"id":"https://openalex.org/C136536468","wikidata":"https://www.wikidata.org/wiki/Q1225894","display_name":"Undersampling","level":2,"score":0.26350000500679016},{"id":"https://openalex.org/C2778753569","wikidata":"https://www.wikidata.org/wiki/Q1960395","display_name":"Span (engineering)","level":2,"score":0.2549999952316284},{"id":"https://openalex.org/C2987487971","wikidata":"https://www.wikidata.org/wiki/Q8096","display_name":"Lexical access","level":3,"score":0.2547000050544739},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.25060001015663147}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2601.20276","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.20276","pdf_url":"https://arxiv.org/pdf/2601.20276","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.2601.20276","doi":"https://doi.org/10.48550/arxiv.2601.20276","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.20276","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:2601.20276","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.20276","pdf_url":"https://arxiv.org/pdf/2601.20276","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/W7126079967.pdf","grobid_xml":"https://content.openalex.org/works/W7126079967.grobid-xml"},"referenced_works_count":11,"referenced_works":["https://openalex.org/W4391709268","https://openalex.org/W4394778349","https://openalex.org/W4399555009","https://openalex.org/W4399991263","https://openalex.org/W4402671766","https://openalex.org/W4403560390","https://openalex.org/W4411119426","https://openalex.org/W4412886690","https://openalex.org/W4415289348","https://openalex.org/W4416982487","https://openalex.org/W7117144337"],"related_works":["https://openalex.org/W4407310644","https://openalex.org/W7117144337","https://openalex.org/W7166727927","https://openalex.org/W4417112348","https://openalex.org/W4404782876","https://openalex.org/W7140144908","https://openalex.org/W4417239995","https://openalex.org/W7169834506","https://openalex.org/W7214023293","https://openalex.org/W4403749280"],"abstract_inverted_index":{"Long-context":[0],"LLM":[1,109],"agents":[2],"must":[3],"access":[4,121,173],"the":[5,16,31,50,188],"right":[6],"evidence":[7,97,120,172],"from":[8,125,148],"large":[9],"environments":[10],"and":[11,30,95,108,141],"use":[12],"it":[13],"faithfully.":[14],"However,":[15],"popular":[17],"Needle-in-a-Haystack":[18],"(NIAH)":[19],"evaluation":[20],"mostly":[21],"measures":[22],"benign":[23,167],"span":[24],"localization.":[25],"The":[26],"needle":[27],"is":[28,33,187],"near-unique,":[29],"haystack":[32],"largely":[34],"irrelevant.":[35],"We":[36,111],"introduce":[37],"EverMemBench-S":[38],"(EMB-S),":[39],"an":[40],"adversarial":[41],"NIAH-style":[42],"benchmark":[43],"built":[44],"on":[45],"a":[46,84,114,145,153,160],"326M-token":[47,156],"MemoryBank.":[48],"While":[49],"full":[51],"MemoryBank":[52],"spans":[53],"326M":[54],"tokens":[55,78],"for":[56,136,191],"retrieval-based":[57],"(RAG)":[58],"evaluation,":[59],"we":[60,158],"evaluate":[61],"native":[62,138],"long-context":[63,139,192],"models":[64],"only":[65],"at":[66,194],"scales":[67],"that":[68,118,165,180],"fit":[69],"within":[70],"each":[71],"model's":[72],"context":[73,184],"window":[74],"(up":[75],"to":[76,82,152],"1M":[77],"in":[79,171],"this":[80],"work)":[81],"ensure":[83],"fair":[85],"comparison.":[86],"EMB-S":[87],"pairs":[88],"queries":[89],"with":[90],"collision-tested":[91],"near-miss":[92],"hard":[93],"negatives":[94],"gold":[96],"sets":[98],"spanning":[99],"one":[100],"or":[101],"more":[102],"documents,":[103],"validated":[104],"via":[105],"human":[106],"screening":[107],"verification.":[110],"also":[112],"propose":[113],"decoupled":[115],"diagnostic":[116],"protocol":[117],"reports":[119],"(document-ID":[122],"localization)":[123],"separately":[124],"end-to-end":[126],"QA":[127],"quality":[128],"under":[129,174],"full-context":[130],"prompting.":[131],"This":[132],"enables":[133],"consistent":[134],"diagnosis":[135],"both":[137],"prompting":[140],"retrieval":[142],"pipelines.":[143],"Across":[144],"reference-corpus":[146],"ladder":[147],"domain-isolated":[149],"64K":[150],"contexts":[151],"globally":[154],"shared":[155],"environment,":[157],"observe":[159],"clear":[161],"reality":[162],"gap.":[163],"Systems":[164],"saturate":[166],"NIAH":[168],"degrade":[169],"sharply":[170],"semantic":[175,181],"interference.":[176],"These":[177],"results":[178],"indicate":[179],"discrimination,":[182],"not":[183],"length":[185],"alone,":[186],"dominant":[189],"bottleneck":[190],"memory":[193],"scale.":[195]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-01-30T00:00:00"}