{"id":"https://openalex.org/W7208745310","doi":"https://doi.org/10.48550/arxiv.2609.04198","title":"Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints","display_name":"Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints","publication_year":2026,"publication_date":"2026-09-03","ids":{"openalex":"https://openalex.org/W7208745310","doi":"https://doi.org/10.48550/arxiv.2609.04198"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.04198","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.04198","pdf_url":"https://arxiv.org/pdf/2609.04198","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/2609.04198","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5150009184","display_name":"Haoyaun Zhu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I91136226","display_name":"University of Sheffield","ror":"https://ror.org/05krs5044","country_code":"GB","type":"education","lineage":["https://openalex.org/I91136226"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Zhu, Haoyaun","raw_affiliation_strings":["R&D Department , Cambridge AI+ Ltd. , Cambridge CB23 3UY , UK","School of Electronic and Electrical Engineering , University of Sheffield , Sheffield S10 2TN , UK"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"R&D Department , Cambridge AI+ Ltd. , Cambridge CB23 3UY , UK","institution_ids":[]},{"raw_affiliation_string":"School of Electronic and Electrical Engineering , University of Sheffield , Sheffield S10 2TN , UK","institution_ids":["https://openalex.org/I91136226"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5149982142","display_name":"Jie Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jie","raw_affiliation_strings":["R&D Department , Cambridge AI+ Ltd. , Cambridge CB23 3UY , UK","R&D Department , Ranplan Wireless Network Design Ltd. , Cambridge CB23 3UY , UK"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"R&D Department , Cambridge AI+ Ltd. , Cambridge CB23 3UY , UK","institution_ids":[]},{"raw_affiliation_string":"R&D Department , Ranplan Wireless Network Design Ltd. , Cambridge CB23 3UY , UK","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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.15119999647140503,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.15119999647140503,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11986","display_name":"Scientific Computing and Data Management","score":0.0851999968290329,"subfield":{"id":"https://openalex.org/subfields/1802","display_name":"Information Systems and Management"},"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.07810000330209732,"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/measurement-instability","display_name":"measurement instability","score":0.5979999899864197}],"concepts":[{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.6736999750137329},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6431999802589417},{"id":"https://openalex.org/C199521495","wikidata":"https://www.wikidata.org/wiki/Q181487","display_name":"Audit","level":2,"score":0.6359999775886536},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.6204000115394592},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.6079000234603882},{"id":"https://openalex.org/C93518851","wikidata":"https://www.wikidata.org/wiki/Q180160","display_name":"Metadata","level":2,"score":0.5289000272750854},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.4415000081062317},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.38760000467300415},{"id":"https://openalex.org/C142853389","wikidata":"https://www.wikidata.org/wiki/Q744778","display_name":"Association (psychology)","level":2,"score":0.37130001187324524},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.35530000925064087},{"id":"https://openalex.org/C133462117","wikidata":"https://www.wikidata.org/wiki/Q4929239","display_name":"Data collection","level":2,"score":0.3411000072956085},{"id":"https://openalex.org/C2779662365","wikidata":"https://www.wikidata.org/wiki/Q5416694","display_name":"Event (particle physics)","level":2,"score":0.3093000054359436},{"id":"https://openalex.org/C79403827","wikidata":"https://www.wikidata.org/wiki/Q3988","display_name":"Real-time computing","level":1,"score":0.30559998750686646},{"id":"https://openalex.org/C20556612","wikidata":"https://www.wikidata.org/wiki/Q4469374","display_name":"Volume (thermodynamics)","level":2,"score":0.29280000925064087},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.2867000102996826},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.28619998693466187},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.2856000065803528},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.27869999408721924},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.2759000062942505},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.272599995136261},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.27239999175071716},{"id":"https://openalex.org/C64943373","wikidata":"https://www.wikidata.org/wiki/Q2651003","display_name":"Alpha (finance)","level":4,"score":0.2687000036239624},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.2667999863624573},{"id":"https://openalex.org/C2778217198","wikidata":"https://www.wikidata.org/wiki/Q556977","display_name":"Cutoff","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C93996380","wikidata":"https://www.wikidata.org/wiki/Q44127","display_name":"Server","level":2,"score":0.2522999942302704},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.25119999051094055}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.04198","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.04198","pdf_url":"https://arxiv.org/pdf/2609.04198","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.2609.04198","doi":"https://doi.org/10.48550/arxiv.2609.04198","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.04198","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:2609.04198","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.04198","pdf_url":"https://arxiv.org/pdf/2609.04198","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":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7208745310.pdf","grobid_xml":"https://content.openalex.org/works/W7208745310.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7213996713","https://openalex.org/W7214756950","https://openalex.org/W7222534895","https://openalex.org/W7169550126","https://openalex.org/W7214125621","https://openalex.org/W7121321095","https://openalex.org/W7169866231","https://openalex.org/W7158423527","https://openalex.org/W7214545680","https://openalex.org/W7202368021"],"abstract_inverted_index":{"Language-model":[0],"judges":[1],"now":[2],"gate":[3,279],"training":[4],"data,":[5],"score":[6],"generations,":[7],"and":[8,73,120,198,225],"drive":[9],"leaderboards.":[10],"The":[11],"judge":[12],"is":[13,264],"then":[14],"a":[15,70,81,97,126,218,226,229,258,261,266,269],"measurement":[16],"instrument,":[17],"resting":[18],"on":[19,139,152,188,199,253,280],"one":[20],"rarely":[21],"stated":[22],"assumption:":[23],"the":[24,29,34,87,95,106,115,140,146,153,172,182,194,205,215,235],"same":[25,30,35],"request,":[26],"sent":[27],"to":[28,176],"model":[31,262],"name,":[32],"reads":[33],"tomorrow.":[36],"We":[37,213],"audited":[38,59],"that":[39,100,128],"assumption":[40],"in":[41,49,245],"two":[42],"preregistered":[43,270],"campaigns":[44],"with":[45,86,202],"every":[46],"threshold":[47],"fixed":[48],"advance;":[50],"neither":[51],"got":[52],"past":[53],"validating":[54],"its":[55,274],"instrument.":[56],"Across":[57],"52,988":[58],"request":[60],"attempts,":[61],"same-window":[62],"repeat":[63],"rankings":[64],"agreed":[65,77],"at":[66,78,90,231],"Spearman":[67],"0.400":[68],"against":[69,80],"required":[71,82],"0.90,":[72],"byte-identical":[74,121],"next-day":[75],"replays":[76],"0.78":[79],"0.99,":[83],"each":[84],"time":[85],"execution":[88],"record":[89],"ceiling.":[91],"Three":[92],"mechanisms":[93],"explain":[94],"gap:":[96],"label-to-meaning":[98],"mapping":[99],"biased":[101],"readouts":[102,130],"as":[103,105],"strongly":[104],"signal;":[107],"candidate":[108],"gaps":[109],"seven":[110],"orders":[111],"of":[112,181,234],"magnitude":[113],"below":[114],"instrument's":[116],"own":[117],"noise":[118,127],"floor;":[119],"inputs":[122],"returning":[123],"different":[124],"rankings,":[125],"exact-permutation":[129],"compound.":[131],"Neither":[132],"metric":[133],"substitution":[134],"nor":[135],"sampling":[136],"repaired":[137],"it":[138],"tested":[141],"grid.":[142],"Preregistered":[143],"follow-ups":[144],"bound":[145],"problem:":[147],"waiting":[148],"did":[149,166],"not":[150,167,211,265],"help":[151,168],"days":[154],"sampled":[155],"(0.805":[156],"versus":[157],"0.800,":[158],"replicated":[159],"over":[160],"five":[161],"further":[162],"days);":[163],"switching":[164],"providers":[165,170],"(four":[169],"share":[171],"floor,":[173],"medians":[174],"0.74":[175],"0.88,":[177],"predicted":[178],"by":[179],"none":[180],"metadata":[183],"fields":[184],"they":[185],"expose);":[186],"self-hosting":[187],"batch-invariant":[189],"kernels":[190],"helped":[191],"only":[192],"while":[193],"server":[195],"was":[196],"quiet;":[197],"constructed":[200],"errors":[201],"known":[203],"gaps,":[204],"readout's":[206],"separation":[207],"tracks":[208],"error":[209],"type,":[210],"size.":[212],"distill":[214],"evidence":[216],"into":[217],"three-level":[219],"snapshot-identity":[220],"ladder,":[221],"eight":[222],"design":[223],"rules,":[224],"reporting":[227],"checklist;":[228],"pilot":[230],"roughly":[232],"2%":[233],"study's":[236],"call":[237],"volume":[238],"would":[239],"have":[240],"exposed":[241],"both":[242],"unreachable":[243],"gates":[244],"advance.":[246],"All":[247],"results":[248],"concern":[249],"externally":[250],"measured":[251],"behaviour":[252],"shared":[254,259],"serving":[255],"infrastructure.":[256],"On":[257],"endpoint,":[260],"name":[263],"frozen":[267],"instrument;":[268],"evaluation":[271],"must":[272],"measure":[273],"instrument":[275],"before":[276],"freezing":[277],"any":[278],"it.":[281]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-05T00:00:00"}