{"meta":{"query_hash":"0e97332707a1","filters":{"venue":"Accreditation and Quality Assurance"},"cohort_total":29,"direct_labels_cover":0,"predictions_cover":29,"exported":29,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/0e97332707a1","api":"https://metacan.xera.ac/api/v1/cohort?venue=Accreditation+and+Quality+Assurance"},"results":[{"id":"W1964146798","doi":"10.1007/s00769-006-0118-8","title":"Quality specifications for evaluation and comparison of performance among external quality assessment schemes in occupational and environmental laboratory medicine","year":2006,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"European Commission","keywords":"Equivalence (formal languages); External quality assessment; Quality assurance; Reliability engineering; Quality (philosophy); Quality assessment; Computer science; Set (abstract data type); Data mining; Risk analysis (engineering); Evaluation methods; Engineering; Mathematics; Operations management; Medicine","score_opus":0.17667360289144882,"score_gpt":0.4736216374677477,"score_spread":0.2969480345762989,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964146798","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8421538,0.0013863596,0.13769898,0.0011109052,0.00025328746,0.0021041033,0.0028792645,0.0011009321,0.011312483],"genre_scores_gemma":[0.9366708,0.00018561668,0.057857,0.00022606755,0.000025078882,0.0010057023,0.003080318,0.00014072037,0.0008086508],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.90014386,0.05770801,0.011857679,0.002065172,0.026243942,0.0019813592],"domain_scores_gemma":[0.68240464,0.17977612,0.022401338,0.017733335,0.09540672,0.002277888],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09415271,0.0005810044,0.00078957516,0.0051041367,0.0017504285,0.004078177,0.002187784,0.002068527,0.0013301794],"category_scores_gemma":[0.21293923,0.00046695553,0.0013389036,0.0039999285,0.0012568651,0.0021227552,0.0018289752,0.00089674536,0.00043999852],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0068009207,0.0029694834,0.43459666,0.0016652748,0.0006377198,0.00016167117,0.0059781075,0.04656763,0.023560518,0.020149847,0.013237949,0.44367418],"study_design_scores_gemma":[0.0010474144,0.014680887,0.6858119,0.001209926,0.00088892394,0.00041822303,0.009476052,0.15137157,0.09955212,0.010311904,0.024819046,0.00041204607],"about_ca_topic_score_codex":0.015785487,"about_ca_topic_score_gemma":0.008553893,"teacher_disagreement_score":0.09415271,"about_ca_system_score_codex":0.006901387,"about_ca_system_score_gemma":0.008946931,"threshold_uncertainty_score":0.4979331},"labels":[],"label_agreement":null},{"id":"W1964182244","doi":"10.1007/s00769-009-0561-4","title":"Criteria to define the standard deviation for proficiency assessment for the determination of essential trace elements in serum: comparison of Z-scores based on the Horwitz function or on biological variability","year":2009,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Trace Elements in Health","field":"Nursing","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"TRACE (psycholinguistics); Standard deviation; Quality (philosophy); Function (biology); Statistics; Mathematics; Reference values; Medicine; Physics","score_opus":0.10825897629631515,"score_gpt":0.46031199485717417,"score_spread":0.352053018560859,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1964182244","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.276513,0.010147248,0.6792278,0.003037193,0.001420756,0.0032165325,0.0025053895,0.002293627,0.02163835],"genre_scores_gemma":[0.6276691,0.00080243306,0.36425236,0.00070129865,0.00018601546,0.0023130854,0.0022020345,0.0004309239,0.0014426807],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.89535743,0.040812783,0.020949785,0.0039388,0.036979575,0.0019616145],"domain_scores_gemma":[0.71440804,0.19402243,0.014983514,0.013473179,0.05948519,0.0036275904],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14544035,0.0014400615,0.0025753728,0.009792268,0.002629602,0.0045928783,0.006318677,0.00454661,0.001909987],"category_scores_gemma":[0.23893262,0.0006124574,0.003360762,0.004694944,0.0035214103,0.0024935803,0.0047180313,0.003026344,0.0011649255],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.009164439,0.0020555053,0.338063,0.004549029,0.0031370313,0.0010947325,0.010061459,0.0143000055,0.04680828,0.06141762,0.027228318,0.48212063],"study_design_scores_gemma":[0.0020677145,0.018767796,0.52507895,0.0039419937,0.0032860788,0.00919356,0.014366834,0.15142947,0.11420739,0.06979398,0.08632416,0.0015420443],"about_ca_topic_score_codex":0.0032285554,"about_ca_topic_score_gemma":0.0024511,"teacher_disagreement_score":0.14544035,"about_ca_system_score_codex":0.0020689887,"about_ca_system_score_gemma":0.005088559,"threshold_uncertainty_score":0.7691713},"labels":[],"label_agreement":null},{"id":"W1967499314","doi":"10.1007/s00769-008-0421-7","title":"External quality assessment in microbiology: comparison of results from Belgian and Canadian laboratories with regard to their ability to identify Streptococcus pyogenes","year":2008,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Bacterial Identification and Susceptibility Testing","field":"Biochemistry, Genetics and Molecular Biology","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Clinical microbiology; External quality assessment; Streptococcus pyogenes; Viridans streptococci; Streptococcus; Microbiology; Medicine; Throat; Medical physics; Biology; Pathology; Surgery; Staphylococcus aureus; Bacteria","score_opus":0.05474549356872136,"score_gpt":0.3630003531813596,"score_spread":0.3082548596126382,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1967499314","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97574514,0.0029751505,0.0035473532,0.0009450072,0.00014390695,0.00017899263,0.0017019713,0.00017239922,0.014590144],"genre_scores_gemma":[0.99299884,0.0006475183,0.0021925345,0.0003081012,0.00002730056,0.00003561228,0.0018939778,0.00006574684,0.0018302705],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9732371,0.0046960805,0.0016470037,0.0021147197,0.0154247945,0.00288029],"domain_scores_gemma":[0.9282221,0.010993293,0.0068655196,0.0020910057,0.048396565,0.0034313963],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020058988,0.00085675134,0.00065861957,0.006625014,0.0027437496,0.004370475,0.0015752128,0.0010586905,0.0012601033],"category_scores_gemma":[0.040413637,0.00039383324,0.00092093734,0.0070375693,0.0024894385,0.00059861457,0.002257087,0.0005843137,0.00032409796],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020112481,0.00024098821,0.90131074,0.0006204683,0.000361176,0.00022185194,0.008345313,0.0008853194,0.0116271265,0.00066589605,0.0028205805,0.07088926],"study_design_scores_gemma":[0.00002662374,0.00019507651,0.9869495,0.000102199294,0.00013043842,0.00013812893,0.0019977186,0.000347128,0.004014303,0.00005117277,0.0060133664,0.00003418165],"about_ca_topic_score_codex":0.7907845,"about_ca_topic_score_gemma":0.7530147,"teacher_disagreement_score":0.20921552,"about_ca_system_score_codex":0.018330457,"about_ca_system_score_gemma":0.024714734,"threshold_uncertainty_score":0.42089522},"labels":[],"label_agreement":null},{"id":"W1970174238","doi":"10.1007/s00769-014-1082-3","title":"Performance of Canadian clinical laboratories processing throat culture proficiency testing surveys","year":2014,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Bacterial Identification and Susceptibility Testing","field":"Biochemistry, Genetics and Molecular Biology","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"University of British Columbia","keywords":"Streptococcus pyogenes; Pharyngitis; Throat; Throat culture; Streptococcus; Pathogen; Clinical microbiology; Medicine; Microbiology; Biology; Surgery; Internal medicine; Bacteria; Staphylococcus aureus","score_opus":0.07274539070457586,"score_gpt":0.3449978617734287,"score_spread":0.27225247106885286,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970174238","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9692947,0.001778929,0.0018338638,0.0020224494,0.00019531425,0.00036844923,0.006054721,0.00083366793,0.01761785],"genre_scores_gemma":[0.9866621,0.00050797214,0.0028758883,0.00051163207,0.000034261397,0.000058930375,0.0035692074,0.00007325791,0.0057067573],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9775178,0.0049552876,0.0013288641,0.0024753658,0.009874941,0.0038477823],"domain_scores_gemma":[0.93661386,0.009591601,0.0034361158,0.0016250556,0.043602895,0.005130507],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.012109794,0.0011964047,0.0008057411,0.0025063811,0.0035672043,0.0031705976,0.0021798208,0.0015187258,0.0038858827],"category_scores_gemma":[0.040655483,0.00082633557,0.00072532706,0.0045659295,0.0013113231,0.00072154315,0.0017510026,0.0008268885,0.001574607],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008518929,0.0006353553,0.7794845,0.0010220067,0.00034959728,0.00048672964,0.008216274,0.009715334,0.01662609,0.0009868139,0.030086512,0.14387189],"study_design_scores_gemma":[0.000106880856,0.001146439,0.955151,0.000118251504,0.00014275844,0.0001972245,0.003057554,0.0056627984,0.009656862,0.00007104796,0.024552954,0.00013619571],"about_ca_topic_score_codex":0.9726355,"about_ca_topic_score_gemma":0.9708892,"teacher_disagreement_score":0.9878902,"about_ca_system_score_codex":0.053359035,"about_ca_system_score_gemma":0.07430685,"threshold_uncertainty_score":0.38714868},"labels":[],"label_agreement":null},{"id":"W1970248714","doi":"10.1007/s00769-006-0144-6","title":"Estimation of the biological variation of glucose-6-phosphate dehydrogenase in dried blood spots","year":2006,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Newborn Screening Ontario","funders":"","keywords":"Coefficient of variation; Mathematics; Matrix (chemical analysis); Analytical Chemistry (journal); Chemistry; Systematic error; Combinatorics; Statistics; Chromatography","score_opus":0.043625093662082386,"score_gpt":0.34454190133102996,"score_spread":0.3009168076689476,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1970248714","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.95508045,0.0037344776,0.037053153,0.0001724269,0.00024874526,0.00009988063,0.0006993583,0.00020754349,0.0027039377],"genre_scores_gemma":[0.9730108,0.0010247341,0.023128655,0.00017728249,0.00006399649,0.000066831,0.0009578605,0.00002994191,0.0015399412],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99742764,0.00086057826,0.00012509135,0.00042188447,0.0010650626,0.00009979336],"domain_scores_gemma":[0.9984061,0.00073699566,0.00020434076,0.00014979346,0.00042640048,0.00007625737],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016704688,0.00040399458,0.00038030196,0.0011729208,0.0002882168,0.00074178615,0.0005002782,0.0006355843,0.00033517563],"category_scores_gemma":[0.004751065,0.00023315565,0.00031300978,0.0014029378,0.0005510734,0.00023340165,0.00018611927,0.00055002497,0.0002269676],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003900006,0.00030025287,0.2131443,0.00033334014,0.00057911914,0.0002994663,0.0006752736,0.0015198423,0.7066338,0.0010175932,0.00074198324,0.070854984],"study_design_scores_gemma":[0.000043449974,0.0010343074,0.5092463,0.000030552572,0.0002033714,0.0018313589,0.00018230572,0.0062887804,0.47746128,0.000547884,0.0030859918,0.000044421817],"about_ca_topic_score_codex":0.0027364956,"about_ca_topic_score_gemma":0.0019850035,"teacher_disagreement_score":0.0027364956,"about_ca_system_score_codex":0.00044155988,"about_ca_system_score_gemma":0.00032342816,"threshold_uncertainty_score":0.008834362},"labels":[],"label_agreement":null},{"id":"W1981139298","doi":"10.1007/s00769-008-0385-7","title":"The European conferences on quality in the spotlight in medical laboratories, Antwerp, Belgium, 20–21 March 2006 and 22–23 March 2007","year":2008,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Shared Health","funders":"","keywords":"Commercial law; Quality (philosophy); Engineering; Library science; Political science; Computer science; Law; Physics","score_opus":0.08485616912799505,"score_gpt":0.40242432933400296,"score_spread":0.3175681602060079,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1981139298","genre_codex":"commentary","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.052715708,0.19441327,0.030940318,0.33585706,0.16615221,0.0014816594,0.006597943,0.0010951923,0.2107466],"genre_scores_gemma":[0.19095668,0.04291696,0.03409665,0.023867968,0.038287528,0.0005938731,0.0057289675,0.0011281383,0.66242313],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99314404,0.002840033,0.00038038104,0.0005389152,0.0021468005,0.0009498219],"domain_scores_gemma":[0.9874407,0.0017452531,0.0013452776,0.00047090297,0.0044051185,0.004592699],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016351528,0.0014715203,0.0009187379,0.0019147723,0.0015927812,0.0044140783,0.0021594707,0.004986051,0.036665387],"category_scores_gemma":[0.011018304,0.0005249207,0.0009792793,0.0017291197,0.0018987455,0.0019346274,0.0037065323,0.0036178848,0.0050994265],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010887916,0.00039130484,0.0028581282,0.0013836797,0.00008494965,0.0006790905,0.0009297465,0.0021906504,0.006743944,0.0061855344,0.81040215,0.16706206],"study_design_scores_gemma":[0.00007529108,0.00015807858,0.014770434,0.0006229162,0.000026561043,0.00039226635,0.00071015966,0.00044988267,0.0022604659,0.0013128123,0.97918105,0.00003996033],"about_ca_topic_score_codex":0.009927678,"about_ca_topic_score_gemma":0.01626734,"teacher_disagreement_score":0.036665387,"about_ca_system_score_codex":0.0052489084,"about_ca_system_score_gemma":0.009087764,"threshold_uncertainty_score":0.122657895},"labels":[],"label_agreement":null},{"id":"W1982922574","doi":"10.1007/s00769-009-0506-y","title":"Instability of mercury in specimens of human urine for external quality assessment","year":2009,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Mercury impact and mitigation studies","field":"Environmental Science","cited_by":8,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"Mercury (programming language); Urine; Chemistry; Chromatography; Uric acid; Urine sample; External quality assessment; Environmental chemistry; Biochemistry; Medicine; Computer science","score_opus":0.07372258331384528,"score_gpt":0.39603436428930616,"score_spread":0.32231178097546087,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1982922574","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9793428,0.0012681307,0.01384922,0.00027796466,0.00011522761,0.000090478905,0.0005299929,0.00014199132,0.0043842355],"genre_scores_gemma":[0.9891866,0.00034355104,0.0069552907,0.00018929025,0.000060908187,0.00006551092,0.00076451234,0.00005356573,0.0023808028],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.9975164,0.000989635,0.00018226274,0.00031897204,0.00072642346,0.00026623756],"domain_scores_gemma":[0.9970114,0.000767315,0.00049858313,0.0003187051,0.0011922317,0.00021178831],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0020640513,0.00051378284,0.00032136758,0.0017510342,0.001580517,0.00080057164,0.0005055462,0.0012521471,0.0017046608],"category_scores_gemma":[0.006150278,0.00025109053,0.0005863338,0.0011097033,0.0010037166,0.00033678758,0.00058979035,0.00078579923,0.000813386],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017079699,0.00018679145,0.05111969,0.00016429905,0.0001731284,0.0006290009,0.0015793346,0.00029937705,0.9296072,0.00042282557,0.00040746355,0.0137030715],"study_design_scores_gemma":[0.000016669144,0.0018165767,0.06069891,0.000041703097,0.00015518758,0.0015404219,0.00065541826,0.0008622408,0.9312194,0.00028120438,0.002675656,0.000036614405],"about_ca_topic_score_codex":0.0028798964,"about_ca_topic_score_gemma":0.0022620705,"teacher_disagreement_score":0.0028798964,"about_ca_system_score_codex":0.00044099244,"about_ca_system_score_gemma":0.0009403241,"threshold_uncertainty_score":0.010915875},"labels":[],"label_agreement":null},{"id":"W1992242927","doi":"10.1007/s00769-006-0108-x","title":"Occupational and environmental laboratory medicine: A network of EQAS organisers","year":2006,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Chemical Safety and Risk Management","field":"Chemical Engineering","cited_by":7,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"External quality assessment; Environmental health; Medical laboratory; Medicine; Traceability; Computer science; Pathology","score_opus":0.01119142992951947,"score_gpt":0.25189131332869574,"score_spread":0.24069988339917628,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1992242927","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85049814,0.004712847,0.020421201,0.039120838,0.0013459733,0.0044184397,0.0035520499,0.0010904173,0.07484005],"genre_scores_gemma":[0.862938,0.0015517513,0.017563846,0.007104246,0.0006286127,0.0013399984,0.0015526535,0.00035845875,0.10696235],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98977715,0.0038173096,0.000704777,0.0012407046,0.0016962602,0.0027637905],"domain_scores_gemma":[0.93621016,0.0075108735,0.0037531867,0.0024607894,0.0117157735,0.038349207],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015564975,0.0004302012,0.0002820545,0.0028779292,0.0053978814,0.003281077,0.0013837143,0.0020646919,0.049700845],"category_scores_gemma":[0.012026077,0.00080120814,0.00046387312,0.0026003208,0.0014749672,0.0023625293,0.0074781803,0.001703774,0.004414864],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0030323225,0.003192354,0.5551071,0.00087400037,0.00019730804,0.003906847,0.018530479,0.003005676,0.013713348,0.0052809245,0.07805743,0.31510216],"study_design_scores_gemma":[0.00040221048,0.001731192,0.518993,0.00043255225,0.000112908456,0.0024694884,0.029999048,0.0049210656,0.0024046386,0.0016760848,0.43669847,0.00015931379],"about_ca_topic_score_codex":0.012144663,"about_ca_topic_score_gemma":0.022032866,"teacher_disagreement_score":0.049700845,"about_ca_system_score_codex":0.005491282,"about_ca_system_score_gemma":0.023438688,"threshold_uncertainty_score":0.1662659},"labels":[],"label_agreement":null},{"id":"W1994851363","doi":"10.1007/s00769-005-0045-0","title":"Chemical and environmental sampling: quality through accreditation, certification and industrial standards","year":2005,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Nordic Life Science Pipeline (Canada)","funders":"European Commission","keywords":"Accreditation; Sampling (signal processing); Certification; Quality assurance; Lot quality assurance sampling; Quality (philosophy); Product certification; Computer science; Engineering; Engineering management; Business; Sampling design; Marketing; Service (business); Political science; Telecommunications; Environmental health; Medicine","score_opus":0.3186829572167391,"score_gpt":0.4813249204239368,"score_spread":0.1626419632071977,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1994851363","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23096085,0.010910757,0.693167,0.019651072,0.00067521934,0.000881948,0.00065697194,0.0014075362,0.041688714],"genre_scores_gemma":[0.78564525,0.0033959972,0.19701523,0.00186438,0.0003371262,0.0002685342,0.00063124957,0.0001648295,0.010677296],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9629301,0.017431585,0.0014495048,0.0014261227,0.016229512,0.0005332514],"domain_scores_gemma":[0.9619631,0.015173369,0.0048464844,0.0033917928,0.013842343,0.0007829166],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03571277,0.00032942707,0.0006417879,0.0025201975,0.0011423232,0.0038317116,0.0011062593,0.0011752945,0.0009923961],"category_scores_gemma":[0.03706427,0.00033772577,0.00034645866,0.002498119,0.0026539417,0.0014374843,0.0012554952,0.001351081,0.00037159843],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002686961,0.0004636427,0.11101873,0.00071340654,0.00015035717,0.00004995626,0.0008128736,0.008882091,0.013181815,0.059400696,0.011903895,0.7931539],"study_design_scores_gemma":[0.00020156476,0.0025975714,0.36181098,0.0009242351,0.0003722663,0.0007538433,0.0028617368,0.13097832,0.16001171,0.12829441,0.21090268,0.0002906962],"about_ca_topic_score_codex":0.011982582,"about_ca_topic_score_gemma":0.018946998,"teacher_disagreement_score":0.03571277,"about_ca_system_score_codex":0.0033528446,"about_ca_system_score_gemma":0.009941885,"threshold_uncertainty_score":0.18886942},"labels":[],"label_agreement":null},{"id":"W1997603456","doi":"10.1007/s00769-007-0282-5","title":"Analysis of 19-norandrosterone in human urine by gas chromatography–isotope-dilution mass spectrometry: method adopted by LGC for participation in the Comité Consultatif pour la Quantité de Matière (CCQM) Pilot Study P68","year":2007,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Mass Spectrometry Techniques and Applications","field":"Chemistry","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"World Anti-Doping Agency","keywords":"Isotope dilution; Chromatography; Chemistry; Mass spectrometry; Calibration; Analytical Chemistry (journal); Mathematics","score_opus":0.039330071977626696,"score_gpt":0.3834681268113606,"score_spread":0.34413805483373394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1997603456","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.74975353,0.02609627,0.19407693,0.002424696,0.002264947,0.002288137,0.007363775,0.0021049415,0.013626797],"genre_scores_gemma":[0.8602557,0.0048944606,0.10812849,0.0026676327,0.0003132417,0.0013162995,0.004764962,0.00040323383,0.017255982],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9973226,0.001059946,0.00013774275,0.00045818262,0.0008309043,0.000190548],"domain_scores_gemma":[0.9989197,0.00019695617,0.00007309896,0.00017850038,0.0004590807,0.00017268442],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0022233075,0.0010006154,0.00079860893,0.0018690813,0.0021209086,0.00087996334,0.0008669024,0.0015668373,0.0017900108],"category_scores_gemma":[0.0032177425,0.0005958794,0.00066827354,0.0012505315,0.001097426,0.00042459104,0.00060769037,0.0010514045,0.0013215044],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0029842074,0.00035243513,0.023599507,0.00027400485,0.0002736933,0.0003620213,0.00043801614,0.0003165522,0.9374658,0.00077117386,0.0014361326,0.031726457],"study_design_scores_gemma":[0.00020968834,0.0017639048,0.05091354,0.000085728345,0.0002022701,0.0016013454,0.00019417668,0.0021276218,0.9264409,0.0003555565,0.016034458,0.000070791786],"about_ca_topic_score_codex":0.0074122557,"about_ca_topic_score_gemma":0.01384518,"teacher_disagreement_score":0.0074122557,"about_ca_system_score_codex":0.0009424827,"about_ca_system_score_gemma":0.0030499771,"threshold_uncertainty_score":0.014738202},"labels":[],"label_agreement":null},{"id":"W2009412568","doi":"10.1007/s00769-012-0906-2","title":"Determination of the standard deviation for proficiency assessment from past participant’s performances","year":2012,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"Standard deviation; Reproducibility; Statistics; Relative standard deviation; Mathematics; Mercury (programming language); Analytical Chemistry (journal); Computer science; Chromatography; Chemistry; Detection limit","score_opus":0.08245211673099916,"score_gpt":0.34273756272306555,"score_spread":0.2602854459920664,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2009412568","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.112848476,0.0013302135,0.8763476,0.0002218207,0.00020023363,0.00031932176,0.0009080591,0.0016533934,0.00617081],"genre_scores_gemma":[0.78400564,0.00027444752,0.21077923,0.00016016346,0.000045153352,0.00066647964,0.0016633214,0.00037454852,0.0020310148],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9597782,0.016968803,0.0024529847,0.0055588284,0.0145431515,0.0006981522],"domain_scores_gemma":[0.9269051,0.047415324,0.0040121092,0.010598567,0.0106534315,0.00041549455],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.040217727,0.0012177147,0.0013785904,0.0028706752,0.0007868254,0.0014881883,0.0029012028,0.001664799,0.0012517842],"category_scores_gemma":[0.088515565,0.00049376994,0.0020965391,0.0016857353,0.0010626024,0.0015311285,0.0016388,0.0014629407,0.0011462551],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014596665,0.0008204159,0.23351228,0.0010923183,0.0024976986,0.00037400794,0.0026224318,0.11220051,0.05214886,0.012850833,0.008878953,0.571542],"study_design_scores_gemma":[0.00010198923,0.0037987714,0.26021907,0.0004288808,0.00082838297,0.0024997662,0.0008891965,0.57174134,0.11670751,0.014763217,0.02737242,0.00064943125],"about_ca_topic_score_codex":0.0032591936,"about_ca_topic_score_gemma":0.0025975383,"teacher_disagreement_score":0.95978224,"about_ca_system_score_codex":0.0012160387,"about_ca_system_score_gemma":0.0013855459,"threshold_uncertainty_score":0.21269423},"labels":[],"label_agreement":null},{"id":"W2010598206","doi":"10.1007/s00769-010-0669-6","title":"Conceptual thinking and metrology concepts","year":2010,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Statistics Education and Methodologies","field":"Mathematics","cited_by":11,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Rehabilitation Institute; University of Toronto","funders":"","keywords":"Competence (human resources); Computer science; Enabling; Operationalization; Metrology; Management science; Engineering; Psychology; Mathematics","score_opus":0.20620552319380506,"score_gpt":0.47691645462167703,"score_spread":0.27071093142787195,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2010598206","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05207562,0.008453377,0.5797103,0.037864156,0.0009662509,0.00013637365,0.00022062143,0.00024677694,0.32032654],"genre_scores_gemma":[0.86129797,0.0017377909,0.12079853,0.0019603935,0.0004729922,0.000359909,0.00019220817,0.000074256866,0.01310588],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99574834,0.0027669235,0.00022414292,0.00044787538,0.0006140272,0.00019875924],"domain_scores_gemma":[0.99259007,0.0051173563,0.0005419318,0.00058140577,0.00094057084,0.0002286276],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0049030576,0.0007087705,0.00048388963,0.0029708375,0.0025487817,0.007036,0.0017212451,0.0023859572,0.0056777042],"category_scores_gemma":[0.009244273,0.0004435069,0.0008403675,0.0023742465,0.021093937,0.011101898,0.0022321153,0.0037557597,0.0006368212],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000017881091,0.0000026533185,0.000024355037,0.000012401881,0.000001217624,0.0000075621915,0.00077149284,0.00006876651,0.00002427793,0.99791497,0.00020555728,0.0009649821],"study_design_scores_gemma":[0.0000035782634,0.00000343626,0.000045550518,0.000022171092,0.000001881348,0.000030964646,0.00049541355,0.00046273816,0.00005796013,0.9930068,0.005866131,0.0000032437968],"about_ca_topic_score_codex":0.0016080023,"about_ca_topic_score_gemma":0.0009988438,"teacher_disagreement_score":0.007036,"about_ca_system_score_codex":0.003860403,"about_ca_system_score_gemma":0.0022157244,"threshold_uncertainty_score":0.028009295},"labels":[],"label_agreement":null},{"id":"W2021927114","doi":"10.1007/s00769-008-0450-2","title":"Uncertainties when using dual-wavelength or single-wavelength spectrophotometry to monitor the course of hexavalent chromium reduction mediated by zero-valent iron","year":2008,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Environmental remediation with nanomaterials","field":"Engineering","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Hexavalent chromium; Spectrophotometry; Chromium; Zerovalent iron; Chemistry; Wavelength; Analytical Chemistry (journal); Contamination; Nuclear chemistry; Environmental chemistry; Chromatography; Optics; Physics; Organic chemistry","score_opus":0.03931189639586145,"score_gpt":0.2735875994995712,"score_spread":0.23427570310370974,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021927114","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5773028,0.003593741,0.40346223,0.0009339794,0.0010672155,0.00029097951,0.0009135005,0.00206471,0.010370861],"genre_scores_gemma":[0.9238038,0.00073182065,0.07211461,0.0004967066,0.000044819528,0.00018272953,0.00053552864,0.0002976619,0.0017923291],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.98491067,0.0033607762,0.00089720835,0.0033707812,0.006923396,0.0005371891],"domain_scores_gemma":[0.97479635,0.015138554,0.0018196894,0.0038579516,0.0041491347,0.00023841101],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014287237,0.0009967914,0.00069735653,0.0016864626,0.0011212988,0.0016062098,0.0017211395,0.0017065469,0.0004569538],"category_scores_gemma":[0.020965895,0.0006698361,0.0010020002,0.0010151616,0.0014339561,0.00078119524,0.0009123086,0.0014851332,0.0002481674],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019516009,0.00040369426,0.040065642,0.0006685336,0.0009490141,0.00023691641,0.0013988737,0.01749468,0.8625424,0.0048878277,0.0010801533,0.06832055],"study_design_scores_gemma":[0.000048766397,0.0005056766,0.033579994,0.00007708003,0.00043196042,0.00030786736,0.00026384293,0.031851586,0.9268405,0.0022903658,0.00368775,0.00011460307],"about_ca_topic_score_codex":0.008002192,"about_ca_topic_score_gemma":0.016859056,"teacher_disagreement_score":0.014287237,"about_ca_system_score_codex":0.002648438,"about_ca_system_score_gemma":0.0013304644,"threshold_uncertainty_score":0.07555902},"labels":[],"label_agreement":null},{"id":"W2033558742","doi":"10.1007/s00769-009-0522-y","title":"Report of the TrainMiC® Convention, held in Celje, Slovenia (10–14 January 2009)","year":2009,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Technology Assessment and Management","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"123 Certification (Canada)","funders":"Joint Research Centre","keywords":"License; Convention; Interactivity; European commission; Commission; Position (finance); Political science; Medical education; Public relations; Focus (optics); Engineering; Business; Computer science; Law; Multimedia; European union; Medicine; Finance","score_opus":0.016889053888468248,"score_gpt":0.28005497479656005,"score_spread":0.2631659209080918,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2033558742","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15754871,0.023262288,0.012612641,0.10322162,0.06618351,0.004907968,0.10201735,0.0023683596,0.52787757],"genre_scores_gemma":[0.14034995,0.0041949153,0.0077728746,0.012036384,0.002398773,0.001612189,0.02669508,0.00058029004,0.8043595],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9969547,0.0002754124,0.00020291393,0.00028950055,0.0013583591,0.0009191337],"domain_scores_gemma":[0.99722433,0.00017281092,0.00027851367,0.00017010301,0.0014905154,0.00066377263],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0055177426,0.00079257647,0.00053757837,0.0014836384,0.0031297987,0.005966742,0.0017379441,0.0053604785,0.013241326],"category_scores_gemma":[0.0047733868,0.00048578635,0.00069283147,0.0011844706,0.0010648703,0.0014251875,0.00273404,0.0044052945,0.004424854],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00045858862,0.00019849381,0.0065214676,0.0005606526,0.00005589774,0.0005651163,0.0010462756,0.0006894786,0.0049781534,0.0074164644,0.9544406,0.023068821],"study_design_scores_gemma":[0.000048948812,0.00010252046,0.056871355,0.00020605251,0.000037356134,0.00008804554,0.0010944625,0.00016783839,0.0017317794,0.00041876762,0.9391727,0.00006016163],"about_ca_topic_score_codex":0.16495754,"about_ca_topic_score_gemma":0.22719681,"teacher_disagreement_score":0.16495754,"about_ca_system_score_codex":0.0045029433,"about_ca_system_score_gemma":0.016519181,"threshold_uncertainty_score":0.3279947},"labels":[],"label_agreement":null},{"id":"W2036990291","doi":"10.1007/s00769-002-0492-9","title":"Addressing issues associated with the development and management of PT programs and with their optimum use","year":2002,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Quality and Safety in Healthcare","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institutes of Health","keywords":"Atlanta; Quality (philosophy); Library science; Political science; Engineering management; Engineering; Geography; Computer science; Metropolitan area","score_opus":0.376334339508065,"score_gpt":0.4494525600640289,"score_spread":0.0731182205559639,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036990291","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1297837,0.011691198,0.19522293,0.55596143,0.0024164442,0.0030138467,0.00032250705,0.001057816,0.10053013],"genre_scores_gemma":[0.73775876,0.008320949,0.19598588,0.025537692,0.0038210882,0.003463617,0.00021087375,0.00023424062,0.024666896],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.93199825,0.037833273,0.0062585627,0.0034006615,0.014470492,0.0060387775],"domain_scores_gemma":[0.6911575,0.12616104,0.048029613,0.014534874,0.06410216,0.056014765],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06943312,0.00058047747,0.00050818705,0.003504414,0.0057156086,0.013248424,0.0059662983,0.0068602553,0.01056646],"category_scores_gemma":[0.15074037,0.0006413186,0.00068392203,0.0026476837,0.0037957446,0.0105127655,0.006819474,0.0088031255,0.0016818054],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025661808,0.0024521218,0.06804495,0.001637612,0.0000677413,0.0006922005,0.0055724117,0.00831876,0.0030761443,0.058612324,0.03234458,0.8189245],"study_design_scores_gemma":[0.00045747292,0.0076797814,0.27730104,0.010584258,0.00028025205,0.00920534,0.058805645,0.036137864,0.013439871,0.15825182,0.42728084,0.00057588663],"about_ca_topic_score_codex":0.0070435395,"about_ca_topic_score_gemma":0.014757545,"teacher_disagreement_score":0.06943312,"about_ca_system_score_codex":0.00915891,"about_ca_system_score_gemma":0.07932888,"threshold_uncertainty_score":0.36720186},"labels":[],"label_agreement":null},{"id":"W2061797984","doi":"10.1007/s00769-002-0500-0","title":"Global Odyssey 2002: International conference on proficiency testing for medical laboratories","year":2002,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Quality and Safety in Healthcare","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Commercial law; Engineering; Medical laboratory; Engineering ethics; Library science; Medical physics; Management science; Political science; Computer science; Medicine; Pathology; Law","score_opus":0.30388766781366894,"score_gpt":0.5129245037421812,"score_spread":0.2090368359285123,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2061797984","genre_codex":"commentary","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03353198,0.0380993,0.16678242,0.42600858,0.10003214,0.005946563,0.007159696,0.004493809,0.21794552],"genre_scores_gemma":[0.16781768,0.022066783,0.10436877,0.08511032,0.014219693,0.0029362487,0.020533793,0.0025720792,0.58037466],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9859822,0.0036724326,0.00090920494,0.001150028,0.006149475,0.002136747],"domain_scores_gemma":[0.95605695,0.003796566,0.0021309066,0.0019730101,0.0263847,0.009657821],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.040544495,0.002011633,0.0013786395,0.0030705652,0.0040634554,0.009890519,0.0039041147,0.007942558,0.02387146],"category_scores_gemma":[0.02022039,0.0010049494,0.0010241249,0.0021013273,0.0040207156,0.004125343,0.009342155,0.007926058,0.007852712],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00026043068,0.000388599,0.003088005,0.00038316296,0.000030994594,0.00023894863,0.00063851144,0.00083350134,0.00415764,0.006803018,0.88324714,0.09993009],"study_design_scores_gemma":[0.000047118032,0.00020002268,0.007373716,0.00040212003,0.000022126169,0.00021102924,0.0006789223,0.0009498282,0.0024824403,0.002707382,0.98488486,0.000040338447],"about_ca_topic_score_codex":0.04336838,"about_ca_topic_score_gemma":0.0621889,"teacher_disagreement_score":0.04336838,"about_ca_system_score_codex":0.005690303,"about_ca_system_score_gemma":0.053322326,"threshold_uncertainty_score":0.21442235},"labels":[],"label_agreement":null},{"id":"W2074327302","doi":"10.1007/s00769-012-0899-x","title":"Using target measurement uncertainty to determine fitness for purpose","year":2012,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":16,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dynamic Systems Analysis (Canada)","funders":"","keywords":"Measurement uncertainty; Process (computing); Uncertainty analysis; Guard (computer science); Computer science; Range (aeronautics); Propagation of uncertainty; Data mining; Risk analysis (engineering); Engineering; Statistics; Simulation; Mathematics; Algorithm","score_opus":0.19003194181979322,"score_gpt":0.34748597376685303,"score_spread":0.1574540319470598,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2074327302","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.52354735,0.0005888298,0.4508214,0.00082057307,0.00016063427,0.0003337536,0.0007235071,0.0015322525,0.021471646],"genre_scores_gemma":[0.97426856,0.000037260877,0.024507347,0.000083313564,0.0000096558415,0.00007578136,0.00031893796,0.00013594983,0.0005633421],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.9812278,0.0062718457,0.0011203594,0.002506163,0.007881244,0.0009925736],"domain_scores_gemma":[0.9202581,0.050505906,0.0055342503,0.01011944,0.012886153,0.00069621793],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.025423432,0.0011474082,0.0009133261,0.0025896353,0.0011618205,0.0037592584,0.0013931209,0.0022379167,0.0018957811],"category_scores_gemma":[0.12103899,0.0005064708,0.0016426226,0.0013918543,0.0011116597,0.0031770742,0.0024261996,0.0014322605,0.000628864],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0022431135,0.00049905153,0.5478384,0.0004970073,0.0013903971,0.0003993505,0.0015915416,0.16125578,0.026802516,0.023004752,0.0025989765,0.2318791],"study_design_scores_gemma":[0.0001368198,0.0017580054,0.23583312,0.0003327508,0.0007897738,0.0007428396,0.0010011875,0.6511582,0.06572732,0.034295157,0.007944294,0.00028057754],"about_ca_topic_score_codex":0.005438514,"about_ca_topic_score_gemma":0.0032543698,"teacher_disagreement_score":0.025423432,"about_ca_system_score_codex":0.0016064905,"about_ca_system_score_gemma":0.0022051225,"threshold_uncertainty_score":0.1344536},"labels":[],"label_agreement":null},{"id":"W2080442435","doi":"10.1007/s007690050442","title":"Reference materials in environmental trace organic analysis","year":2000,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Water Quality Monitoring and Analysis","field":"Environmental Science","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Research Council Canada; Bundesministerium für Bildung und Forschung","keywords":"TRACE (psycholinguistics); Quality assurance; Environmental quality; Pollutant; Environmental science; Environmental analysis; Environmental chemistry; Engineering; Chemistry; Political science; Law","score_opus":0.025781274176230863,"score_gpt":0.2754081212659894,"score_spread":0.24962684708975855,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2080442435","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.040631402,0.18215732,0.46530092,0.011231378,0.022850834,0.002143135,0.008128138,0.008316889,0.25924],"genre_scores_gemma":[0.18975258,0.10019453,0.46975002,0.01002349,0.0033076468,0.0025678251,0.023329055,0.0019951276,0.19907975],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99192166,0.0027158775,0.0007627658,0.0005307526,0.0036108934,0.0004580732],"domain_scores_gemma":[0.99335116,0.0012768302,0.00053910655,0.001238886,0.0033241091,0.0002699698],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007082636,0.0017509669,0.0013357411,0.01095006,0.0027950823,0.0028076929,0.005243573,0.006464929,0.01594454],"category_scores_gemma":[0.01313953,0.000934385,0.0009080583,0.0057962416,0.0012273098,0.0034666078,0.0029437826,0.0025885752,0.01834927],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041897877,0.00067687465,0.0021708128,0.005318385,0.000116157404,0.0023080718,0.0004281626,0.0033592272,0.10566305,0.08290523,0.12905622,0.6675788],"study_design_scores_gemma":[0.0000414514,0.00028370417,0.0010244269,0.0013431106,0.00012452871,0.0012037666,0.00020308496,0.00188196,0.16979785,0.014642839,0.8093879,0.00006529167],"about_ca_topic_score_codex":0.0042782268,"about_ca_topic_score_gemma":0.0066618533,"teacher_disagreement_score":0.01594454,"about_ca_system_score_codex":0.0013696359,"about_ca_system_score_gemma":0.003894832,"threshold_uncertainty_score":0.05333978},"labels":[],"label_agreement":null},{"id":"W2083486886","doi":"10.1007/s00769-007-0283-4","title":"The WADA Proficiency Testing Program as an integral part of the fight against doping in sport","year":2007,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Hormonal and reproductive studies","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"World Anti-Doping Agency","funders":"","keywords":"Accreditation; Context (archaeology); Wada test; Agency (philosophy); Psychology; Medicine; Medical education; Epistemology; History; Philosophy","score_opus":0.060638034808893605,"score_gpt":0.37427991111969716,"score_spread":0.31364187631080354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2083486886","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.49330303,0.011579207,0.061144248,0.17797191,0.008273783,0.0052408264,0.0034622457,0.004810239,0.23421447],"genre_scores_gemma":[0.81074935,0.0023880762,0.041384287,0.0220197,0.0007794835,0.0017037345,0.0019140645,0.00029545228,0.11876581],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99425757,0.00236561,0.0004165646,0.00039980837,0.0018822032,0.0006781996],"domain_scores_gemma":[0.96112704,0.0069230213,0.0041369456,0.00234544,0.019491795,0.005975791],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01384945,0.0003368041,0.00027872925,0.0025545622,0.002426645,0.0025488737,0.0017176609,0.0020269868,0.0044990503],"category_scores_gemma":[0.019329302,0.00030076457,0.00027065407,0.000980083,0.0012818762,0.001141383,0.0017584814,0.0030207816,0.001403151],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004510413,0.0019120061,0.19945993,0.00042380544,0.00006535472,0.0004063987,0.0020847796,0.0011897689,0.020441806,0.008014434,0.100717425,0.66483325],"study_design_scores_gemma":[0.00011348879,0.002681479,0.61579704,0.0006422239,0.000046639714,0.000966833,0.0032645897,0.0024112447,0.011605805,0.0031576916,0.3592078,0.00010509339],"about_ca_topic_score_codex":0.024299126,"about_ca_topic_score_gemma":0.054292448,"teacher_disagreement_score":0.024299126,"about_ca_system_score_codex":0.0019320971,"about_ca_system_score_gemma":0.023682695,"threshold_uncertainty_score":0.0732438},"labels":[],"label_agreement":null},{"id":"W2086650703","doi":"10.1007/s00769-013-1031-6","title":"TDRM/TDLM Workshop at the AOAC INTERNATIONAL Annual Meeting 2013","year":2014,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Meteorological Phenomena and Simulations","field":"Earth and Planetary Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"National Research Council Canada","funders":"","keywords":"Commercial law; Library science; Engineering; Analytical Chemistry (journal); Political science; Computer science; Chemistry; Environmental chemistry; Law","score_opus":0.03974803267383914,"score_gpt":0.28924179587303483,"score_spread":0.2494937631991957,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086650703","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04723317,0.01249828,0.2511209,0.16906027,0.08748543,0.0019450785,0.010369912,0.008898293,0.41138873],"genre_scores_gemma":[0.12550151,0.005430532,0.091506824,0.010158083,0.0074537564,0.0008306832,0.010435844,0.0023492328,0.7463335],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99670404,0.0007467726,0.000105532046,0.00030608318,0.001635076,0.00050252693],"domain_scores_gemma":[0.9936699,0.0002782381,0.00019885803,0.00038526845,0.0043137306,0.0011539579],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011276883,0.00095054257,0.0007427839,0.001125527,0.0025176876,0.004831012,0.0030585248,0.0027356083,0.056062154],"category_scores_gemma":[0.0057764384,0.0005229302,0.0010106311,0.00076501776,0.0009482657,0.0026558936,0.004792187,0.003581618,0.022433378],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00045461024,0.0006017171,0.0026425305,0.00030177334,0.000038655435,0.00031043962,0.0008495615,0.004258214,0.010872993,0.0069055427,0.76626223,0.20650172],"study_design_scores_gemma":[0.00004124149,0.00014985567,0.0036361192,0.00033592,0.000021326316,0.0001122533,0.001419737,0.0054862886,0.008008329,0.003492624,0.97723174,0.00006450365],"about_ca_topic_score_codex":0.026506662,"about_ca_topic_score_gemma":0.046560895,"teacher_disagreement_score":0.056062154,"about_ca_system_score_codex":0.003441306,"about_ca_system_score_gemma":0.013040992,"threshold_uncertainty_score":0.18754661},"labels":[],"label_agreement":null},{"id":"W2092953325","doi":"10.1007/s00769-014-1049-4","title":"Harmonization and transferability of performance assessment: experience from four serum aluminum proficiency testing schemes","year":2014,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institut National de Santé Publique du Québec","funders":"","keywords":"Harmonization; Standard deviation; Transferability; Computer science; Statistics; Reliability engineering; Mathematics; Engineering","score_opus":0.06021817124645472,"score_gpt":0.288018029644113,"score_spread":0.22779985839765826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2092953325","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9728074,0.00039235718,0.017820565,0.0012143223,0.00004521118,0.0012384887,0.00009611913,0.00017542069,0.0062100496],"genre_scores_gemma":[0.9780131,0.0001770983,0.01712929,0.00036586283,0.00002149846,0.00026554032,0.00030065735,0.00007382626,0.003653158],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9352406,0.04560613,0.0025403856,0.003805119,0.0090854075,0.003722337],"domain_scores_gemma":[0.9398355,0.025928084,0.0033884526,0.008696728,0.019333838,0.0028174794],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0883948,0.0008482941,0.0006001951,0.0012197839,0.004021722,0.003008529,0.003999805,0.0022398566,0.0017300491],"category_scores_gemma":[0.05780286,0.00043663752,0.00082133495,0.0014561296,0.0029284146,0.0014451741,0.004608922,0.0018703715,0.00067358295],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0070716357,0.021061804,0.13272373,0.0008201332,0.00037981372,0.0011061435,0.08705256,0.0109380465,0.05892919,0.0060580224,0.0059048347,0.66795415],"study_design_scores_gemma":[0.0027651084,0.08936017,0.39094654,0.00071533874,0.0008763645,0.0033141214,0.06428467,0.026215978,0.26771882,0.00692768,0.14636824,0.0005069184],"about_ca_topic_score_codex":0.016429765,"about_ca_topic_score_gemma":0.011947322,"teacher_disagreement_score":0.9116052,"about_ca_system_score_codex":0.0064019556,"about_ca_system_score_gemma":0.013098402,"threshold_uncertainty_score":0.46748197},"labels":[],"label_agreement":null},{"id":"W2897236559","doi":"10.1007/s00769-018-1349-1","title":"Cooperation in publicly funded reference material production","year":2018,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Scientific Measurement and Uncertainty Evaluation","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"National Research Council Canada","funders":"International Atomic Energy Agency; Gobierno del Principado de Asturias","keywords":"Certification; Production (economics); European commission; Commercial law; Accounting; Business; Commission; Library science; Engineering; Political science; Computer science; European union; Finance; Economics; Law","score_opus":0.47412113293856345,"score_gpt":0.48009324608865606,"score_spread":0.005972113150092606,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2897236559","genre_codex":"other","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.14715968,0.017353907,0.07038068,0.19874363,0.018061833,0.0034237423,0.0034035984,0.0024184273,0.5390545],"genre_scores_gemma":[0.5680918,0.0041056992,0.040048163,0.013632498,0.0031324946,0.0025569296,0.007193468,0.0014572099,0.3597817],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.8424703,0.09942668,0.006452507,0.008176176,0.031820595,0.011653662],"domain_scores_gemma":[0.8832509,0.05037913,0.0077640866,0.0219173,0.025716426,0.010972176],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13212174,0.00095561444,0.001164433,0.0024338723,0.006335744,0.011639723,0.005441964,0.009334939,0.041519847],"category_scores_gemma":[0.09741312,0.000854434,0.0012217171,0.0037689875,0.002846045,0.0075370795,0.023835184,0.0041797585,0.010243843],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011376103,0.0007378301,0.0066982647,0.0018626645,0.00013568268,0.0039156256,0.043797873,0.0028502166,0.011040664,0.11114379,0.38894373,0.42773604],"study_design_scores_gemma":[0.00008790009,0.00027444752,0.0035642204,0.0006416588,0.000022674918,0.0005431263,0.009303746,0.0009365111,0.0028123197,0.010539067,0.97120124,0.00007310038],"about_ca_topic_score_codex":0.004195128,"about_ca_topic_score_gemma":0.004506418,"teacher_disagreement_score":0.86787826,"about_ca_system_score_codex":0.008972411,"about_ca_system_score_gemma":0.024049584,"threshold_uncertainty_score":0.6987349},"labels":[],"label_agreement":null},{"id":"W2900128974","doi":"10.1007/s00769-018-1357-1","title":"A proficiency testing scheme to evaluate the effectiveness of laboratory sample reduction of a soil sample","year":2018,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Pesticide Residue Analysis and Safety","field":"Agricultural and Biological Sciences","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Environment and Climate Change Canada","keywords":"Sample (material); Accreditation; Homogeneity (statistics); Computer science; Engineering; Reliability engineering; Statistics; Mathematics; Chemistry; Chromatography; Medical education; Medicine","score_opus":0.060860853854059795,"score_gpt":0.3278827809757692,"score_spread":0.26702192712170936,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2900128974","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9102661,0.00013281757,0.066199645,0.00041900174,0.00013820104,0.013260039,0.0014673318,0.0012566228,0.0068602543],"genre_scores_gemma":[0.80643135,0.00015507056,0.16402332,0.00086262234,0.00006809229,0.008980704,0.00560852,0.00015895216,0.013711461],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.98819065,0.004952987,0.0011834848,0.0017667917,0.003453648,0.0004524076],"domain_scores_gemma":[0.97917706,0.005002711,0.0019569872,0.0039792005,0.008993267,0.0008907727],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015459241,0.0010974816,0.00068957015,0.0013535757,0.0018609673,0.0007467488,0.0015327637,0.002202956,0.0036916411],"category_scores_gemma":[0.015970148,0.0004459556,0.00083421107,0.0007617137,0.0013224739,0.00068750675,0.0010754408,0.0008881082,0.0019530971],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0042816484,0.009983501,0.03878324,0.00029209777,0.00013721758,0.0001712469,0.0010329784,0.0017852599,0.8903419,0.00074479426,0.00195555,0.050490532],"study_design_scores_gemma":[0.0008142719,0.067085065,0.054204136,0.00006391003,0.00029167748,0.000828346,0.00048474793,0.008570116,0.8551425,0.00032363576,0.0120618045,0.00012983078],"about_ca_topic_score_codex":0.0025504709,"about_ca_topic_score_gemma":0.0029761544,"teacher_disagreement_score":0.015459241,"about_ca_system_score_codex":0.0009688626,"about_ca_system_score_gemma":0.0030740502,"threshold_uncertainty_score":0.08175725},"labels":[],"label_agreement":null},{"id":"W2964970392","doi":"10.1007/s00769-019-01397-5","title":"Non-carcinogenic health risk assessment of nitrate in bottled drinking waters sold in Iranian markets: A Monte Carlo simulation","year":2019,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Water Treatment and Disinfection","field":"Environmental Science","cited_by":65,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Hospital for Sick Children","keywords":"Hazard quotient; Nitrate; Bottled water; Health risk assessment; Health risk; Toxicology; Environmental health; Risk assessment; Environmental science; Human health; Environmental chemistry; Chemistry; Medicine; Environmental engineering; Biology","score_opus":0.0177469213794428,"score_gpt":0.3059776679947836,"score_spread":0.2882307466153408,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2964970392","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97929907,0.0003863573,0.012804596,0.00027271372,0.000031854852,0.00007838075,0.0005911574,0.000060713854,0.0064752186],"genre_scores_gemma":[0.99428505,0.00015350459,0.0025495063,0.00005317939,0.00000955318,0.000050516755,0.00024997036,0.000014106531,0.0026345411],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9997055,0.00010607003,0.00001092692,0.000057953708,0.000041695865,0.00007783567],"domain_scores_gemma":[0.9974401,0.0019750884,0.00020517691,0.000057614314,0.0002453123,0.0000766753],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011015602,0.0006148542,0.00094536634,0.0008850456,0.0006931631,0.0011280747,0.0013915038,0.0016582625,0.0030018003],"category_scores_gemma":[0.0027098726,0.00081477495,0.0018286,0.00074824115,0.0006557823,0.00073680514,0.0005016618,0.0008625248,0.0001731491],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009471183,0.000063936466,0.004150545,0.00002106612,0.000042876592,0.00006560265,0.000017780034,0.993768,0.00016108935,0.00069954304,0.00012387187,0.00079107046],"study_design_scores_gemma":[0.000042922504,0.00008781671,0.0016348487,0.0000068624936,0.00005824081,0.00002586531,0.00004642043,0.99721247,0.0002158908,0.0005293087,0.00012574701,0.000013621757],"about_ca_topic_score_codex":0.12865442,"about_ca_topic_score_gemma":0.06313994,"teacher_disagreement_score":0.12865442,"about_ca_system_score_codex":0.002135216,"about_ca_system_score_gemma":0.0019753012,"threshold_uncertainty_score":0.25581115},"labels":[],"label_agreement":null},{"id":"W3028443178","doi":"10.1007/s00769-020-01433-9","title":"Development of low and elevated level multivitamin and mineral supplement certified reference materials: VITA-1 and VITB-1","year":2020,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Vitamin D Research Studies","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"National Research Council Canada","funders":"","keywords":"Certified reference materials; Multivitamin; Analyte; Chemistry; Mathematics; Chromatography; Vitamin; Detection limit","score_opus":0.18094198931547836,"score_gpt":0.3795725873741518,"score_spread":0.19863059805867342,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3028443178","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.55092394,0.011134724,0.37528467,0.0038101994,0.0018671002,0.0062206476,0.00990952,0.003658929,0.037190255],"genre_scores_gemma":[0.51156926,0.0026714285,0.4350293,0.0031063235,0.00020414389,0.003660971,0.019597646,0.00064339506,0.023517588],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.97996855,0.0063799424,0.0017836792,0.0028026672,0.0084996475,0.0005655858],"domain_scores_gemma":[0.9834087,0.0021902134,0.0020970847,0.0014753076,0.010359799,0.0004690112],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.020555131,0.0011465417,0.0008257421,0.0055977204,0.0018266889,0.0029217007,0.0038181143,0.0036539587,0.0037489808],"category_scores_gemma":[0.025337106,0.0008093717,0.0009150334,0.0017048166,0.0014946153,0.001093607,0.0023592825,0.0017412622,0.002764275],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012458899,0.0023738043,0.11130804,0.0015379172,0.00026150187,0.00036572866,0.0016361132,0.0028331333,0.6261184,0.011580197,0.009836322,0.23090295],"study_design_scores_gemma":[0.00028680247,0.0029231093,0.080152266,0.0013392814,0.00050842995,0.0016214503,0.0011755592,0.008496134,0.74789375,0.0030629926,0.1523845,0.00015565797],"about_ca_topic_score_codex":0.0064645736,"about_ca_topic_score_gemma":0.009806589,"teacher_disagreement_score":0.020555131,"about_ca_system_score_codex":0.0014946237,"about_ca_system_score_gemma":0.006010272,"threshold_uncertainty_score":0.10870719},"labels":[],"label_agreement":null},{"id":"W4284689332","doi":"10.1007/s00769-022-01505-y","title":"Interlaboratory comparisons of chemical measurements: Quo Vadis?","year":2022,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Scientific Measurement and Uncertainty Evaluation","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"National Research Council Canada","funders":"National Research Council Canada","keywords":"Statistician; Criticism; The arts; Status quo; Value (mathematics); Management science; Engineering; Statistics; Political science; Mathematics; Law","score_opus":0.42892369295191496,"score_gpt":0.4588551741003888,"score_spread":0.029931481148473826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4284689332","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0028153607,0.10233895,0.035015058,0.66278803,0.18681471,0.00007365311,0.00025212084,0.00020865582,0.009693386],"genre_scores_gemma":[0.23694047,0.053324338,0.039438475,0.38034052,0.27505463,0.00033009984,0.0002833131,0.00079930766,0.013488807],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.8542489,0.07848942,0.009318948,0.011520525,0.044796202,0.001625954],"domain_scores_gemma":[0.68350697,0.2157528,0.012253225,0.015726108,0.070440315,0.002320557],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.13469502,0.0008300032,0.0023331272,0.002823829,0.0032703904,0.012619499,0.00359378,0.004884219,0.0031199623],"category_scores_gemma":[0.28033432,0.0005768412,0.0015986154,0.002907925,0.017206244,0.011192439,0.0044921977,0.014166731,0.0011005806],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00061070645,0.00009902478,0.00294334,0.003621677,0.0009854765,0.00024738046,0.005680471,0.0012671069,0.0015120464,0.3190765,0.43168908,0.23226728],"study_design_scores_gemma":[0.00007661244,0.00016779413,0.0022111926,0.0033334752,0.00018754473,0.00016735595,0.0032772066,0.001801386,0.0016042738,0.3259517,0.66105187,0.00016959663],"about_ca_topic_score_codex":0.005186459,"about_ca_topic_score_gemma":0.0050051543,"teacher_disagreement_score":0.86530495,"about_ca_system_score_codex":0.006466942,"about_ca_system_score_gemma":0.0076077557,"threshold_uncertainty_score":0.7123439},"labels":[],"label_agreement":null},{"id":"W4304614705","doi":"10.1007/s00769-022-01515-w","title":"The units of rate constants in chemical kinetics","year":2022,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Spectroscopy and Quantum Chemical Studies","field":"Physics and Astronomy","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Open Text (Canada)","funders":"","keywords":"Reaction rate constant; Arrhenius equation; Logarithm; Thermodynamics; Constant (computer programming); Base (topology); Rate equation; Kinetics; Chemistry; Mathematics; Physical chemistry; Computer science; Physics; Activation energy; Mathematical analysis","score_opus":0.03169094056643566,"score_gpt":0.3058511319264604,"score_spread":0.27416019136002473,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4304614705","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008494832,0.038147435,0.8015483,0.0068688127,0.010988134,0.00022731558,0.0010296687,0.0014368705,0.1312586],"genre_scores_gemma":[0.3773596,0.032563537,0.48555025,0.0064005405,0.008610075,0.0019218951,0.0013586348,0.003314564,0.08292092],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9913862,0.0032330398,0.00091349444,0.0018569512,0.0021841966,0.00042608372],"domain_scores_gemma":[0.98317415,0.008583406,0.0020506913,0.0034949495,0.002311215,0.0003855781],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008626996,0.003620703,0.0016915689,0.0034800658,0.0014175017,0.007859313,0.005763727,0.0034160858,0.015492291],"category_scores_gemma":[0.029337585,0.0013552675,0.002230194,0.0033268887,0.009035095,0.0195814,0.0031800703,0.013872139,0.007148747],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016215263,0.000011692645,0.000062482,0.00015623056,0.000011295738,0.000027598324,0.00012747591,0.00090385816,0.0006356702,0.9897453,0.0020703494,0.006231907],"study_design_scores_gemma":[0.000018664408,0.0000771133,0.00017410611,0.00020679695,0.00003638179,0.0002636856,0.00009539163,0.0071639386,0.0031671373,0.86160606,0.12711605,0.00007456179],"about_ca_topic_score_codex":0.0011053055,"about_ca_topic_score_gemma":0.0005091368,"teacher_disagreement_score":0.015492291,"about_ca_system_score_codex":0.0037977847,"about_ca_system_score_gemma":0.0016621518,"threshold_uncertainty_score":0.051826894},"labels":[],"label_agreement":null},{"id":"W4396736053","doi":"10.1007/s00769-024-01595-w","title":"New statistical framework for interlaboratory evaluation of anti-doping testing results by WADA","year":2024,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"World Anti-Doping Agency; National Research Council Canada","funders":"National Research Council Canada","keywords":"Commercial law; Statistical analysis; Analytical Chemistry (journal); Statistics; Engineering; Accounting; Mathematics; Chemistry; Business; Chromatography; Political science; Law","score_opus":0.6683357042775011,"score_gpt":0.6183950536970836,"score_spread":0.04994065058041752,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396736053","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012154682,0.00027422296,0.9972036,0.00036149085,0.000043545857,0.000056640347,0.000077505785,0.00007329519,0.0006941809],"genre_scores_gemma":[0.18481244,0.0011617823,0.8076814,0.0005841368,0.0005922829,0.0013287463,0.0006298743,0.00020749371,0.0030017528],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9257404,0.056611624,0.0029901064,0.005725678,0.007881188,0.0010510293],"domain_scores_gemma":[0.8145973,0.14903933,0.010617295,0.009431109,0.014994249,0.0013207226],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09493107,0.002021804,0.0029552025,0.006302426,0.0009782088,0.0066249697,0.0061012013,0.0026074986,0.0033176835],"category_scores_gemma":[0.14939138,0.0015750554,0.0029086482,0.0042724432,0.006601382,0.0053853057,0.0053859954,0.0047738585,0.0009132557],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007469963,0.00009010478,0.0028047222,0.00022855058,0.00035984887,0.00021563376,0.0003251661,0.0950547,0.00052792276,0.8550369,0.0019734327,0.043308306],"study_design_scores_gemma":[0.000023451536,0.00010424171,0.0009541986,0.000118914904,0.00008619504,0.00009309149,0.00006731564,0.6481696,0.0002781264,0.3453319,0.0047265547,0.000046496345],"about_ca_topic_score_codex":0.0069766114,"about_ca_topic_score_gemma":0.0042749373,"teacher_disagreement_score":0.09493107,"about_ca_system_score_codex":0.003952155,"about_ca_system_score_gemma":0.004971669,"threshold_uncertainty_score":0.5020495},"labels":[],"label_agreement":null},{"id":"W4403718788","doi":"10.1007/s00769-024-01621-x","title":"Production and certification of a freeze-dried reference material for human urinary steroids to ensure quality of longitudinal profiling data","year":2024,"lang":"en","type":"article","venue":"Accreditation and Quality Assurance","topic":"Hormonal and reproductive studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"World Anti-Doping Agency","keywords":"Certification; Profiling (computer programming); Certified reference materials; Commercial law; Quality assurance; Chromatography; Chemistry; Engineering; Process engineering; External quality assessment; Computer science; Operations management; Management; Political science; Economics","score_opus":0.267195335066961,"score_gpt":0.4401583683803643,"score_spread":0.1729630333134033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403718788","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20777658,0.0033043975,0.7291322,0.0026984783,0.002095829,0.011137696,0.013851827,0.007500239,0.02250281],"genre_scores_gemma":[0.26128852,0.002291817,0.63790494,0.0024536003,0.00059553614,0.0070720892,0.05438044,0.002464075,0.031549],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9850144,0.0039107264,0.0014829031,0.0018341043,0.0069491803,0.0008086168],"domain_scores_gemma":[0.961522,0.0033111945,0.0027685042,0.01272642,0.018706314,0.00096559763],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015829999,0.0010475552,0.0009275401,0.0053674006,0.003176955,0.0020075906,0.0026979411,0.0029196811,0.007178751],"category_scores_gemma":[0.026325049,0.0010057593,0.0013270248,0.0022932214,0.0024696095,0.0010519951,0.0017121622,0.0019286274,0.008594167],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047498447,0.0004462647,0.013231021,0.000495566,0.000088090834,0.00064617384,0.00090754183,0.0010075489,0.8888189,0.0028993534,0.0106547745,0.08032975],"study_design_scores_gemma":[0.00017374883,0.0010251969,0.03376909,0.00054274144,0.00017650754,0.0018089493,0.0003960682,0.002430722,0.8062338,0.001178257,0.15214516,0.00011972611],"about_ca_topic_score_codex":0.005933316,"about_ca_topic_score_gemma":0.006791444,"teacher_disagreement_score":0.015829999,"about_ca_system_score_codex":0.0014045667,"about_ca_system_score_gemma":0.006641261,"threshold_uncertainty_score":0.083718},"labels":[],"label_agreement":null}]}