{"meta":{"query_hash":"bfc1c98d030d","filters":{"venue":"Evaluation Review"},"cohort_total":27,"direct_labels_cover":1,"predictions_cover":27,"exported":27,"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/bfc1c98d030d","api":"https://metacan.xera.ac/api/v1/cohort?venue=Evaluation+Review"},"results":[{"id":"W15985522","doi":"10.1177/0193841x05276654","title":"The construction of social imaginaries on terrorism in newspapers from Montreal and Guadalajara : a comparative study among three cultural groups","year":2008,"lang":"en","type":"dissertation","venue":"Evaluation Review","topic":"Cultural and political discourse analysis","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Terrorism; Newspaper; Construct (python library); Media studies; Political science; The Imaginary; Reading (process); Sociology; Order (exchange); Law; Psychology","score_opus":0.07795567414794159,"score_gpt":0.4310312717312078,"score_spread":0.35307559758326623,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W15985522","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":[0.994478,0.00033038447,0.000050841954,0.00024327291,0.000011187885,0.00003686846,0.000036514106,0.0000023920752,0.004810572],"genre_scores_gemma":[0.9975073,0.00042016845,0.00008882238,0.000087627974,0.000015968046,0.00003201087,0.000036281162,0.000006083231,0.0018057895],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.99742067,0.001282727,0.000050271585,0.0001794169,0.000305985,0.000760977],"domain_scores_gemma":[0.995181,0.0022924005,0.00093481323,0.00015750741,0.00080879935,0.00062539685],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0031999957,0.0007243628,0.00046133576,0.0046038697,0.011380052,0.0043602907,0.0013717636,0.0008423148,0.0036405725],"category_scores_gemma":[0.0072903335,0.0004423184,0.00024213862,0.003926352,0.008130371,0.0012963606,0.003039924,0.0010927229,0.00017663278],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000047086203,0.000044834353,0.021161065,0.0000921905,0.0000101553505,0.00089754903,0.97064036,0.000015552623,0.00073362334,0.0004165046,0.00021299566,0.0057280445],"study_design_scores_gemma":[0.0000030999267,0.000045507415,0.07377825,0.000036752892,0.000008807169,0.00010523801,0.9217948,0.000015500667,0.00015830493,0.000027542847,0.004014794,0.0000113988235],"about_ca_topic_score_codex":0.57528925,"about_ca_topic_score_gemma":0.772145,"teacher_disagreement_score":0.57528925,"about_ca_system_score_codex":0.018807575,"about_ca_system_score_gemma":0.007657047,"threshold_uncertainty_score":0.8544239},"labels":[],"label_agreement":null},{"id":"W1983148067","doi":"10.1177/0193841x0002400103","title":"An Application of Theory-Driven Evaluation to a Drop-In Youth Center","year":2000,"lang":"en","type":"review","venue":"Evaluation Review","topic":"Youth Development and Social Support","field":"Social Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Montréal; McGill University","funders":"","keywords":"Drop out; Context (archaeology); Theory of change; Intervention (counseling); Program evaluation; Data collection; Psychology; Center (category theory); Management science; Applied psychology; Sociology; Political science; Engineering; Social science; Economics; Public administration","score_opus":0.1513130385005866,"score_gpt":0.490324462403093,"score_spread":0.3390114239025064,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983148067","genre_codex":"methods","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15551358,0.06326997,0.4559668,0.008000332,0.0023558086,0.2502271,0.00093447993,0.0006819795,0.06304999],"genre_scores_gemma":[0.35313296,0.019053869,0.53390515,0.0021785588,0.0000952695,0.08820722,0.00038983623,0.0001393753,0.0028977892],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.74976355,0.22220524,0.0055988915,0.0027905258,0.018298492,0.0013433009],"domain_scores_gemma":[0.8151452,0.1408849,0.004675804,0.00645602,0.03171938,0.0011187074],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.19952464,0.0016769137,0.0026037474,0.007768119,0.0024060921,0.004274416,0.0037767068,0.0015662436,0.0022966573],"category_scores_gemma":[0.21546793,0.00090250897,0.0016798403,0.0045125466,0.0029486765,0.0026784216,0.004647024,0.0018926276,0.00044772343],"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":[0.0006937516,0.0020295943,0.0041582226,0.032345314,0.00061845355,0.00046038526,0.03234264,0.0050275577,0.002155566,0.023359025,0.005285014,0.8915245],"study_design_scores_gemma":[0.008252993,0.03559115,0.04229523,0.17322053,0.0058038267,0.0023200535,0.1595198,0.04109171,0.05432069,0.10685471,0.36956483,0.0011645141],"about_ca_topic_score_codex":0.009177628,"about_ca_topic_score_gemma":0.01761253,"teacher_disagreement_score":0.19952464,"about_ca_system_score_codex":0.013931323,"about_ca_system_score_gemma":0.02829167,"threshold_uncertainty_score":0.9871286},"labels":[],"label_agreement":null},{"id":"W1990843122","doi":"10.1177/0193841x07307532","title":"The Interaction of Co-Occurring Mental Disorders and Recovery Management Checkups on Substance Abuse Treatment Participation and Recovery","year":2008,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Addiction and Mental Health","funders":"National Institute on Drug Abuse","keywords":"Substance abuse; Intervention (counseling); Psychiatry; Substance use; Mental health; Psychology; Clinical psychology; Randomized controlled trial; Substance abuse treatment; Medicine; Internal medicine","score_opus":0.08272507338278332,"score_gpt":0.3832698896296803,"score_spread":0.300544816246897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990843122","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":[0.99093175,0.0055879722,0.00029771624,0.0004876831,0.000040274226,0.00031286146,0.00009315725,0.000015944564,0.0022326643],"genre_scores_gemma":[0.9958891,0.0018468648,0.0008613963,0.0001323492,0.00007212355,0.0003659133,0.00006979609,0.0000050561566,0.00075726217],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9726299,0.023255067,0.00070779654,0.000729926,0.0018617534,0.0008154516],"domain_scores_gemma":[0.92209315,0.06868354,0.005695687,0.0010531336,0.0007216941,0.0017526416],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013523159,0.0003771779,0.0017061838,0.00060102885,0.00058087124,0.00078538485,0.0009279913,0.0009194822,0.003246303],"category_scores_gemma":[0.04134613,0.00031114873,0.001312952,0.0004465041,0.000807147,0.00092201406,0.0009049446,0.0013202432,0.0002579068],"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":[0.1505003,0.048048086,0.16022564,0.0031508268,0.0063591907,0.00016859894,0.001668493,0.0030128718,0.0018765705,0.001091804,0.0010253977,0.62287223],"study_design_scores_gemma":[0.010924314,0.29857686,0.6737298,0.00082786626,0.0049611856,0.00024203122,0.0010390094,0.0025455006,0.0023928904,0.00052042754,0.004159799,0.00008029677],"about_ca_topic_score_codex":0.0033128806,"about_ca_topic_score_gemma":0.0071691973,"teacher_disagreement_score":0.013523159,"about_ca_system_score_codex":0.0011967443,"about_ca_system_score_gemma":0.0016661786,"threshold_uncertainty_score":0.07151818},"labels":[],"label_agreement":null},{"id":"W2001415027","doi":"10.1177/0193841x06296947","title":"A Repeated Observation Approach for Estimating the Street Homeless Population","year":2007,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Census and Population Estimation","field":"Mathematics","cited_by":38,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Downtown; Population; Identification (biology); Estimation; Sampling (signal processing); Statistics; Visibility; Geography; Confidence interval; Econometrics; Demography; Psychology; Computer science; Mathematics; Sociology; Engineering","score_opus":0.26078212160035724,"score_gpt":0.44689302585433927,"score_spread":0.18611090425398202,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2001415027","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":[0.10017275,0.0016151067,0.8942221,0.00025340402,0.00009823561,0.00076070894,0.00082524307,0.00031800105,0.0017344271],"genre_scores_gemma":[0.5204864,0.0013930829,0.47283334,0.00012693378,0.00008313509,0.0016848292,0.00089500536,0.000033296703,0.0024639175],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.992141,0.005667248,0.00030521466,0.0011566555,0.0006361192,0.000093775576],"domain_scores_gemma":[0.99183273,0.0045284275,0.0013750193,0.001502414,0.0006972317,0.000064184016],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0072630076,0.00059605093,0.001000695,0.0015513699,0.0003697965,0.0005034745,0.0020825386,0.0007362335,0.0012264514],"category_scores_gemma":[0.022841893,0.00039288448,0.0011774763,0.0013176084,0.00037044397,0.00064423366,0.0005636377,0.000716146,0.00034783423],"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":[0.0004000202,0.00053838047,0.22860776,0.0013033325,0.0045576887,0.00064830505,0.0021537766,0.09088725,0.003625563,0.019067407,0.0039753662,0.6442352],"study_design_scores_gemma":[0.0002769954,0.002847435,0.21507947,0.00035338994,0.0022997211,0.0013677558,0.0015121639,0.71925336,0.0045266063,0.029232224,0.022967169,0.00028378],"about_ca_topic_score_codex":0.022316644,"about_ca_topic_score_gemma":0.021762893,"teacher_disagreement_score":0.022316644,"about_ca_system_score_codex":0.0007506805,"about_ca_system_score_gemma":0.0012248228,"threshold_uncertainty_score":0.044373453},"labels":[],"label_agreement":null},{"id":"W2015434985","doi":"10.1177/0193841x14567313","title":"An Alternative Framework to Investigating and Understanding Intraindividual Processes in Substance Abuse Recovery","year":2015,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Nomothetic and idiographic; Substance abuse; Craving; Psychology; Abstinence; Affect (linguistics); Clinical psychology; Addiction; Developmental psychology; Social psychology; Psychiatry","score_opus":0.28539059405864137,"score_gpt":0.4279131534136509,"score_spread":0.14252255935500951,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2015434985","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":[0.15236875,0.0047040465,0.7175252,0.028025277,0.0005026189,0.001052615,0.0008929534,0.0002582914,0.094670214],"genre_scores_gemma":[0.83692265,0.0013831279,0.15534042,0.0020351224,0.00018533121,0.0015324809,0.00039790946,0.00004652035,0.0021564001],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9915291,0.0060686227,0.00035133664,0.0012110576,0.0004872245,0.00035273278],"domain_scores_gemma":[0.9866214,0.008456392,0.0017332833,0.001699624,0.0010044992,0.00048475416],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010324624,0.0010825811,0.0009374187,0.004609152,0.0022934156,0.0050909687,0.0025798278,0.0016751185,0.0050156238],"category_scores_gemma":[0.017868754,0.00039777372,0.0018206822,0.0034323984,0.011639738,0.005779178,0.0046245637,0.0028495912,0.00052222284],"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":[0.000117660566,0.000322014,0.036405925,0.00042871147,0.00015121582,0.00031188622,0.032943837,0.0031729387,0.0007400208,0.8864699,0.0011270316,0.03780882],"study_design_scores_gemma":[0.00006811405,0.00042534072,0.041459084,0.00057504297,0.00018625014,0.00093714136,0.024038732,0.027155403,0.00041459396,0.8634806,0.04116626,0.00009351483],"about_ca_topic_score_codex":0.008476177,"about_ca_topic_score_gemma":0.0058259415,"teacher_disagreement_score":0.010324624,"about_ca_system_score_codex":0.0043218597,"about_ca_system_score_gemma":0.005870351,"threshold_uncertainty_score":0.054602504},"labels":[],"label_agreement":null},{"id":"W2025194027","doi":"10.1177/0193841x12458103","title":"Measuring Stakeholder Participation in Evaluation","year":2012,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Trois-Rivières; Université Laval","funders":"","keywords":"Intraclass correlation; Reliability (semiconductor); Stakeholder; Sample (material); Convergent validity; Psychology; Generalizability theory; Sample size determination; Scale (ratio); Statistics; Applied psychology; Psychometrics; Clinical psychology; Mathematics; Economics; Developmental psychology; Geography","score_opus":0.8475988642781547,"score_gpt":0.6249450003263772,"score_spread":0.22265386395177744,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2025194027","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":[0.8186346,0.0017141928,0.094240256,0.00656454,0.00018967736,0.0030631488,0.0002299195,0.00019229777,0.07517123],"genre_scores_gemma":[0.98274094,0.0002644606,0.014632012,0.00031674677,0.000030727024,0.0013288629,0.00007236111,0.000022028811,0.000591878],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.66450083,0.27621567,0.011482747,0.00444963,0.03858576,0.004765378],"domain_scores_gemma":[0.5297639,0.32791403,0.045685317,0.015456135,0.072087474,0.009093072],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19961834,0.0007223355,0.0006918487,0.0053752773,0.0033475792,0.005383664,0.001529637,0.0012065017,0.0027567965],"category_scores_gemma":[0.29190803,0.00036734852,0.00090692216,0.0032359445,0.0067654145,0.0054532737,0.0114008915,0.0018877023,0.0004754014],"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":[0.00038478308,0.0010593561,0.3337183,0.0032121178,0.00038550998,0.00036547697,0.18758343,0.0034275465,0.002959521,0.028522534,0.0049496214,0.4334317],"study_design_scores_gemma":[0.00018720227,0.0041056597,0.48288947,0.0078077116,0.0004132517,0.0014804265,0.27540156,0.014225676,0.012990241,0.088872366,0.11114161,0.0004848323],"about_ca_topic_score_codex":0.0014136402,"about_ca_topic_score_gemma":0.0014997305,"teacher_disagreement_score":0.80038166,"about_ca_system_score_codex":0.006642837,"about_ca_system_score_gemma":0.009788483,"threshold_uncertainty_score":0.9870131},"labels":[],"label_agreement":null},{"id":"W2034770421","doi":"10.1177/0193841x05279182","title":"Facilitating Self-Determination in Adolescents with Intellectual Disabilities","year":2005,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Disability Education and Employment","field":"Social Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université du Québec à Trois-Rivières","funders":"","keywords":"Curriculum; Identity (music); Psychology; Self-determination; Intellectual disability; Field (mathematics); Pedagogy; Medical education; Applied psychology; Mathematics education; Political science; Medicine","score_opus":0.09025164283672198,"score_gpt":0.42462906028296526,"score_spread":0.3343774174462433,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2034770421","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":[0.9767994,0.012942067,0.00039596946,0.00074489316,0.000036209087,0.00022740585,0.000028796198,0.000010283535,0.008814927],"genre_scores_gemma":[0.98728395,0.009321656,0.001850955,0.00012662234,0.000013591376,0.00015131473,0.000040787218,0.0000024651547,0.00120886],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99771166,0.001597316,0.0000898345,0.00004360196,0.00035958842,0.00019812175],"domain_scores_gemma":[0.9983222,0.00082999107,0.00026506503,0.000053198968,0.00031460953,0.0002149401],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002991974,0.000103630045,0.00027115797,0.00052790996,0.00043124554,0.0005066229,0.00031228308,0.00022538111,0.0006492733],"category_scores_gemma":[0.005803831,0.0000526695,0.00016431544,0.00032417738,0.00035449615,0.00030273636,0.00080812594,0.0002564344,0.00008063531],"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":[0.00014189114,0.0017655793,0.054310836,0.002725678,0.00005352105,0.0004914181,0.021242278,0.00039469762,0.0015057096,0.0019375188,0.0027832675,0.9126476],"study_design_scores_gemma":[0.00035416515,0.0067085945,0.69406325,0.008703141,0.0002958334,0.0031397874,0.0737372,0.000635482,0.0082600815,0.0022609746,0.20177698,0.0000644939],"about_ca_topic_score_codex":0.0057282858,"about_ca_topic_score_gemma":0.018856693,"teacher_disagreement_score":0.0057282858,"about_ca_system_score_codex":0.00085416617,"about_ca_system_score_gemma":0.004828844,"threshold_uncertainty_score":0.015823245},"labels":[],"label_agreement":null},{"id":"W2047617457","doi":"10.1177/0193841x02026004003","title":"The Development of Science Achievement in Middle and High Schoolr","year":2002,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Educational Environments and Student Outcomes","field":"Social Sciences","cited_by":68,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; Canadian Institute for Advanced Research","funders":"","keywords":"Socioeconomic status; Multilevel model; Academic achievement; Context (archaeology); Psychology; Mathematics education; Autonomy; Science education; Demography; Sociology; Geography; Political science; Mathematics; Statistics","score_opus":0.14697315987348716,"score_gpt":0.41420795312102526,"score_spread":0.26723479324753807,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2047617457","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":[0.9882933,0.0027386781,0.00034946596,0.00020472342,0.000022168446,0.000055650315,0.0015316833,0.000051305327,0.0067530554],"genre_scores_gemma":[0.99644756,0.0007915176,0.00051834364,0.000031494263,0.000007003368,0.0000605863,0.0009685584,0.000004588451,0.001170243],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9985481,0.0004370284,0.00013947993,0.00016468077,0.00056054525,0.00015011623],"domain_scores_gemma":[0.99553645,0.0008829561,0.0017318067,0.00021302226,0.0011851585,0.00045066414],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034712283,0.00026525356,0.0004916912,0.0022366098,0.00039357747,0.0014897791,0.0005454435,0.000506396,0.001262167],"category_scores_gemma":[0.0077523817,0.00018666791,0.00073459145,0.0017756112,0.00034194888,0.00058142043,0.001027242,0.0006842043,0.00044193593],"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":[0.00016541516,0.00027014708,0.9657227,0.00013595253,0.00017393136,0.000049184953,0.00068016985,0.00017904841,0.0002777603,0.00029999725,0.0005977207,0.031448092],"study_design_scores_gemma":[0.000002974995,0.00014748164,0.99843735,0.000011331607,0.000021919432,0.000026010415,0.0001158994,0.00010388759,0.00014057412,0.000054578308,0.0009350994,0.000002899554],"about_ca_topic_score_codex":0.0147376815,"about_ca_topic_score_gemma":0.0220792,"teacher_disagreement_score":0.0147376815,"about_ca_system_score_codex":0.0010210936,"about_ca_system_score_gemma":0.0012555969,"threshold_uncertainty_score":0.029303849},"labels":[],"label_agreement":null},{"id":"W2065805300","doi":"10.1177/0193841x06293411","title":"Does Receiving an Earnings Supplement Affect Union Formation? Estimating Effects for Program Participants Using Propensity Score Matching","year":2006,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Gender, Labor, and Family Dynamics","field":"Social Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Propensity score matching; Receipt; Earnings; Matching (statistics); Disadvantaged; Affect (linguistics); Context (archaeology); Demographic economics; Random assignment; Average treatment effect; Psychology; Economics; Econometrics; Medicine; Accounting; Economic growth; Geography","score_opus":0.13375186684758997,"score_gpt":0.42968491301862294,"score_spread":0.295933046171033,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065805300","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":[0.94648534,0.003430493,0.044234544,0.0010252863,0.000085400396,0.0014411415,0.0006842465,0.0000678288,0.0025457],"genre_scores_gemma":[0.9821326,0.00096465985,0.015284513,0.00018686685,0.000027268281,0.00064216956,0.00021661041,0.0000060589387,0.0005393009],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.96752155,0.028017087,0.0010208148,0.0012582681,0.001791087,0.00039114387],"domain_scores_gemma":[0.96418333,0.028455831,0.0047274022,0.0019884224,0.00049463724,0.00015044674],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03367358,0.00039015617,0.0010788476,0.0010042355,0.00047591826,0.0007021778,0.0008597038,0.0008422233,0.0016203091],"category_scores_gemma":[0.05746154,0.00024839488,0.0019260058,0.0014954382,0.00086422684,0.0005302378,0.00078263663,0.00059798564,0.00015727272],"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":[0.0057974993,0.002871907,0.6071601,0.0017706559,0.012205306,0.00013527567,0.0019075525,0.009074802,0.001711087,0.008011852,0.0010641665,0.34828982],"study_design_scores_gemma":[0.0021861678,0.014808717,0.9059773,0.00064770645,0.019941056,0.00015416482,0.001956835,0.026011867,0.0069158264,0.011917292,0.009380723,0.00010227466],"about_ca_topic_score_codex":0.015354732,"about_ca_topic_score_gemma":0.010519582,"teacher_disagreement_score":0.03367358,"about_ca_system_score_codex":0.00092485204,"about_ca_system_score_gemma":0.0021229822,"threshold_uncertainty_score":0.17808509},"labels":[],"label_agreement":null},{"id":"W2093553995","doi":"10.1177/0193841x03261484","title":"Picking Up the Challenge","year":2004,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Homelessness and Social Issues","field":"Health Professions","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; Child Welfare League of Canada; Royal Ottawa Mental Health Centre","funders":"","keywords":"Enumeration; Population; Computer science; Data science; Management science; Sociology; Mathematics; Engineering; Demography","score_opus":0.2812821955724736,"score_gpt":0.5541621417145391,"score_spread":0.27287994614206557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2093553995","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0045911497,0.031074354,0.009259,0.9319632,0.009405094,0.00010839619,0.0006752119,0.00013648803,0.012787087],"genre_scores_gemma":[0.18183044,0.070366785,0.051263735,0.6321428,0.016027061,0.0008031011,0.0013129385,0.00065958337,0.04559358],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9611962,0.016785778,0.00211125,0.0046349578,0.011743404,0.003528468],"domain_scores_gemma":[0.9151109,0.024566196,0.0032157218,0.0046422128,0.04277273,0.009692255],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06007451,0.0013454951,0.0027087398,0.00531082,0.010027439,0.014286185,0.0053208796,0.008223857,0.01548268],"category_scores_gemma":[0.11800695,0.00058560044,0.0013523265,0.004802318,0.017187305,0.023944754,0.012297369,0.018272761,0.0077508707],"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":[0.00022022941,0.000111036104,0.009103739,0.0033113048,0.0002600941,0.00042819764,0.012319979,0.00077835144,0.0006080727,0.15265276,0.5229038,0.29730245],"study_design_scores_gemma":[0.000040061368,0.0001239557,0.0071736407,0.0063569965,0.00010085287,0.00043976464,0.053607736,0.00074543385,0.00039376516,0.16827945,0.7625695,0.00016893173],"about_ca_topic_score_codex":0.3765837,"about_ca_topic_score_gemma":0.54344386,"teacher_disagreement_score":0.3765837,"about_ca_system_score_codex":0.022628292,"about_ca_system_score_gemma":0.064489335,"threshold_uncertainty_score":0.74878347},"labels":[],"label_agreement":null},{"id":"W2099407725","doi":"10.1177/0193841x0102500602","title":"Characteristics of Responding-, Nonresponding- and Refusing-Parents in an Adolescent Lifestyle Choice Study","year":2001,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Adolescent Sexual and Reproductive Health","field":"Health Professions","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Brock University; Queen's University","funders":"","keywords":"Parental consent; Informed consent; Psychology; Scale (ratio); Clinical psychology; Developmental psychology; Medicine; Alternative medicine","score_opus":0.2809426709750991,"score_gpt":0.5545025882861809,"score_spread":0.2735599173110818,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099407725","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":[0.9988041,0.0002999115,0.00018285429,0.00007879029,0.000009265001,0.00015107401,0.000086619584,0.0000028310549,0.0003845772],"genre_scores_gemma":[0.99816567,0.00042586826,0.00037432008,0.00015516978,0.000008621937,0.000298514,0.00019252404,0.0000041320836,0.00037517835],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99128294,0.0053802957,0.0014145507,0.0003763368,0.001178318,0.00036764768],"domain_scores_gemma":[0.97935176,0.009248453,0.006725852,0.0013417738,0.0021020551,0.0012301969],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008668701,0.00014788359,0.000463744,0.0008651736,0.00070762885,0.0007080413,0.0002819881,0.00048551572,0.0011833519],"category_scores_gemma":[0.030504534,0.00023531301,0.00031985488,0.00089172303,0.00034872774,0.0004633262,0.00035909042,0.000515703,0.0003675668],"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":[0.00030880212,0.0003184466,0.98148876,0.0000493266,0.000027555116,0.0008047542,0.006649291,0.000017721677,0.00051065336,0.00005234306,0.0002132431,0.009559067],"study_design_scores_gemma":[0.000035996214,0.0016133946,0.97821563,0.00008689897,0.000041178493,0.003281424,0.013859704,0.0001606287,0.00074124936,0.00006056779,0.0018811745,0.000022172586],"about_ca_topic_score_codex":0.0014203355,"about_ca_topic_score_gemma":0.001898246,"teacher_disagreement_score":0.008668701,"about_ca_system_score_codex":0.0002914195,"about_ca_system_score_gemma":0.0006451227,"threshold_uncertainty_score":0.04584503},"labels":[],"label_agreement":null},{"id":"W2112067598","doi":"10.1177/0193841x0002400601","title":"The Evaluation of Information Campaigns to Promote Voluntary Household Water Conservation","year":2000,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Environmental Education and Sustainability","field":"Environmental Science","cited_by":246,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Persuasion; Popularity; Summative assessment; Water conservation; Public relations; Marketing; Business; Environmental resource management; Psychology; Environmental economics; Political science; Social psychology; Formative assessment; Economics; Ecology; Water resources; Pedagogy","score_opus":0.03931785094118342,"score_gpt":0.31836773981942185,"score_spread":0.27904988887823845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2112067598","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":[0.8192203,0.09504247,0.007958623,0.0040109013,0.00069458777,0.013856355,0.0005098584,0.00012251572,0.058584508],"genre_scores_gemma":[0.9458874,0.033603895,0.01142157,0.00090183405,0.00022730809,0.0047571724,0.00031232278,0.000019267329,0.0028692377],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.8981279,0.08146503,0.0035175013,0.0010104275,0.014667078,0.0012120458],"domain_scores_gemma":[0.83706397,0.13491492,0.009379298,0.0021664924,0.015242421,0.0012329966],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0606851,0.0010104805,0.0011023565,0.002954985,0.00082571845,0.0020963678,0.0009149275,0.0013013658,0.0022764748],"category_scores_gemma":[0.12187141,0.00028756933,0.0007097275,0.0022981,0.00088712,0.0013941089,0.0010419129,0.0007808465,0.00034358862],"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":[0.003488295,0.007714644,0.0069134436,0.027403295,0.0007295411,0.00012237077,0.005213833,0.0011988281,0.0017354904,0.0026345951,0.002851688,0.939994],"study_design_scores_gemma":[0.009759265,0.291812,0.2957586,0.09603596,0.013058293,0.0012906092,0.034958255,0.008797301,0.07488319,0.01150014,0.1615313,0.0006150429],"about_ca_topic_score_codex":0.0034314787,"about_ca_topic_score_gemma":0.0053725145,"teacher_disagreement_score":0.0606851,"about_ca_system_score_codex":0.004698758,"about_ca_system_score_gemma":0.006607172,"threshold_uncertainty_score":0.32093734},"labels":[],"label_agreement":null},{"id":"W2112624513","doi":"10.1177/0193841x06293412","title":"Addressing Program Fidelity Using Onsite Observations and Program Provider Descriptions of Program Delivery","year":2006,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Youth Development and Social Support","field":"Social Sciences","cited_by":82,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Fidelity; Program evaluation; Data collection; Quarter (Canadian coin); Computer science; Program Design Language; Population; Medical education; Medicine; Environmental health; Political science; Software engineering","score_opus":0.41773105899385166,"score_gpt":0.49081063493772575,"score_spread":0.0730795759438741,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2112624513","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":[0.75168383,0.005519317,0.17476155,0.0015684665,0.00024480812,0.040491875,0.0027106097,0.00039915967,0.022620428],"genre_scores_gemma":[0.7830386,0.0050168755,0.16282973,0.0006437697,0.00010544745,0.043035045,0.0017677348,0.00012261633,0.0034401144],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.81549865,0.14016555,0.022276334,0.0042958143,0.016111875,0.001651831],"domain_scores_gemma":[0.7360022,0.14399984,0.04422114,0.027073512,0.047899056,0.00080423435],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.14334062,0.0012745927,0.0014174101,0.0040501193,0.001925411,0.0025096629,0.0022549457,0.0009712207,0.0009970024],"category_scores_gemma":[0.2444236,0.001157569,0.001399509,0.0042191427,0.0014940888,0.0033683283,0.0031946355,0.0015978257,0.00029681556],"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":[0.001117984,0.0025471626,0.2957157,0.011166734,0.0011260499,0.00038965043,0.091626525,0.003074201,0.0074138353,0.0022979656,0.003369136,0.58015496],"study_design_scores_gemma":[0.00095111073,0.013857259,0.76595116,0.014163829,0.0022280319,0.0010064314,0.08089122,0.012381461,0.051654078,0.0047205887,0.051565275,0.0006295529],"about_ca_topic_score_codex":0.011302095,"about_ca_topic_score_gemma":0.031001203,"teacher_disagreement_score":0.14334062,"about_ca_system_score_codex":0.0039905002,"about_ca_system_score_gemma":0.0069925324,"threshold_uncertainty_score":0.7580667},"labels":[],"label_agreement":null},{"id":"W2113827930","doi":"10.1177/0193841x0002400501","title":"Health Outcomes of Elementary School Students in New Brunswick","year":2000,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Obesity, Physical Activity, Diet","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Alberta; Canadian Institute for Advanced Research","funders":"","keywords":"Socioeconomic status; Psychology; Mental health; Physical health; Multilevel model; Healthy eating; Physical activity; Gerontology; Environmental health; Medicine; Physical therapy; Psychiatry; Population","score_opus":0.05501168353014899,"score_gpt":0.43534261547478337,"score_spread":0.38033093194463435,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113827930","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":[0.993006,0.0037899304,0.000034104054,0.00020513899,0.000011978684,0.000058724785,0.00093250064,0.000003926339,0.0019577132],"genre_scores_gemma":[0.99463576,0.0034581027,0.00012844539,0.000103387574,0.0000057399507,0.00006616123,0.0007573059,0.0000017079768,0.0008435416],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.999696,0.00004607123,0.000031932395,0.000045274937,0.00010719414,0.00007355347],"domain_scores_gemma":[0.9994622,0.000063687025,0.00018553903,0.000016766331,0.000152937,0.00011891507],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005174384,0.0002070967,0.00048096772,0.00113329,0.00054981816,0.0006068824,0.000409327,0.00022725172,0.0008931179],"category_scores_gemma":[0.00088977005,0.00017274574,0.00024050959,0.0019847315,0.00031844978,0.00019217802,0.00042375104,0.00038503757,0.00015326981],"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":[0.0001298031,0.00017296545,0.97227514,0.00016570686,0.00008108182,0.00013318198,0.0009889782,0.00007636315,0.0006110908,0.00004352348,0.0005640127,0.024758084],"study_design_scores_gemma":[0.0000058836586,0.000046571837,0.99886143,0.000025672285,0.000015993444,0.000019054258,0.00048507727,0.000014899455,0.00006623107,0.000005760177,0.0004503032,0.0000030675094],"about_ca_topic_score_codex":0.5392546,"about_ca_topic_score_gemma":0.82152474,"teacher_disagreement_score":0.4607454,"about_ca_system_score_codex":0.0038435955,"about_ca_system_score_gemma":0.005042858,"threshold_uncertainty_score":0.9269176},"labels":[],"label_agreement":null},{"id":"W2124187889","doi":"10.1177/0193841x05284090","title":"How to Begin a New Topic in Mathematics","year":2006,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Mathematics Education and Teaching Techniques","field":"Social Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Mathematics education; Computer science; Contrast (vision); Multilevel model; Mathematics","score_opus":0.12386235465861056,"score_gpt":0.4557607482728444,"score_spread":0.33189839361423384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124187889","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":[0.79910296,0.027064761,0.031575907,0.020075347,0.00077680347,0.0010438106,0.00047060256,0.00060406054,0.11928578],"genre_scores_gemma":[0.80935603,0.03070368,0.13189533,0.0035671808,0.0001937081,0.00044193017,0.0006029367,0.0001518951,0.023087226],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99668354,0.0008561483,0.00018272676,0.00035016434,0.0016682401,0.0002591566],"domain_scores_gemma":[0.983074,0.008436046,0.0025422885,0.0011679202,0.0030960902,0.0016836617],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034981787,0.00030759876,0.00067293295,0.0015632347,0.001395129,0.0029682836,0.0010454078,0.00091149315,0.0037499303],"category_scores_gemma":[0.029622387,0.0002807292,0.00063024886,0.0009921986,0.0012685456,0.0030780449,0.0015102905,0.0018021162,0.0011673493],"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":[0.00020413045,0.0013159213,0.069656946,0.002271532,0.00013119608,0.0005551609,0.055716824,0.0008061314,0.010779717,0.014627937,0.016531125,0.8274033],"study_design_scores_gemma":[0.00023456854,0.0017108986,0.2639416,0.0038253209,0.00054167287,0.0016492716,0.05145964,0.0012561466,0.014144236,0.019813951,0.6412277,0.00019494227],"about_ca_topic_score_codex":0.033055704,"about_ca_topic_score_gemma":0.12123934,"teacher_disagreement_score":0.033055704,"about_ca_system_score_codex":0.003034403,"about_ca_system_score_gemma":0.009809674,"threshold_uncertainty_score":0.06572658},"labels":[],"label_agreement":null},{"id":"W2131308570","doi":"10.1177/0193841x03254349","title":"Outpatient Marijuana Treatment for Adolescents","year":2003,"lang":"en","type":"review","venue":"Evaluation Review","topic":"Prenatal Substance Exposure Effects","field":"Medicine","cited_by":60,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"National Institute on Drug Abuse; U.S. Public Health Service","keywords":"Economic evaluation; Economic cost; Economic analysis; Substance abuse; Intervention (counseling); Drug treatment; Quarter (Canadian coin); Medicine; Treatment and control groups; Cost–benefit analysis; Substance abuse treatment; Cost analysis; Total cost; Environmental health; Demography; Psychiatry; Economics; Internal medicine; Agricultural economics; Geography; Political science; Operations research","score_opus":0.16997458196408038,"score_gpt":0.45066889266219834,"score_spread":0.28069431069811795,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2131308570","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008397925,0.98950106,0.00006883674,0.00023893284,0.00005515559,0.00012386266,0.00008302244,0.000004820163,0.0015264886],"genre_scores_gemma":[0.035310533,0.963481,0.00041989223,0.00015460192,0.000060300903,0.00014230715,0.00007166246,0.0000013808742,0.0003582817],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9991153,0.0005079909,0.00009525825,0.000041382376,0.0002013552,0.000038606257],"domain_scores_gemma":[0.99934965,0.00037605816,0.00017638195,0.000012129857,0.000054147207,0.000031564396],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012825168,0.0004014753,0.0012415556,0.0014320858,0.00020301715,0.00053172756,0.0005662919,0.00055308326,0.0029931248],"category_scores_gemma":[0.0027647323,0.00018523281,0.0011007169,0.0017968256,0.00014584964,0.00028243716,0.00024538563,0.00043387298,0.00021998023],"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":[0.0012662181,0.000417428,0.0015352841,0.06445342,0.0010855463,0.00030147555,0.00006747079,0.00021750282,0.00039945837,0.00041213303,0.0048043584,0.9250397],"study_design_scores_gemma":[0.018434927,0.025597882,0.15769185,0.25024757,0.032612164,0.016055554,0.0015748368,0.0023236275,0.0057774913,0.0031203202,0.48643896,0.00012472634],"about_ca_topic_score_codex":0.003365192,"about_ca_topic_score_gemma":0.010176654,"teacher_disagreement_score":0.003365192,"about_ca_system_score_codex":0.0009932616,"about_ca_system_score_gemma":0.0016405447,"threshold_uncertainty_score":0.010013044},"labels":[],"label_agreement":null},{"id":"W2166236442","doi":"10.1177/0193841x14531260","title":"Grief, Anger, and Relationality","year":2014,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Emotional Labor in Professions","field":"Social Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Public Health Ontario; Toronto Rehabilitation Institute; University of Toronto; University Health Network","funders":"Institute of Gender and Health; Canadian Institutes of Health Research; Toronto Rehabilitation Institute; Ontario Ministry of Health and Long-Term Care","keywords":"Grief; Anger; Empathy; Psychology; Context (archaeology); Neurorehabilitation; Nursing; Thematic analysis; Focus group; Psychosocial; Rehabilitation; Health care; Feeling; Psychotherapist; Medicine; Qualitative research; Clinical psychology; Psychiatry; Physical therapy; Social psychology","score_opus":0.08182174885329145,"score_gpt":0.47375205392300074,"score_spread":0.3919303050697093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2166236442","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":[0.91428065,0.008835912,0.0033835834,0.0043061273,0.00010650101,0.00007055128,0.00010672143,0.000040040693,0.06886995],"genre_scores_gemma":[0.99727863,0.0010234797,0.00049254467,0.0001779846,0.000018746565,0.000016142465,0.000015818832,0.0000034222762,0.00097320334],"study_design_codex":"observational","study_design_gemma":"qualitative","domain_scores_codex":[0.9987305,0.0007229454,0.000034983783,0.00008348301,0.0002803221,0.00014772332],"domain_scores_gemma":[0.9978986,0.0006796832,0.0008653078,0.00008996073,0.0001389104,0.00032745578],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001525042,0.00018972448,0.00016385828,0.00043734818,0.00098946,0.0015564404,0.00027786798,0.00034212557,0.002912004],"category_scores_gemma":[0.0048382087,0.00008547275,0.00015633108,0.0002404532,0.0037163447,0.00075637805,0.0015611955,0.0009338128,0.00018934929],"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":[0.0009814503,0.00088785775,0.43164757,0.0012632129,0.00036523648,0.0021365767,0.16836886,0.0012771083,0.009122081,0.05523082,0.006950372,0.32176882],"study_design_scores_gemma":[0.00007216324,0.0007752204,0.7996441,0.00089269923,0.00015940506,0.0058063655,0.09898143,0.0010864334,0.0024286709,0.041895013,0.048160166,0.00009835028],"about_ca_topic_score_codex":0.0013727869,"about_ca_topic_score_gemma":0.0017788867,"teacher_disagreement_score":0.002912004,"about_ca_system_score_codex":0.0009440446,"about_ca_system_score_gemma":0.0005660162,"threshold_uncertainty_score":0.009741664},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2168096943","doi":"10.1177/0193841x12439704","title":"Homelessness Outcome Reporting Normative Framework","year":2012,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Homelessness and Social Issues","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Victoria","funders":"","keywords":"Normative; Report card; Outcome (game theory); Psychology; Political science; Economics","score_opus":0.3399599014597608,"score_gpt":0.588645813129939,"score_spread":0.24868591167017823,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2168096943","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029520957,0.013503073,0.6437519,0.031348046,0.0021046833,0.06956446,0.048154093,0.0016640408,0.16038871],"genre_scores_gemma":[0.20419295,0.0065997387,0.51617074,0.0055160504,0.00047427797,0.22098668,0.039625924,0.00023813659,0.0061955103],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7717471,0.15103014,0.03971881,0.00407094,0.030476574,0.0029564567],"domain_scores_gemma":[0.76857066,0.100457676,0.023731863,0.012012941,0.09302799,0.0021988899],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.19979735,0.0013002106,0.0021923,0.013225903,0.0018935647,0.0047063394,0.0056852438,0.002318866,0.0054818424],"category_scores_gemma":[0.22235143,0.00055813405,0.003691676,0.007668984,0.004354449,0.005734625,0.0064705396,0.0041877446,0.0019380732],"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":[0.0006532879,0.00065029465,0.046859883,0.011497857,0.00057105476,0.00019647791,0.0064404076,0.0047446648,0.0004337713,0.512068,0.064086154,0.3517982],"study_design_scores_gemma":[0.00092334603,0.0028766657,0.07989924,0.029776707,0.0015282213,0.0013013806,0.012998903,0.016630951,0.0035432554,0.33752364,0.5124594,0.00053829746],"about_ca_topic_score_codex":0.011627299,"about_ca_topic_score_gemma":0.010536948,"teacher_disagreement_score":0.80020267,"about_ca_system_score_codex":0.0121940635,"about_ca_system_score_gemma":0.02651025,"threshold_uncertainty_score":0.9867923},"labels":[],"label_agreement":null},{"id":"W2302098269","doi":"10.1177/0193841x16637950","title":"How Do Evaluators Differentiate Successful From Less-Than-Successful Experiences With Collaborative Approaches to Evaluation?","year":2016,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University; Queen's University; University of Ottawa","funders":"University of Ottawa","keywords":"Stakeholder; Antecedent (behavioral psychology); Psychology; Control (management); Computer-assisted web interviewing; Applied psychology; Knowledge management; Medical education; Social psychology; Public relations; Computer science; Political science; Business; Marketing; Medicine","score_opus":0.4512052275898277,"score_gpt":0.4632224952179479,"score_spread":0.01201726762812022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2302098269","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":[0.9807966,0.0025598758,0.0042316634,0.003216593,0.00009307466,0.00014268844,0.00004131558,0.000027801845,0.008890506],"genre_scores_gemma":[0.9968502,0.00058644585,0.001597352,0.00039924457,0.00003063066,0.00013093652,0.00003221087,0.0000140850725,0.0003588552],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.8942397,0.08140605,0.0066054724,0.0029222625,0.011678676,0.0031478177],"domain_scores_gemma":[0.6568459,0.23765358,0.053669665,0.01033164,0.03338679,0.008112387],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.088899545,0.000310519,0.000659669,0.0041816323,0.0016823787,0.006900586,0.0009411572,0.0015130565,0.0012370009],"category_scores_gemma":[0.30771148,0.00037906753,0.0005462221,0.0026334783,0.0039435225,0.00729974,0.0046064444,0.0011910695,0.00028415062],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007093258,0.00039172953,0.33458292,0.0018286312,0.00032176328,0.00041024093,0.4114484,0.0002523866,0.0014583709,0.00478912,0.0025109642,0.2412962],"study_design_scores_gemma":[0.00012171861,0.0008798745,0.365412,0.0037457321,0.00021974312,0.00078731054,0.5947846,0.0011010038,0.0023624601,0.008107229,0.022256918,0.00022135024],"about_ca_topic_score_codex":0.0016844101,"about_ca_topic_score_gemma":0.0036164997,"teacher_disagreement_score":0.088899545,"about_ca_system_score_codex":0.0022626885,"about_ca_system_score_gemma":0.0030749005,"threshold_uncertainty_score":0.47015136},"labels":[],"label_agreement":null},{"id":"W2889669442","doi":"10.1177/0193841x18799093","title":"The Sensitivity of Impact Estimates to Data Sources Used: Analysis From an Access to Postsecondary Education Experiment","year":2018,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Canada Millennium Scholarship Foundation","keywords":"Context (archaeology); Psychological intervention; Survey data collection; Randomized experiment; Psychology; Estimation; Political science; Demographic economics; Statistics; Economics; Geography","score_opus":0.3855258045495976,"score_gpt":0.6285228913714911,"score_spread":0.24299708682189347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889669442","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":"methods","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":[0.7231783,0.01514079,0.21505216,0.002893357,0.0006468006,0.027576905,0.00614904,0.00044922496,0.008913378],"genre_scores_gemma":[0.95752984,0.0009099806,0.029091528,0.0012030374,0.000101132966,0.00837252,0.0019232525,0.000101670164,0.00076701463],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.40552235,0.5205953,0.023644956,0.015819116,0.031371534,0.0030468304],"domain_scores_gemma":[0.09525147,0.83441186,0.032185595,0.028636627,0.0088482965,0.00066613697],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.5028804,0.0018089629,0.00461076,0.0020730924,0.0022652014,0.0041350266,0.0037531916,0.0031970022,0.003450271],"category_scores_gemma":[0.6265934,0.001194841,0.010184716,0.0046418086,0.00532116,0.0032983252,0.0041069873,0.0050752237,0.00037296678],"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":[0.12699214,0.009734073,0.35286063,0.021118695,0.16403422,0.0017906543,0.00931769,0.0898468,0.0048028156,0.023755874,0.0060436856,0.18970272],"study_design_scores_gemma":[0.028280007,0.06432137,0.48397365,0.0051848292,0.14180261,0.0015962182,0.006818704,0.13925721,0.03308961,0.051939327,0.041929137,0.0018073848],"about_ca_topic_score_codex":0.04169134,"about_ca_topic_score_gemma":0.018934136,"teacher_disagreement_score":0.5028804,"about_ca_system_score_codex":0.008787745,"about_ca_system_score_gemma":0.008683087,"threshold_uncertainty_score":0.613037},"labels":[],"label_agreement":null},{"id":"W4226126191","doi":"10.1177/0193841x221085355","title":"Analyzing the Nexus Between Geopolitical Risk, Policy Uncertainty, and Tourist Arrivals: Evidence From the United States","year":2022,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Market Dynamics and Volatility","field":"Economics, Econometrics and Finance","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Nexus (standard); Geopolitics; Tourism; Political science; Economics; Regional science; Geography; Computer science","score_opus":0.10686984929016416,"score_gpt":0.3450513100099599,"score_spread":0.23818146071979576,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4226126191","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":[0.8907594,0.08126471,0.0023301418,0.0058809374,0.00026366714,0.00006445907,0.00463917,0.000018302999,0.014779213],"genre_scores_gemma":[0.96028125,0.03642889,0.00047406656,0.00047597854,0.00009433362,0.000016924158,0.0017094471,0.0000052443493,0.00051381934],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9996371,0.00013870711,0.00003999007,0.000053315285,0.00008592248,0.000044830427],"domain_scores_gemma":[0.994942,0.0027253414,0.0014364689,0.0001411983,0.000627992,0.00012697981],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011851139,0.00025995905,0.00030682026,0.001661307,0.0003178589,0.0011808409,0.00026438475,0.0003102172,0.0016238001],"category_scores_gemma":[0.0049408795,0.00014326027,0.0006564855,0.0036118408,0.00039298864,0.0007901012,0.00071681454,0.0008777856,0.00013536387],"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":[0.0002031663,0.00019097378,0.89429647,0.0023247183,0.0017979868,0.00048900174,0.0009675635,0.0038585807,0.00023833373,0.008865528,0.0063923113,0.080375336],"study_design_scores_gemma":[0.000021377542,0.0001878555,0.9563416,0.001690183,0.0013690032,0.00016812686,0.0067769443,0.0045854687,0.00043813718,0.0040477947,0.02431881,0.000054682994],"about_ca_topic_score_codex":0.04237693,"about_ca_topic_score_gemma":0.060920354,"teacher_disagreement_score":0.04237693,"about_ca_system_score_codex":0.0005356294,"about_ca_system_score_gemma":0.0013370425,"threshold_uncertainty_score":0.08426052},"labels":[],"label_agreement":null},{"id":"W4281261681","doi":"10.1177/0193841x221098211","title":"Post-Graduation Work Visas and Loopholes: Insights into Support Provision for International Graduates from the Perspectives of Migration Agents, Universities, and International Graduates","year":2022,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Impulse Buying and Technology Impacts","field":"Economics, Econometrics and Finance","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Impact","funders":"","keywords":"Graduation (instrument); Nexus (standard); Work (physics); Scope (computer science); International education; Public relations; Political science; Career Pathways; Higher education; Medical education; Medicine; Engineering; Law","score_opus":0.05548297578315207,"score_gpt":0.3011743031318248,"score_spread":0.24569132734867274,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4281261681","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":[0.98001546,0.0004593723,0.0005537934,0.005602161,0.000057909634,0.0000352362,0.00001563655,0.000008804858,0.013251516],"genre_scores_gemma":[0.997156,0.00035234002,0.00021347422,0.000476082,0.00001637656,0.00001793646,0.000008461408,0.0000071851723,0.0017520903],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.99532616,0.002706596,0.00011735906,0.00018883681,0.00041453095,0.0012465024],"domain_scores_gemma":[0.9913579,0.004779561,0.0009526782,0.00023778487,0.00054740434,0.0021246618],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0078645535,0.00037916546,0.00038200757,0.001394745,0.010642449,0.009211637,0.0015034004,0.0023236587,0.0039267084],"category_scores_gemma":[0.010863036,0.0003630631,0.00030233254,0.0010756416,0.010057234,0.004722617,0.0070804777,0.004776845,0.00033503983],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020532423,0.00009151776,0.009477773,0.00005043618,0.000003563192,0.00066182116,0.97463846,0.000029737206,0.00025908067,0.0057932236,0.000679212,0.008294643],"study_design_scores_gemma":[0.0000013849434,0.000026711896,0.0033014156,0.000056223267,0.0000016260675,0.000051471492,0.992195,0.000031555817,0.000033041495,0.00036711647,0.0039311494,0.0000033016274],"about_ca_topic_score_codex":0.009084501,"about_ca_topic_score_gemma":0.021620879,"teacher_disagreement_score":0.010642449,"about_ca_system_score_codex":0.005528405,"about_ca_system_score_gemma":0.010299776,"threshold_uncertainty_score":0.04159224},"labels":[],"label_agreement":null},{"id":"W4315752656","doi":"10.1177/0193841x221149809","title":"The Dynamic Nexus Between International Tourism and Environmental Degradation in Top Twenty Tourist Destinations: New Insights From Quantile-on-Quantile Approach","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Energy, Environment, Economic Growth","field":"Economics, Econometrics and Finance","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Quantile; Tourism; Nexus (standard); Quantile regression; Granger causality; Destinations; Econometrics; China; Geography; Economic geography; Economics; Computer science","score_opus":0.05660189426987145,"score_gpt":0.2808662194890804,"score_spread":0.22426432521920897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315752656","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":[0.88962775,0.074177645,0.015631158,0.0026356773,0.00009443773,0.00006151748,0.004884243,0.000057697693,0.012829803],"genre_scores_gemma":[0.98629004,0.010810516,0.0011217757,0.00008816561,0.000042850013,0.00001603263,0.0009932934,0.000008272677,0.00062919495],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99963,0.00015905601,0.000025671,0.0000648938,0.00006127339,0.00005920323],"domain_scores_gemma":[0.9986871,0.0006785399,0.0002697566,0.00006139157,0.00024177389,0.00006138829],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001301822,0.00021079481,0.00032608898,0.0014713759,0.0001305672,0.00081066036,0.0002757401,0.00021494881,0.0017492024],"category_scores_gemma":[0.0025677911,0.00010486377,0.00076684175,0.002885273,0.00025501396,0.0005886646,0.00047775684,0.00057388237,0.00014931695],"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":[0.0001626888,0.00010581355,0.8406815,0.0012927126,0.0012441652,0.0004275725,0.0010005112,0.018403076,0.00035686747,0.011588896,0.004026896,0.12070932],"study_design_scores_gemma":[0.000006507339,0.000098421966,0.9634287,0.00034618622,0.0004503661,0.00012089103,0.002125123,0.01631997,0.00019114144,0.005370491,0.0115158055,0.000026348516],"about_ca_topic_score_codex":0.02206331,"about_ca_topic_score_gemma":0.018343229,"teacher_disagreement_score":0.02206331,"about_ca_system_score_codex":0.00055516366,"about_ca_system_score_gemma":0.00063503237,"threshold_uncertainty_score":0.043869793},"labels":[],"label_agreement":null},{"id":"W4366256521","doi":"10.1177/0193841x231166973","title":"Examining the Effects of Renewable Energy and Economic Growth on Carbon Emission in Canada: Evidence from the Nonlinear ARDL Approaches","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Energy, Environment, Economic Growth","field":"Economics, Econometrics and Finance","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Renewable energy; Per capita; Economics; Natural resource economics; Greenhouse gas; Energy consumption; Fossil fuel; Shock (circulatory); Energy intensity; Environmental science; Consumption (sociology); Unit root; Econometrics; Ecology; Engineering; Waste management","score_opus":0.10403570464279333,"score_gpt":0.24364218887594705,"score_spread":0.13960648423315372,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366256521","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":[0.97349894,0.002696304,0.0026368587,0.0025723872,0.00005815089,0.00006709749,0.0039579635,0.0000880935,0.014424371],"genre_scores_gemma":[0.99064493,0.0014240647,0.0008248609,0.0001732695,0.000016581815,0.000018217052,0.0023030436,0.000027628943,0.0045674276],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99761814,0.0003594157,0.00012898419,0.00041602412,0.00076459063,0.00071279105],"domain_scores_gemma":[0.99099976,0.0032192874,0.0011219959,0.0003908697,0.0036462923,0.00062183087],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003302377,0.0008285792,0.0008936291,0.0015693479,0.0019192758,0.0021131206,0.0018253034,0.00067146367,0.0038922485],"category_scores_gemma":[0.010756076,0.00039707567,0.0021345324,0.0030759166,0.0013720727,0.0009618661,0.002147979,0.0021682903,0.00049833086],"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":[0.0005392337,0.0003092101,0.89153385,0.0003453546,0.0010830497,0.0012437031,0.0013677916,0.05041207,0.0006314473,0.016854836,0.006006526,0.029672833],"study_design_scores_gemma":[0.00013778699,0.00023597486,0.768404,0.00027481356,0.0019989442,0.00023482142,0.0069809845,0.19462255,0.0022490697,0.0060163373,0.01862772,0.00021703415],"about_ca_topic_score_codex":0.9878628,"about_ca_topic_score_gemma":0.9781285,"teacher_disagreement_score":0.02186938,"about_ca_system_score_codex":0.02186938,"about_ca_system_score_gemma":0.041669514,"threshold_uncertainty_score":0.15867418},"labels":[],"label_agreement":null},{"id":"W4386533916","doi":"10.1177/0193841x231198706","title":"Building a Logic Model to Foster Engagement and Learning Using the Case of a Province-Wide Multispecies Antimicrobial Use Monitoring System","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre Hospitalier de l’Université de Montréal; Université de Montréal","funders":"Ministère de l'Agriculture, des Pêcheries et de l'Alimentation","keywords":"Logic model; Computer science; Process (computing); Framing (construction); Corporate governance; Reflexivity; Management science; Knowledge management; Process management; Engineering; Sociology; Business; Social science","score_opus":0.8525169825921586,"score_gpt":0.6806434946012924,"score_spread":0.1718734879908662,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386533916","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":[0.12144165,0.0017202842,0.51491505,0.084098555,0.0002641719,0.0032964551,0.00029176925,0.00039498485,0.27357706],"genre_scores_gemma":[0.7457041,0.00092717365,0.24154964,0.00247224,0.000028594985,0.0018699446,0.0001594424,0.0000702779,0.0072185216],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9624364,0.0315432,0.0007002189,0.0011192926,0.002333491,0.0018673734],"domain_scores_gemma":[0.94936144,0.041362643,0.0014309898,0.0020165548,0.0044469666,0.0013814054],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04450496,0.00067795394,0.000526726,0.0037781547,0.010326642,0.011006696,0.0038126677,0.004194214,0.0058916886],"category_scores_gemma":[0.0373742,0.0005895469,0.0010117235,0.0033287846,0.016107937,0.01014255,0.007893235,0.0044868933,0.0005818471],"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":[0.000061677456,0.00024711248,0.0028653266,0.0006376936,0.000026297676,0.0010129713,0.04414202,0.012992013,0.0006013335,0.8843488,0.004415654,0.04864911],"study_design_scores_gemma":[0.00025027423,0.00039417524,0.0025073746,0.0029674126,0.00013510993,0.00054290355,0.13376048,0.05775742,0.0024721872,0.55957985,0.23949043,0.0001423939],"about_ca_topic_score_codex":0.08689316,"about_ca_topic_score_gemma":0.14655876,"teacher_disagreement_score":0.08689316,"about_ca_system_score_codex":0.046301346,"about_ca_system_score_gemma":0.061696228,"threshold_uncertainty_score":0.33594137},"labels":[],"label_agreement":null},{"id":"W4389722681","doi":"10.1177/0193841x231221303","title":"When Who Matters: Interviewer Effects and Survey Modality","year":2023,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"International Development Research Centre","keywords":"Interview; Phone; Modality (human–computer interaction); Psychology; Leverage (statistics); Modalities; Social psychology; Applied psychology; Survey data collection; Statistics; Computer science; Sociology","score_opus":0.4999020040566802,"score_gpt":0.557242604081877,"score_spread":0.05734060002519675,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389722681","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.41776508,0.036524814,0.43599272,0.026716193,0.003111252,0.033764344,0.003363408,0.0006735254,0.042088658],"genre_scores_gemma":[0.8431472,0.003109897,0.11858665,0.0077792224,0.00062880025,0.023203459,0.0004668292,0.00022403056,0.002854014],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.28644183,0.63667226,0.031796046,0.01697257,0.025523202,0.0025941068],"domain_scores_gemma":[0.17215739,0.75834244,0.026705993,0.030884044,0.010988127,0.0009220224],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.55336905,0.0009108789,0.0018398567,0.0026867099,0.0023614326,0.0047968146,0.0030063218,0.0031660888,0.0054152454],"category_scores_gemma":[0.7002502,0.0016890938,0.0033685493,0.0038304417,0.0077608847,0.006065155,0.0054149963,0.0032344377,0.00077908946],"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":[0.0070795743,0.0011369255,0.28112224,0.021570796,0.0097101135,0.0008114848,0.10517019,0.003517317,0.0068979333,0.04741219,0.011006542,0.50456476],"study_design_scores_gemma":[0.0035393178,0.0098008495,0.56991655,0.029917344,0.00888185,0.0024122256,0.053172436,0.017929753,0.030349411,0.13198557,0.14119823,0.0008964317],"about_ca_topic_score_codex":0.0037548856,"about_ca_topic_score_gemma":0.004279462,"teacher_disagreement_score":0.44663095,"about_ca_system_score_codex":0.003728111,"about_ca_system_score_gemma":0.0043347874,"threshold_uncertainty_score":0.5507755},"labels":[],"label_agreement":null},{"id":"W4409153149","doi":"10.1177/0193841x251331723","title":"A Critical Reflection of Generalization in Mixed Methods Research","year":2025,"lang":"en","type":"article","venue":"Evaluation Review","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Centre for Interdisciplinary Research in Rehabilitation","funders":"","keywords":"Generalization; Constructive; Multimethodology; Relevance (law); Management science; Computer science; Reflection (computer programming); Psychology; Epistemology; Mathematics education; Process (computing); Political science","score_opus":0.7561451234814716,"score_gpt":0.7902642498582428,"score_spread":0.03411912637677128,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409153149","genre_codex":"commentary","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007424458,0.02051372,0.25026864,0.6918397,0.014969317,0.002460383,0.00008825384,0.00029745945,0.012138076],"genre_scores_gemma":[0.2958269,0.014926763,0.42995572,0.235182,0.006831939,0.012620247,0.000077770084,0.0006904249,0.0038882284],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.18528564,0.69849735,0.042196184,0.016552178,0.054011576,0.0034570668],"domain_scores_gemma":[0.11112911,0.7448978,0.01748067,0.059275325,0.06471221,0.002504895],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6921039,0.0025856576,0.004043718,0.007271818,0.014548474,0.024782516,0.0086423475,0.020153148,0.0025001515],"category_scores_gemma":[0.74565065,0.0028861389,0.004632846,0.0047865687,0.08621913,0.046384405,0.023427853,0.061030217,0.00087157905],"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":[0.00027802808,0.00013995003,0.001515746,0.009278689,0.00031047908,0.0010717083,0.2569403,0.0011910805,0.0018313229,0.5890148,0.04048135,0.097946495],"study_design_scores_gemma":[0.00022177427,0.00042550114,0.0011905641,0.02987457,0.00031447515,0.001922274,0.057337984,0.0041868766,0.007962887,0.49186245,0.40432933,0.00037134375],"about_ca_topic_score_codex":0.0039021797,"about_ca_topic_score_gemma":0.0029338875,"teacher_disagreement_score":0.30789608,"about_ca_system_score_codex":0.027901195,"about_ca_system_score_gemma":0.047659997,"threshold_uncertainty_score":0.3796907},"labels":[],"label_agreement":null}]}