{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":27,"total_is_capped":false,"direct_labels_cover":1,"predictions_cover":27,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"query_hash":"bfc1c98d030d","filters":{"venue":"Evaluation Review"}},"results":[{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.03931785094118342,"gpt":0.3183677398194218,"spread":0.2790498888782385,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.008522069,0.0001080142,0.0001273912,0.0000205185,0.0001979175,0.00002536556,0.0001755772,0.00003353524,0.02909968],"category_scores_gemma":[0.0004146308,0.00006898178,0.00005686679,0.0002456905,0.00008692288,0.0005413591,0.00004205388,0.00006523972,0.001714038],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006102493,"about_ca_system_score_gemma":0.00006443036,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001379434,"about_ca_topic_score_gemma":0.00003626954,"domain_scores_codex":[0.9971328,0.000574542,0.0005478917,0.0001679962,0.001407017,0.0001697987],"domain_scores_gemma":[0.9992425,0.0000341724,0.0001160321,0.0004310971,0.0001051564,0.00007101958],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001465148,0.00007608457,0.004355144,0.000127443,0.000008371612,2.316822e-8,0.001451439,0.006513578,0.0001729934,0.00002405823,0.03002317,0.9572331],"study_design_scores_gemma":[0.0005637066,0.00009860824,0.3343098,0.000244792,0.0002841328,0.000003210238,0.0003605195,0.0151729,0.0009633874,0.002433717,0.6453086,0.0002566797],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9716162,0.001611837,0.00007630087,0.01218273,0.0001120108,0.004501005,0.000006875099,0.00002394281,0.009869121],"genre_scores_gemma":[0.9887487,0.002035556,0.0004205857,0.006955296,0.00002322971,0.0009249232,0.0001961446,0.00001087705,0.000684663],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9569764,"threshold_uncertainty_score":0.9990633,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.4177310589938517,"gpt":0.4908106349377258,"spread":0.0730795759438741,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003933822,0.0001354788,0.0002941571,0.00006108786,0.0007281493,0.0001407482,0.000114128,0.00009806615,0.000217032],"category_scores_gemma":[0.000724255,0.0001330639,0.0001057631,0.0007745603,0.0002839718,0.0005010892,0.00003760715,0.0001027938,0.000003878833],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002222355,"about_ca_system_score_gemma":0.00132428,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001640019,"about_ca_topic_score_gemma":0.001400999,"domain_scores_codex":[0.9971701,0.000601035,0.0006355171,0.000259591,0.001041899,0.000291847],"domain_scores_gemma":[0.9980239,0.00005736289,0.0003572881,0.0001491615,0.001327203,0.00008511316],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000002246015,0.0004113764,0.05159513,0.0006373801,0.00002383345,5.879195e-7,0.001075117,0.000006329652,0.000035382,0.001507144,0.001690979,0.9430145],"study_design_scores_gemma":[0.0008434724,0.0001819013,0.6391178,0.007551423,0.001333339,0.000003069884,0.001100662,0.005366595,0.00002460271,0.004865526,0.3387843,0.0008273429],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8895828,0.07275344,0.0001836874,0.002449463,0.0003588371,0.01989389,0.00002884721,0.0006886286,0.01406043],"genre_scores_gemma":[0.8445495,0.01220033,0.1381912,0.0004819226,0.0003642499,0.003134976,0.0004073691,0.00003562596,0.0006347392],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9421871,"threshold_uncertainty_score":0.560041,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.1469731598734872,"gpt":0.4142079531210253,"spread":0.2672347932475381,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006101822,0.00003219589,0.00006689497,0.00002454656,0.000441427,0.00001991998,0.0001403121,0.000007509102,0.0005000825],"category_scores_gemma":[0.0005146426,0.00002156914,0.000009624521,0.0002277724,0.0002564819,0.00008718786,0.00002731165,0.00002674705,0.00002424927],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001307115,"about_ca_system_score_gemma":0.0001470755,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002946814,"about_ca_topic_score_gemma":0.00022829,"domain_scores_codex":[0.9986176,0.0001335407,0.0002289481,0.0000918991,0.000834991,0.0000930133],"domain_scores_gemma":[0.9996834,0.00007343141,0.00007987999,0.00007734232,0.00005583197,0.00003007712],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[8.276463e-7,0.0002457505,0.07872472,0.0001772896,0.00001349079,5.568775e-8,0.01983819,0.00000871645,0.000029054,0.07124815,0.00117943,0.8285343],"study_design_scores_gemma":[0.0001615156,0.000007820983,0.7566562,0.0004623121,0.00001637953,5.030259e-8,0.001851988,0.00005325259,0.00001183083,0.001000166,0.2397086,0.00006986756],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7474932,0.1710936,0.00002207817,0.04270848,0.0004141703,0.00190357,7.193343e-7,0.000008044585,0.0363561],"genre_scores_gemma":[0.9453784,0.05174657,0.001018299,0.0004151953,0.00001786592,0.00009268076,8.820853e-7,0.000001783548,0.001328334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8284644,"threshold_uncertainty_score":0.5475553,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.1699745819640804,"gpt":0.4506688926621983,"spread":0.280694310698118,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001818345,0.000601585,0.003093373,0.0001251455,0.00005772881,0.00001470964,0.000129409,0.0002488247,0.000462602],"category_scores_gemma":[0.0009180235,0.00041192,0.001301837,0.0003815917,0.00002227392,0.00005114497,0.00001964165,0.000192773,0.0005673441],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001659174,"about_ca_system_score_gemma":0.001040212,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001570115,"about_ca_topic_score_gemma":0.000002510892,"domain_scores_codex":[0.9962899,0.0008321722,0.001090443,0.0006450634,0.0008111657,0.000331221],"domain_scores_gemma":[0.9976773,0.000155042,0.0007158339,0.0008759795,0.0003801604,0.0001956667],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000006601637,0.000162373,0.000002306718,0.2143278,0.0001469411,0.000002659793,0.000004877371,5.111688e-8,3.314724e-8,0.00001785741,0.005423326,0.7799052],"study_design_scores_gemma":[0.0009141027,0.0003984353,0.000004814814,0.2984732,0.01045236,0.00002729242,3.368238e-7,0.000007908689,4.135323e-7,0.00001298579,0.6894883,0.0002199166],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[2.315016e-7,0.9705626,0.00009261779,0.00005171854,0.0006850623,0.0272033,0.00005663485,0.00004877331,0.001299106],"genre_scores_gemma":[0.000001950754,0.9882061,0.0005082043,0.000577131,0.0001941755,0.007862859,0.001740985,0.00009171775,0.0008168239],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7796853,"threshold_uncertainty_score":0.9998333,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2607821216003572,"gpt":0.4468930258543393,"spread":0.186110904253982,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009854336,0.0001524116,0.0002881814,0.00005888226,0.000268266,0.00003309076,0.0001236886,0.00006864298,0.00007283298],"category_scores_gemma":[0.003811213,0.0001100146,0.0001148918,0.0004349775,0.00001404219,0.000200778,0.00001542038,0.00008794235,0.00000445065],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001429924,"about_ca_system_score_gemma":0.00003691419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000264493,"about_ca_topic_score_gemma":0.00004646304,"domain_scores_codex":[0.9978265,0.0002319393,0.0008670551,0.0002458568,0.0006385429,0.0001901081],"domain_scores_gemma":[0.9977015,0.0005473845,0.0006670585,0.0003920997,0.0006523204,0.00003969224],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002995044,0.0001845559,0.01592738,0.006773208,0.00006737549,2.118391e-7,0.0004283026,0.007539414,0.00004448927,0.07929745,0.007860454,0.8818472],"study_design_scores_gemma":[0.0006104342,0.00003022286,0.07606599,0.001102106,0.0004429787,0.000004143083,0.00004036001,0.8857867,0.00002065102,0.03466309,0.00100708,0.0002263053],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06063867,0.003525156,0.9276569,0.0009074908,0.0002959404,0.005875107,0.00001230453,0.0001549929,0.0009333874],"genre_scores_gemma":[0.5472116,0.0001589976,0.4492462,0.0005755618,0.0003146212,0.0004829002,0.001878844,0.00003722238,0.00009406124],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8816209,"threshold_uncertainty_score":0.4562654,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.08272507338278332,"gpt":0.3832698896296803,"spread":0.300544816246897,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00040174,0.0001299876,0.0002419358,0.00004732988,0.0001241232,0.000008014927,0.00002177927,0.00002088048,0.00006054346],"category_scores_gemma":[0.00003659648,0.00008325312,0.000068141,0.0001025756,0.00004799043,0.0001370187,0.000003114907,0.00003795163,0.00001236659],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002065552,"about_ca_system_score_gemma":0.00001599907,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000003945996,"about_ca_topic_score_gemma":0.00005105724,"domain_scores_codex":[0.9989886,0.0001085688,0.0003079419,0.0001897676,0.0002979231,0.0001071962],"domain_scores_gemma":[0.9994007,0.0001432243,0.000174289,0.0001971341,0.0000448809,0.0000397487],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0005814527,0.0003073663,0.08114472,0.0004829185,0.0004130939,0.00000470904,0.00257178,0.00002977148,0.00005076162,0.00007809319,0.00103491,0.9133004],"study_design_scores_gemma":[0.02218011,0.004641865,0.8502916,0.0217997,0.005676661,0.00008120426,0.002612398,0.003474741,0.004313145,0.0007310978,0.08331231,0.0008851928],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"review","genre_scores_codex":[0.9432707,0.0533506,0.000009266117,0.0004695751,0.0001271813,0.00158504,0.000001830786,0.0000152004,0.001170555],"genre_scores_gemma":[0.4513455,0.5478571,0.000081233,0.0002282876,0.00001431647,0.0001964264,0.00003607387,0.000007210756,0.0002338486],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9124152,"threshold_uncertainty_score":0.3394963,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2809426709750991,"gpt":0.5545025882861809,"spread":0.2735599173110818,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009240041,0.0001807196,0.000587937,0.0002929891,0.000259486,0.000007225496,0.0001396874,0.00008785821,0.0003012068],"category_scores_gemma":[0.00254147,0.0001485942,0.00002881182,0.0007927479,0.00003650286,0.00025917,0.00009457317,0.0004498969,0.00004897218],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002847955,"about_ca_system_score_gemma":0.0003286173,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007560711,"about_ca_topic_score_gemma":0.0001817751,"domain_scores_codex":[0.9928931,0.004510595,0.001210556,0.0005065288,0.0005418982,0.0003373148],"domain_scores_gemma":[0.9978727,0.0002240845,0.0006875162,0.0005124942,0.0005502262,0.0001529839],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001492905,0.001058316,0.9751489,0.002907385,0.000006946454,0.000003137131,0.001481655,7.606637e-7,0.00009879198,0.00005269251,0.0001737137,0.01891844],"study_design_scores_gemma":[0.001093114,0.0003021534,0.977722,0.007197579,0.00008282934,0.000002017356,0.0007319268,0.0001076302,0.000001419442,0.00003181844,0.01258714,0.0001403642],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9844543,0.009123866,0.00002647716,0.001026466,0.0004399494,0.004688606,0.000007672382,0.00002701069,0.0002055958],"genre_scores_gemma":[0.9846287,0.01262055,0.00004630632,0.001223315,0.0002977931,0.0001847398,0.00004557393,0.00002400896,0.0009290169],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01877808,"threshold_uncertainty_score":0.6059497,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.05501168353014899,"gpt":0.4353426154747834,"spread":0.3803309319446344,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001589347,0.0001161015,0.0006808646,0.00005314321,0.0000174334,0.000004658243,0.0001113014,0.00002253133,0.008334897],"category_scores_gemma":[0.0002105204,0.00009584516,0.0001092726,0.0002999877,0.00001416392,0.0001046897,0.00002804246,0.0001357895,0.0004986352],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002273523,"about_ca_system_score_gemma":0.0009999975,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001352947,"about_ca_topic_score_gemma":0.0005573278,"domain_scores_codex":[0.9977014,0.0003254355,0.0005692873,0.0002038539,0.001040008,0.0001599892],"domain_scores_gemma":[0.9992254,0.00004851395,0.0001587734,0.0003193067,0.00008005089,0.0001679918],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000006992152,0.001999575,0.8223652,0.001259629,0.0000398778,6.562985e-7,0.00002368352,0.000002609185,0.000002646945,0.00005583644,0.00558662,0.1686567],"study_design_scores_gemma":[0.001492129,0.0001856338,0.989897,0.002697116,0.000152272,0.000001215588,0.000004961118,0.00003419081,0.00001301833,0.0003604232,0.005094992,0.00006704271],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9364591,0.04542128,0.000002277662,0.01321935,0.00005282728,0.002850153,0.000002143777,0.00002116368,0.001971687],"genre_scores_gemma":[0.939469,0.04351452,0.0004246769,0.01436665,0.0000945404,0.00002984377,0.00007136937,0.00001796179,0.002011438],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1685896,"threshold_uncertainty_score":0.9925716,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Nexus (standard); Geopolitics; Tourism; Political science; Economics; Regional science; Geography; Computer science","retraction":{"nature":"Expression of concern","openalex_flagged":false},"screen_n_in":null,"score":{"opus":0.1068698492901642,"gpt":0.3450513100099599,"spread":0.2381814607197958,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01135692,0.0001275861,0.0003465902,0.00007576479,0.0006521938,0.0001039961,0.0003757759,0.00002460781,0.002219213],"category_scores_gemma":[0.004544131,0.00009089265,0.0001051087,0.0006642952,0.00009293205,0.00008955182,0.000239228,0.0003272945,0.00002617636],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002128489,"about_ca_system_score_gemma":0.00008012902,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01469276,"about_ca_topic_score_gemma":0.00012576,"domain_scores_codex":[0.9976046,0.001034247,0.0006685035,0.0003403409,0.0001432672,0.0002090052],"domain_scores_gemma":[0.9967813,0.002034365,0.0004692925,0.0005634421,0.00008658192,0.00006495147],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001305866,0.00003600716,0.9143319,0.0002308404,0.0002082903,5.573488e-7,0.0009995161,0.001486071,1.064189e-7,0.03433916,0.008102235,0.04025229],"study_design_scores_gemma":[0.0001682632,0.00003045562,0.2388498,0.0001896115,0.0001539064,8.003263e-7,0.0001242152,0.4164659,5.33807e-8,0.2425739,0.101287,0.0001560463],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.535503,0.4028267,0.002184984,0.05343141,0.0002056184,0.001788322,0.002070542,0.00003300036,0.001956414],"genre_scores_gemma":[0.9288128,0.06804453,0.00005235277,0.002228009,0.000215488,0.0001664626,0.0003883308,0.000013174,0.0000788382],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6754821,"threshold_uncertainty_score":0.9986929,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.1513130385005866,"gpt":0.490324462403093,"spread":0.3390114239025064,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02164221,0.0002924701,0.001361306,0.0002732878,0.0001854147,0.00003977411,0.000550732,0.0002762719,0.003582294],"category_scores_gemma":[0.0007442302,0.0002677153,0.0003280238,0.001402511,0.00007198304,0.0002743365,0.00002974352,0.0001829233,0.0004777499],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007743801,"about_ca_system_score_gemma":0.002391969,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001923745,"about_ca_topic_score_gemma":0.0004958441,"domain_scores_codex":[0.9896031,0.006053203,0.001339462,0.0005523263,0.002094107,0.0003577883],"domain_scores_gemma":[0.997954,0.0001141035,0.000710221,0.000470461,0.0005873983,0.0001638462],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000002028353,0.00007074104,0.00006471948,0.002679606,0.0000222545,1.182781e-7,0.004524738,0.000001997716,1.642439e-8,0.001067497,0.000122405,0.9914439],"study_design_scores_gemma":[0.0001977484,0.00002432593,0.00009975346,0.02578347,0.001571657,2.42975e-7,0.0002198097,0.0000444801,3.017592e-8,0.0007170102,0.9709998,0.000341714],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001740162,0.9658009,0.0001443568,0.0001090175,0.000177091,0.01018837,0.00005334918,0.00003340822,0.02347612],"genre_scores_gemma":[0.0005752715,0.995446,0.0001463261,0.0002272315,0.0001967679,0.001543436,0.00167139,0.00003151152,0.0001620128],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9911022,"threshold_uncertainty_score":0.9999775,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8475988642781547,"gpt":0.6249450003263772,"spread":0.2226538639517774,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.1042227,0.0001572819,0.0003388576,0.0003105557,0.0001096437,0.00009458444,0.0002778601,0.0000572667,0.02495021],"category_scores_gemma":[0.01195731,0.0001211532,0.00009204786,0.001459309,0.00002389697,0.001565213,0.00004417504,0.0001396639,0.002704043],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000415951,"about_ca_system_score_gemma":0.0004585415,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008936398,"about_ca_topic_score_gemma":0.00006749839,"domain_scores_codex":[0.9864475,0.003579991,0.001510536,0.0003878839,0.007679221,0.0003949034],"domain_scores_gemma":[0.9963818,0.0004863139,0.000517922,0.0006554922,0.001820913,0.0001375609],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.000005376884,0.0001720019,0.08618295,0.00008377084,0.000008492564,7.532351e-8,0.0007156972,0.002130831,0.00006354203,0.0008978101,0.005772951,0.9039665],"study_design_scores_gemma":[0.001191527,0.00003484978,0.8213325,0.0009253917,0.0002246177,0.000002684009,0.000230445,0.1206068,0.0001753445,0.004824496,0.05014956,0.0003017812],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6679611,0.187464,0.002532478,0.009724199,0.002581865,0.008335941,0.000006760418,0.00008862073,0.121305],"genre_scores_gemma":[0.9943848,0.002374301,0.0003353478,0.001576749,0.0001498385,0.0008956148,0.00004401637,0.00001048849,0.0002288839],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9036647,"threshold_uncertainty_score":0.9980724,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Quantile; Tourism; Nexus (standard); Quantile regression; Granger causality; Destinations; Econometrics; China; Geography; Economic geography; Economics; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.05660189426987145,"gpt":0.2808662194890804,"spread":0.224264325219209,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00133473,0.0001828454,0.0003360574,0.0002086088,0.0001557947,0.00009555586,0.0002874946,0.0000761417,0.0003008984],"category_scores_gemma":[0.0003648038,0.0001802083,0.00007517645,0.0002352427,0.00005596747,0.0003009415,0.0000939025,0.0001575339,0.0005990287],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003977363,"about_ca_system_score_gemma":0.00002695097,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001406427,"about_ca_topic_score_gemma":0.00004953091,"domain_scores_codex":[0.9981271,0.0001206949,0.0008730513,0.0005375473,0.0001542263,0.0001873642],"domain_scores_gemma":[0.9987254,0.0002893419,0.0005211222,0.0003831371,0.000008430039,0.00007262297],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00004045216,0.0005671537,0.3451363,0.000354808,0.0005648482,0.000008811171,0.00160967,0.0120634,0.00007421556,0.1695925,0.02417519,0.4458127],"study_design_scores_gemma":[0.001158626,0.0000433043,0.6869699,0.0002807018,0.00005047399,0.000001521999,0.0001580647,0.1925405,0.00001516671,0.06834165,0.05004507,0.0003950121],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9012896,0.07578073,0.003356689,0.006211338,0.0008003011,0.001952839,0.0003106882,0.0000763877,0.01022143],"genre_scores_gemma":[0.9488093,0.04681316,0.000639759,0.0003173321,0.0001397936,0.0002283889,0.001956691,0.00003065975,0.001064922],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4454177,"threshold_uncertainty_score":0.7699502,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.1040357046427933,"gpt":0.243642188875947,"spread":0.1396064842331537,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00264861,0.000141709,0.0004065917,0.00006087702,0.00005500268,0.00001583961,0.0002546576,0.00004085248,0.00006846304],"category_scores_gemma":[0.0009872736,0.0001092127,0.00004089525,0.0001695304,0.00003768907,0.00008576559,0.00008630059,0.00008214864,0.00001422509],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005353052,"about_ca_system_score_gemma":0.0001747965,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.4531868,"about_ca_topic_score_gemma":0.2022767,"domain_scores_codex":[0.9984996,0.0002001521,0.000647179,0.0004066791,0.00007871123,0.0001676213],"domain_scores_gemma":[0.9978076,0.001196978,0.000494067,0.000455631,0.000007439799,0.0000382859],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00003649517,0.00007809598,0.8805548,0.002300233,0.0002698874,0.000005497663,0.0009058607,0.03795373,0.0002040077,0.009532426,0.01559755,0.05256141],"study_design_scores_gemma":[0.001015742,0.0001289409,0.6306985,0.005607964,0.0001180968,0.000001571117,0.0001676567,0.3383105,0.001631604,0.01087478,0.01089022,0.0005544242],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8575931,0.1354706,0.00001811612,0.002613446,0.0003151658,0.0006022375,0.00002595007,0.000009942486,0.003351393],"genre_scores_gemma":[0.8935496,0.105447,0.00004077326,0.00055482,0.00006939352,0.0001815104,0.00002806265,0.00001885391,0.0001099917],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3003568,"threshold_uncertainty_score":0.8122796,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.13375186684759,"gpt":0.4296849130186229,"spread":0.295933046171033,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01105132,0.0001556219,0.0003067193,0.00005026462,0.0009921237,0.0001824539,0.0001460267,0.00005953643,0.00004394076],"category_scores_gemma":[0.001039346,0.0001150583,0.00008431436,0.0003153377,0.00004682883,0.0007478382,0.00003129736,0.00009276564,0.000005804692],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002877343,"about_ca_system_score_gemma":0.0002079755,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008143677,"about_ca_topic_score_gemma":0.001430289,"domain_scores_codex":[0.9966604,0.001461684,0.0004892058,0.0002908553,0.0006599687,0.0004379005],"domain_scores_gemma":[0.9986667,0.0001811265,0.0004334202,0.0001909385,0.0004394337,0.00008838181],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002458652,0.0009662483,0.3011659,0.02622821,0.0001104803,0.000003534452,0.03651638,0.0239012,0.001960844,0.01155356,0.00149067,0.5960784],"study_design_scores_gemma":[0.001755582,0.0004328878,0.3328753,0.01531295,0.001231726,0.000003084034,0.002594429,0.6151956,0.0001376839,0.02350044,0.005871208,0.001089164],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9271009,0.001858032,0.06122055,0.0005501242,0.0005982621,0.008306177,0.00000671827,0.000165984,0.0001932753],"genre_scores_gemma":[0.9199229,0.0004073479,0.07815892,0.0001779333,0.0003118245,0.0007438354,0.0002163721,0.00002204889,0.00003879666],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5949892,"threshold_uncertainty_score":0.7630714,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2853905940586414,"gpt":0.4279131534136509,"spread":0.1425225593550095,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001701816,0.0001367363,0.0003297667,0.0001081198,0.00003043505,0.00002670479,0.00005852464,0.00004293987,0.00006008543],"category_scores_gemma":[0.001946089,0.0001121779,0.00002021668,0.0006285322,0.00003320345,0.0003110359,0.000004688174,0.0001155391,0.00001915859],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004055639,"about_ca_system_score_gemma":0.0002287174,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004576074,"about_ca_topic_score_gemma":0.0001849955,"domain_scores_codex":[0.9985747,0.000171575,0.0003039,0.0002706535,0.0005201361,0.0001590702],"domain_scores_gemma":[0.9990587,0.0002128286,0.0001141375,0.0001877964,0.0002232549,0.0002032525],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002218279,0.0003412537,0.7392154,0.003666906,0.0002251328,0.00006303735,0.1632565,0.0005229351,0.00008951478,0.00277468,0.001906353,0.08771653],"study_design_scores_gemma":[0.0312207,0.005473448,0.1765932,0.3360594,0.004674951,0.0002433903,0.06102114,0.007075288,0.003071923,0.3668001,0.004047357,0.003719165],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9355708,0.0597648,0.001077599,0.001621021,0.00009619196,0.001461871,0.000001350608,0.00003724315,0.0003690841],"genre_scores_gemma":[0.9621578,0.02211427,0.01051977,0.004830126,0.0001065591,0.0001847776,0.00004268497,0.00002088605,0.00002315658],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5626222,"threshold_uncertainty_score":0.4574482,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.05548297578315207,"gpt":0.3011743031318248,"spread":0.2456913273486727,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001029658,0.0001117297,0.0002176115,0.0001793887,0.0002298217,0.00004296363,0.0001978113,0.00003632544,0.0003026344],"category_scores_gemma":[0.0005866761,0.0001036323,0.00006349192,0.000186167,0.00008730563,0.0002608357,0.0001151007,0.0001019155,0.000007964999],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002058378,"about_ca_system_score_gemma":0.00006274191,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003170347,"about_ca_topic_score_gemma":0.00003011734,"domain_scores_codex":[0.9989838,0.00007923468,0.0004178098,0.000297913,0.0001364642,0.00008477529],"domain_scores_gemma":[0.9989049,0.0001524791,0.000479097,0.000158541,0.0002806338,0.00002436436],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004966412,0.0009548842,0.1323973,0.0009392884,0.001716167,0.000003825046,0.1037564,0.0004214743,0.0003719682,0.2729991,0.05750207,0.4284409],"study_design_scores_gemma":[0.006248095,0.001203068,0.1938786,0.00150089,0.000616161,0.00002031955,0.04313193,0.04060987,0.000365261,0.3990287,0.3122154,0.001181612],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8913289,0.07159904,0.0007840061,0.03387218,0.000622167,0.001211144,0.0002170413,0.00003171685,0.0003337887],"genre_scores_gemma":[0.9767811,0.02097064,0.0005525587,0.0004820823,0.00005192899,0.0001178143,0.000799568,0.00001074216,0.0002335204],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4272593,"threshold_uncertainty_score":0.4226004,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.2812821955724736,"gpt":0.5541621417145391,"spread":0.2728799461420656,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.003942708,0.0000692418,0.0002450062,0.00001260562,0.0006200742,0.000004860008,0.0001224349,0.00005436107,0.003673002],"category_scores_gemma":[0.0005670001,0.0000411764,0.0000584437,0.0001556098,0.00001976027,0.00006037037,0.00003186331,0.0002270361,0.001524555],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001286288,"about_ca_system_score_gemma":0.0002802328,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000111089,"about_ca_topic_score_gemma":0.0005940385,"domain_scores_codex":[0.9981011,0.0008807564,0.0003373508,0.0001189552,0.0003922992,0.0001695846],"domain_scores_gemma":[0.9991457,0.0001362586,0.0001614991,0.000245657,0.0002774217,0.00003348916],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003775545,0.00009372372,0.00397015,0.008367574,0.00006661751,0.000001789151,0.1022377,0.00003774441,0.000007466545,0.2702205,0.002682806,0.6123101],"study_design_scores_gemma":[0.002469785,0.00007786295,0.03566716,0.03417498,0.0005738381,2.927097e-7,0.03554402,0.00006648571,0.00001004645,0.08293495,0.8079817,0.0004988554],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.04304935,0.8026453,0.0004222663,0.06223267,0.002403604,0.005736138,0.000003538171,0.0001459433,0.08336125],"genre_scores_gemma":[0.8842191,0.112592,0.00002395966,0.002004816,0.0003170378,0.000540075,0.000009717712,0.000009529813,0.0002837171],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8411698,"threshold_uncertainty_score":0.9992529,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.09025164283672198,"gpt":0.4246290602829653,"spread":0.3343774174462433,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003573218,0.00007754361,0.000134271,0.00004498247,0.0001274987,0.00003568128,0.0001031222,0.0000263366,0.001984334],"category_scores_gemma":[0.005941437,0.00006775757,0.0000338242,0.0003271755,0.00009413237,0.0002838282,0.00001034418,0.00006699379,0.0002974708],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006122967,"about_ca_system_score_gemma":0.0003174189,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001373416,"about_ca_topic_score_gemma":0.002827264,"domain_scores_codex":[0.9979028,0.0007831537,0.0003065381,0.0001870084,0.0006592926,0.0001612598],"domain_scores_gemma":[0.9992738,0.000245108,0.00006875413,0.0001396724,0.0002149292,0.00005770209],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000007304116,0.0004939566,0.01807885,0.001227885,0.000004773269,4.745235e-8,0.1788671,0.00005245122,0.000001178165,0.0009690039,0.001671132,0.7986263],"study_design_scores_gemma":[0.002390797,0.0003103945,0.03754926,0.02904826,0.0003089125,0.000002792279,0.2580985,0.01037135,0.00003187327,0.002583935,0.6580333,0.001270728],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8915529,0.01930655,0.000228221,0.0284839,0.0003186532,0.004952396,0.000002861554,0.0002190775,0.05493545],"genre_scores_gemma":[0.9954304,0.002303706,0.0006892208,0.000847874,0.00007784949,0.0002470105,0.000006947118,0.00000507114,0.0003919489],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7973556,"threshold_uncertainty_score":0.998928,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.08182174885329145,"gpt":0.4737520539230007,"spread":0.3919303050697093,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008270741,0.00004294807,0.00009430989,0.00001683059,0.0003561829,0.00001429797,0.00006845102,0.00003870901,0.0024801],"category_scores_gemma":[0.004905861,0.00003796718,0.00002520936,0.0001948968,0.00009720088,0.0001467853,0.00001796394,0.00006493048,0.0001599031],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005024173,"about_ca_system_score_gemma":0.0001539733,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004505376,"about_ca_topic_score_gemma":0.00009762585,"domain_scores_codex":[0.9974698,0.001506634,0.0001744016,0.0001277679,0.0006374296,0.00008395069],"domain_scores_gemma":[0.9990515,0.0003353658,0.00008958454,0.000117265,0.0003442645,0.00006209367],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[2.93445e-7,0.000009436711,0.002219,0.00008734815,0.000002857627,2.018906e-8,0.0002188158,0.00000102448,0.000001665169,0.9314386,0.01188877,0.05413218],"study_design_scores_gemma":[0.00009824541,0.000005526345,0.04649836,0.0007094136,0.00005492748,2.939949e-7,0.00003495615,0.0001431951,6.957817e-7,0.4694909,0.4828925,0.00007105372],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.004490334,0.0669491,0.0003391152,0.05057369,0.0004416257,0.0009312311,0.000004639089,0.00006074645,0.8762095],"genre_scores_gemma":[0.9519466,0.03559459,0.001858522,0.006798548,0.0003642609,0.0001218131,0.00005192188,0.000007535082,0.003256187],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9474563,"threshold_uncertainty_score":0.9984317,"prediction_status":"machine_predicted_unvalidated"},"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":"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Mathematics education; Computer science; Contrast (vision); Multilevel model; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.1238623546586106,"gpt":0.4557607482728444,"spread":0.3318983936142338,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003435434,0.00005834465,0.0001460002,0.00007115903,0.00006673844,0.00007001984,0.0001475111,0.00002978066,0.0008564319],"category_scores_gemma":[0.001691365,0.00005379604,0.00003589901,0.00034237,0.00001291688,0.00008065976,0.00001220451,0.00005380819,0.0000778363],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001313503,"about_ca_system_score_gemma":0.0002526996,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005400511,"about_ca_topic_score_gemma":0.001323836,"domain_scores_codex":[0.9988692,0.0002424782,0.0002146305,0.0001066029,0.0004535004,0.0001136332],"domain_scores_gemma":[0.9995085,0.00007547003,0.00008997104,0.0001807967,0.00009063133,0.00005463135],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[1.578336e-7,0.0000917476,0.0001669645,0.0003922849,0.000001752023,1.85785e-7,0.004447951,0.000001362946,0.000009687007,0.7345206,0.1733908,0.08697648],"study_design_scores_gemma":[0.00006917994,0.000006779366,0.0002645439,0.001371353,0.00002225658,3.573014e-7,0.0002074795,0.00008788129,0.00001558953,0.3725164,0.62534,0.00009811601],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"methods","genre_scores_codex":[0.004167468,0.03211106,0.07969458,0.2883002,0.0004343022,0.006902133,0.000001505033,0.0004040502,0.5879847],"genre_scores_gemma":[0.3290936,0.0175544,0.4849736,0.01145751,0.001846383,0.001646855,0.00005735518,0.00006928646,0.1533009],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.4519492,"threshold_uncertainty_score":0.9377328,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.4512052275898277,"gpt":0.4632224952179479,"spread":0.01201726762812022,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.02544826,0.0006103596,0.001089301,0.0004763198,0.0004919277,0.001267669,0.00151798,0.0001376005,0.0231819],"category_scores_gemma":[0.006840893,0.0003164972,0.0002197864,0.003283017,0.0002480286,0.002438165,0.0002162964,0.0001682136,0.001180368],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004785463,"about_ca_system_score_gemma":0.002078918,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002438141,"about_ca_topic_score_gemma":0.0003772673,"domain_scores_codex":[0.9772109,0.00516171,0.001603834,0.00178119,0.01366385,0.0005784689],"domain_scores_gemma":[0.9890987,0.001665906,0.001405476,0.001820398,0.005571951,0.0004375296],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001303742,0.0002080382,0.03701642,0.00009240625,0.0001819634,0.000001343766,0.007926441,0.000252876,0.0001170215,0.001008857,0.01014433,0.9429199],"study_design_scores_gemma":[0.02573119,0.003795957,0.3330787,0.02531322,0.006681724,0.00003217205,0.2330134,0.07742056,0.0108191,0.06644071,0.2094437,0.008229624],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8915538,0.03533217,0.01225706,0.04444286,0.001119952,0.008307938,0.0001232295,0.0001087165,0.006754286],"genre_scores_gemma":[0.9875416,0.002454579,0.001413111,0.001146587,0.0002551433,0.006001501,0.00008972424,0.00004012029,0.001057634],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9346903,"threshold_uncertainty_score":0.9999287,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.7561451234814716,"gpt":0.7902642498582428,"spread":0.03411912637677128,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1237473,0.00007601793,0.0003260754,0.0008254117,0.0001011294,0.00006104937,0.0002947686,0.00006238579,0.003974897],"category_scores_gemma":[0.05347765,0.00005885717,0.00007578693,0.004820862,0.00007870713,0.0003281501,0.00007010957,0.0001643374,0.0001193963],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001942039,"about_ca_system_score_gemma":0.0007832148,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001954656,"about_ca_topic_score_gemma":0.00008094717,"domain_scores_codex":[0.9825403,0.01215385,0.001455338,0.0003983249,0.003265872,0.0001862961],"domain_scores_gemma":[0.9918004,0.002914638,0.0001820613,0.0005307549,0.004536528,0.00003567457],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000009694826,0.00007889364,0.001758325,0.0003373392,0.000005509396,9.647043e-8,0.0001181221,0.0004529586,0.001231755,0.04759291,0.01839804,0.9300163],"study_design_scores_gemma":[0.0009645883,0.0001486285,0.09582551,0.004445506,0.0001155571,0.000001949782,0.0004482165,0.2964904,0.004209167,0.3791482,0.217993,0.0002093143],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02642872,0.2161309,0.3893906,0.07450326,0.004043583,0.006459859,0.000007251276,0.00006773447,0.2829681],"genre_scores_gemma":[0.8900823,0.03489848,0.06545601,0.004164876,0.0001142427,0.001122667,0.00005305173,0.00001684691,0.004091512],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9298071,"threshold_uncertainty_score":0.9969356,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.8525169825921586,"gpt":0.6806434946012924,"spread":0.1718734879908662,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01790403,0.0001167501,0.0002821652,0.0001526542,0.001296168,0.00002862688,0.0001184798,0.00003720103,0.00003225926],"category_scores_gemma":[0.008019641,0.00008502791,0.00003599218,0.0006721594,0.00003183409,0.0002430752,0.0002622019,0.0003057497,0.00004197228],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003717534,"about_ca_system_score_gemma":0.0005522406,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002276231,"about_ca_topic_score_gemma":0.00005636528,"domain_scores_codex":[0.9937076,0.004283885,0.0009354841,0.0002689602,0.0004219258,0.000382171],"domain_scores_gemma":[0.996274,0.002380078,0.0005016767,0.0002548248,0.0004856292,0.0001038096],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000871209,0.00007588263,0.06112046,0.1365702,0.000176202,0.0001818309,0.2000299,0.2355048,0.05656067,0.007564952,0.04507449,0.2570535],"study_design_scores_gemma":[0.001413157,0.00009112113,0.007221616,0.03446103,0.0003692546,0.00007881175,0.01940871,0.9206579,0.0003351935,0.00007437605,0.01545903,0.0004298376],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9720006,0.0006764837,0.01367763,0.007221821,0.0002875428,0.005967696,0.00001758591,0.0000856723,0.0000649872],"genre_scores_gemma":[0.9694942,0.001224529,0.02194304,0.00648626,0.000117982,0.0005788199,0.000002737934,0.00002220758,0.0001302552],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6851531,"threshold_uncertainty_score":0.9969208,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Victoria","funders":"","keywords":"Normative; Report card; Outcome (game theory); Psychology; Political science; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.3399599014597608,"gpt":0.588645813129939,"spread":0.2486859116701782,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.01589316,0.0001599008,0.0009033907,0.00004618695,0.000626235,0.000008572465,0.0001341247,0.0001784055,0.00700877],"category_scores_gemma":[0.010922,0.000125931,0.0001358312,0.0004232129,0.00002625095,0.0003539641,0.00007331522,0.0005485794,0.002294539],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002136961,"about_ca_system_score_gemma":0.0001756558,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005380958,"about_ca_topic_score_gemma":0.00004954136,"domain_scores_codex":[0.9944052,0.001988825,0.002207624,0.0001711628,0.0007248844,0.000502353],"domain_scores_gemma":[0.99567,0.0005834257,0.002509213,0.0003803658,0.0006971565,0.0001598744],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000003185067,0.0000762141,0.8041794,0.007635277,0.00003995375,9.128202e-7,0.04320598,0.000002055662,0.000002332454,0.01913517,0.0005535409,0.125166],"study_design_scores_gemma":[0.0007777304,0.00003327821,0.7580004,0.02830599,0.0008323097,8.557414e-7,0.0738166,0.0002110586,0.00001404487,0.01098969,0.1260771,0.0009408747],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5356824,0.3670297,0.004101579,0.002564986,0.005066707,0.006441473,0.00001425132,0.0003402498,0.07875858],"genre_scores_gemma":[0.9860975,0.009819901,0.0006589536,0.001487945,0.0007783576,0.0008717691,0.00004097363,0.00002424367,0.0002203269],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4504151,"threshold_uncertainty_score":0.9984823,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.3855258045495976,"gpt":0.6285228913714911,"spread":0.2429970868218935,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01460245,0.000135259,0.0003526245,0.0002151158,0.0002263338,0.0004681443,0.001174441,0.00002019565,0.004480011],"category_scores_gemma":[0.007822216,0.00007859281,0.0001030412,0.001658742,0.00004793152,0.0009241071,0.0003926483,0.00004006,0.0002243304],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001137447,"about_ca_system_score_gemma":0.0009041511,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007242469,"about_ca_topic_score_gemma":0.001023289,"domain_scores_codex":[0.9956521,0.0009699126,0.0007650487,0.0005642118,0.001904584,0.0001441607],"domain_scores_gemma":[0.9934931,0.001418521,0.0004630261,0.001870534,0.002580629,0.0001742394],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00002301129,0.00034859,0.05552524,0.00002084095,0.0003100172,4.01626e-8,0.001274397,0.000150352,0.002533385,0.00007632827,0.1328283,0.8069095],"study_design_scores_gemma":[0.0001032838,0.0002513797,0.9411926,0.0002996905,0.0008455678,2.875638e-7,0.00110976,0.01657825,0.004483202,0.003074203,0.03183115,0.0002306234],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9683639,0.01046919,0.003213324,0.01492237,0.0006599692,0.001757788,0.0001794077,0.00001234164,0.0004216917],"genre_scores_gemma":[0.9927411,0.0002819131,0.002299536,0.003601598,0.0003042959,0.0001318423,0.0005462805,0.000007149438,0.00008626941],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8856674,"threshold_uncertainty_score":0.99643,"prediction_status":"machine_predicted_unvalidated"},"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,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.4999020040566802,"gpt":0.557242604081877,"spread":0.05734060002519675,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2126838,0.00007557413,0.0002565381,0.00005121024,0.0001851837,0.0000242499,0.0001246422,0.00006337833,0.001266322],"category_scores_gemma":[0.05532127,0.00006647151,0.00004842919,0.000420692,0.0001180981,0.0001237076,0.00004613223,0.00007832929,0.000506497],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003582154,"about_ca_system_score_gemma":0.0001242374,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005643748,"about_ca_topic_score_gemma":0.0009725565,"domain_scores_codex":[0.9034745,0.09536491,0.0002547289,0.0002477612,0.0004613085,0.0001967593],"domain_scores_gemma":[0.9877156,0.01172017,0.0000882924,0.0001883379,0.0002198718,0.00006772403],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0001920975,0.00003969223,0.04609106,0.002759877,0.00008684873,0.000005065988,0.008722382,0.000001676,0.00002768082,0.001073529,0.3724712,0.568529],"study_design_scores_gemma":[0.0001699737,0.00001726917,0.9178243,0.001066646,0.00006585312,4.441647e-7,0.00005337705,0.00007351626,0.000007033178,0.005091648,0.0755233,0.0001066973],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"review","genre_scores_codex":[0.5357648,0.3043905,0.001542615,0.1414883,0.003086158,0.006006572,0.00004052882,0.0004488523,0.007231596],"genre_scores_gemma":[0.2702304,0.6147259,0.002797984,0.09205358,0.0007550142,0.001670155,0.0005038597,0.00009050169,0.01717255],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.8717332,"threshold_uncertainty_score":0.9996467,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"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,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"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","retraction":null,"screen_n_in":null,"score":{"opus":0.07795567414794159,"gpt":0.4310312717312078,"spread":0.3530755975832662,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007462905,0.0001883688,0.0005623171,0.00004191091,0.0006530841,0.00007595771,0.0001413838,0.00009724654,0.0002269995],"category_scores_gemma":[0.0002693047,0.0001171022,0.0001501446,0.0002973692,0.0004798901,0.0002090467,0.00001061303,0.0002163897,0.000008429128],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001160112,"about_ca_system_score_gemma":0.0001618615,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.02870184,"about_ca_topic_score_gemma":0.3363519,"domain_scores_codex":[0.9972064,0.001003996,0.0004975985,0.0002612198,0.0008629038,0.0001678972],"domain_scores_gemma":[0.9988496,0.0002031294,0.0004168845,0.0001049016,0.0003771471,0.00004836514],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0002596017,0.0005492279,0.0335981,0.0004467636,0.001026099,0.000007009699,0.6973194,0.00001144292,0.00001896401,0.02355452,0.007796692,0.2354122],"study_design_scores_gemma":[0.0003836998,0.00004776244,0.8121269,0.0007408524,0.0009350288,1.88771e-7,0.1800536,0.00006220776,0.000001827075,0.004355466,0.001097075,0.0001954067],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.956735,0.02202052,1.755548e-7,0.002149258,0.0001851312,0.001513371,0.00001959984,0.00001011816,0.01736684],"genre_scores_gemma":[0.9779975,0.02034815,0.000006882835,0.00005044251,0.000175449,0.0001834255,0.000455128,0.000005419212,0.0007776321],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7785287,"threshold_uncertainty_score":0.9777661,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}