{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":2407,"total_is_capped":false,"direct_labels_cover":84,"predictions_cover":2407,"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":"f27e7d734805","filters":{"topic":"Evaluation and Performance Assessment"}},"results":[{"id":"W2113537253","doi":"10.46743/2160-3715/2003.1870","title":"Understanding Reliability and Validity in Qualitative Research","year":2015,"lang":"en","type":"article","venue":"The Qualitative Report","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":5194,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Reliability (semiconductor); Qualitative research; Validity; Psychology; Triangulation; Perspective (graphical); Test (biology); Positivism; Test validity; Construct validity; Social psychology; Epistemology; Computer science; Sociology; Psychometrics; Mathematics; Artificial intelligence; Social science; Developmental psychology","retraction":null,"screen_n_in":null,"score":{"opus":0.9638247936679258,"gpt":0.7706150627950911,"spread":0.1932097308728348,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3081921,0.0001190018,0.0003004927,0.0002912848,0.0002840724,0.0001795511,0.000361872,0.00006142782,0.0001389538],"category_scores_gemma":[0.05178552,0.00006799261,0.00004829415,0.001444473,0.0008844418,0.0004983063,0.0002056753,0.0004337043,0.0001251255],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005754951,"about_ca_system_score_gemma":0.0007012338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004384541,"about_ca_topic_score_gemma":0.0003634444,"domain_scores_codex":[0.9719276,0.02134613,0.001274848,0.0006241332,0.004455629,0.0003715936],"domain_scores_gemma":[0.9701804,0.02647717,0.0004318794,0.0009023612,0.001813182,0.0001950226],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001547457,0.00007797162,0.01182598,0.000007105491,0.00001055936,0.00002952129,0.8952657,0.0001465633,0.00002141069,0.08257321,0.009509596,0.0003776448],"study_design_scores_gemma":[0.0002222081,0.0001214068,0.004355489,0.000009896335,0.000001817519,0.00001299557,0.482209,0.000694068,0.0000272603,0.5100931,0.002188784,0.00006396449],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9229519,0.00006143944,0.009572461,0.01609521,0.0002163889,0.0004698351,0.000006882644,0.00001982825,0.050606],"genre_scores_gemma":[0.9959808,0.00001113927,0.0008066493,0.0001019201,0.0000396685,0.00004381664,0.000004089605,0.00000674009,0.003005156],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4275199,"threshold_uncertainty_score":0.9562017,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2088547373","doi":"10.1258/1355819054308576","title":"Systematically reviewing qualitative and quantitative evidence to inform management and policy-making in the health field","year":2005,"lang":"en","type":"review","venue":"Journal of Health Services Research & Policy","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1384,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Service Delivery and Organisation Programme; Canadian Health Services Research Foundation","keywords":"Management science; Qualitative research; Narrative; Thematic analysis; Computer science; Systematic review; Psychological intervention; Knowledge management; Data science; Sociology; MEDLINE; Medicine; Political science; Social science; Nursing; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.7490632737818417,"gpt":0.7650935679213311,"spread":0.01603029413948942,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.143679,0.0003160359,0.002344802,0.004315164,0.0006503607,0.000851533,0.001708536,0.0001070412,0.00003540324],"category_scores_gemma":[0.005809135,0.0001672766,0.0001897763,0.005296538,0.00009194561,0.001002008,0.0005587741,0.001251539,0.00005563728],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001220986,"about_ca_system_score_gemma":0.005294756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0012672,"about_ca_topic_score_gemma":0.00104022,"domain_scores_codex":[0.9741814,0.01323876,0.005371395,0.0004623275,0.005772272,0.0009738437],"domain_scores_gemma":[0.9727998,0.02099072,0.003512457,0.0007173757,0.001458162,0.0005214397],"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":[0.00002204947,0.00002600142,0.000009851327,0.03772283,0.00003504739,0.000004282987,0.08474569,0.000004689152,3.498952e-9,0.01339564,0.0007633883,0.8632705],"study_design_scores_gemma":[0.0001738381,0.0015249,0.0003108837,0.1533204,0.00001816687,0.00009258556,0.07202397,0.0001406278,6.300476e-9,0.002799034,0.7694582,0.0001373482],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00005489775,0.9038398,0.0001928522,0.09249664,0.00005351576,0.002408402,0.00000840733,0.00000312797,0.000942338],"genre_scores_gemma":[0.001025162,0.9811282,0.003472327,0.0138236,0.0003067913,0.0000919478,0.000001103722,0.00001494938,0.0001359434],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8631332,"threshold_uncertainty_score":0.9392671,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2889852576","doi":"","title":"Creswell, J. W. (2013). Qualitative inquiry and research design. Choosing among five approaches (3e éd.). London : Sage.","year":2015,"lang":"fr","type":"article","venue":"Approches inductives Travail intellectuel et construction des connaissances","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":963,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"SAGE; Qualitative research; Sociology; Social science; Physics","retraction":null,"screen_n_in":null,"score":{"opus":0.6383697011419519,"gpt":0.5228253934375306,"spread":0.1155443077044213,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.02977441,0.0009168424,0.001157604,0.001517613,0.001592719,0.001912313,0.001104168,0.0007413598,0.001075628],"category_scores_gemma":[0.01117669,0.0008319388,0.0002260822,0.003316165,0.06182592,0.005333652,0.0005238166,0.002180059,0.0002704514],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006222549,"about_ca_system_score_gemma":0.001686795,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006787247,"about_ca_topic_score_gemma":0.0003406993,"domain_scores_codex":[0.9786366,0.01263921,0.001854802,0.002174121,0.003316369,0.001378907],"domain_scores_gemma":[0.9851583,0.009162313,0.0009442307,0.00089995,0.002937301,0.0008979419],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0007474014,0.0003129364,0.002556469,0.0001906502,0.0003624391,0.00001213571,0.4967475,0.001463831,0.0001767748,0.01620141,0.00986647,0.471362],"study_design_scores_gemma":[0.001767978,0.00138361,0.005304899,0.001088924,0.0001288168,0.0003516384,0.8464528,0.0165051,0.00591357,0.09393272,0.02603152,0.001138417],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6649458,0.1171895,0.1978824,0.00570081,0.003767533,0.002154181,0.00004487759,0.0001735228,0.008141285],"genre_scores_gemma":[0.9603761,0.009681649,0.02262721,0.0002295815,0.0006971447,0.0003240045,0.00002045229,0.00008785477,0.005956045],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4702236,"threshold_uncertainty_score":0.9998375,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2131259125","doi":"10.1177/104973230001000308","title":"Evaluating Interpretive Inquiry: Reviewing the Validity Debate and Opening the Dialogue","year":2000,"lang":"en","type":"review","venue":"Qualitative Health Research","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":951,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Foundationalism; Epistemology; Lifeworld; Field (mathematics); External validity; Trustworthiness; Qualitative research; Ontology; Psychology; Sociology; Engineering ethics; Social psychology; Social science; Philosophy","retraction":null,"screen_n_in":null,"score":{"opus":0.975849557890781,"gpt":0.8206803585145628,"spread":0.1551691993762182,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","scholarly_communication","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.3761952,0.0004493676,0.002025991,0.0005715187,0.003394076,0.001171879,0.002372696,0.0001817146,0.001330349],"category_scores_gemma":[0.05324278,0.0001914916,0.0003638086,0.003186913,0.001355579,0.000451912,0.0009575284,0.002948734,0.00108725],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006707384,"about_ca_system_score_gemma":0.005104302,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000270593,"about_ca_topic_score_gemma":0.00007957742,"domain_scores_codex":[0.812084,0.1728331,0.003708076,0.001528043,0.008469529,0.001377227],"domain_scores_gemma":[0.8344231,0.159446,0.001743685,0.001894422,0.002099738,0.0003930367],"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":[0.00001170636,0.00001372523,0.000002776512,0.001473656,0.00003348584,0.000001157283,0.1282162,0.000002860528,2.12312e-8,0.0013349,0.004806342,0.8641031],"study_design_scores_gemma":[0.0001089157,0.0004298811,0.00001925026,0.01183127,0.00004096306,0.0000166865,0.0447877,0.0007096003,8.927292e-8,0.005513892,0.9363356,0.0002061786],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0001105915,0.9834147,0.000304228,0.008291163,0.000393175,0.004452936,0.00007133531,0.000018573,0.002943294],"genre_scores_gemma":[0.0003605435,0.9951425,0.0004701236,0.0009978937,0.0003733441,0.001202245,0.00003701929,0.00004068649,0.001375688],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9315292,"threshold_uncertainty_score":0.999865,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2189153528","doi":"","title":"Framework Analysis: A Qualitative Methodology for Applied Policy Research","year":2009,"lang":"en","type":"article","venue":"SSRN Electronic Journal","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":854,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"MacEwan University","funders":"","keywords":"Qualitative research; Management science; Computer science; Political science; Sociology; Economics; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.6192653351074647,"gpt":0.7019230970789018,"spread":0.08265776197143704,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1263644,0.000135941,0.0004474086,0.002029389,0.0006259507,0.0002576015,0.0009085363,0.0001429944,0.0002563502],"category_scores_gemma":[0.009798103,0.00009551478,0.0003119753,0.004588728,0.000100068,0.00022246,0.000037348,0.002191803,0.0001308712],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000911694,"about_ca_system_score_gemma":0.00619019,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002615325,"about_ca_topic_score_gemma":0.0003793373,"domain_scores_codex":[0.9913538,0.002816112,0.0008199787,0.0004146873,0.002163163,0.002432257],"domain_scores_gemma":[0.9900681,0.007769687,0.0003674177,0.0004368912,0.00121179,0.0001460531],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001600678,0.00003515944,0.00008069808,4.348345e-7,0.0002478764,1.705794e-7,0.006065797,0.0003558452,0.0001399837,0.8545552,0.000341873,0.1380169],"study_design_scores_gemma":[0.0004049453,0.0006590345,0.001106832,0.00000285326,0.00006782715,0.00001210203,0.04798083,0.0008079889,0.00007229988,0.943561,0.005218952,0.0001053659],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03234607,0.0006088752,0.9259866,0.03501193,0.0001043165,0.0003324642,0.000004003226,0.00001867532,0.005587034],"genre_scores_gemma":[0.9705904,0.0004658102,0.02421607,0.0009791125,0.0008421768,0.00003034036,0.000003884313,0.000007981389,0.002864233],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9382443,"threshold_uncertainty_score":0.9994438,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3203243032","doi":"10.1186/s13643-021-01821-3","title":"Scoping reviews: reinforcing and advancing the methodology and application","year":2021,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":834,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"St. Michael's Hospital; Queen's University; McMaster University; Impact; Toronto Rehabilitation Institute; Sunnybrook Health Science Centre; Royal College of Physicians and Surgeons of Canada; Institute for Work & Health; Ottawa Hospital; University of Toronto","funders":"Canada Research Chairs; World Health Organization","keywords":"Systematic review; Rigour; Terminology; CLARITY; Medicine; Presentation (obstetrics); Stakeholder; Management science; Consistency (knowledge bases); Process management; MEDLINE; Computer science; Engineering; Public relations","retraction":null,"screen_n_in":null,"score":{"opus":0.4738319348272704,"gpt":0.5753126436374377,"spread":0.1014807088101674,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05740091,0.0001272912,0.00105218,0.00005822624,0.0002362656,0.0001996219,0.0002099978,0.0000362438,0.0001093189],"category_scores_gemma":[0.02069556,0.00005796809,0.0000848458,0.000450271,0.00004309059,0.000255783,0.0001147492,0.00009086326,0.000221675],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002343036,"about_ca_system_score_gemma":0.00006986187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004165572,"about_ca_topic_score_gemma":0.00005519073,"domain_scores_codex":[0.9917234,0.00523393,0.002034813,0.000344172,0.0005138884,0.0001497763],"domain_scores_gemma":[0.9934841,0.004344857,0.001088294,0.0008179675,0.0002042875,0.00006050828],"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":[0.000005365697,0.00001821304,0.003147971,0.1993988,0.00004895764,0.000002977847,0.008452037,0.0000954654,0.00278329,0.02744794,0.005726162,0.7528728],"study_design_scores_gemma":[0.0009814993,0.0001470338,0.001776973,0.1458226,0.0004964185,0.0008609022,0.01356913,0.05977868,0.0007375722,0.02719965,0.7476785,0.0009511411],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.003135967,0.3539882,0.620078,0.003788632,0.0004948018,0.01284989,7.712685e-7,0.00002632482,0.005637373],"genre_scores_gemma":[0.1492342,0.4436778,0.3532421,0.02273836,0.0009702218,0.01568797,0.0000231557,0.00008980794,0.01433629],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7519217,"threshold_uncertainty_score":0.9875535,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2124445109","doi":"10.26522/brocked.v12i2.38","title":"The Use of Think-aloud Methods in Qualitative Research An Introduction to Think-aloud Methods","year":2003,"lang":"en","type":"article","venue":"Brock Education Journal","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":755,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Think aloud protocol; Qualitative research; Interpretation (philosophy); Reciprocity (cultural anthropology); Task (project management); Psychology; Reading aloud; Participant observation; Qualitative property; Triangulation; Process (computing); Computer science; Social psychology; Human–computer interaction; Linguistics; Sociology; Reading (process); Mathematics; Engineering; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.6737008481614124,"gpt":0.7246427914241172,"spread":0.05094194326270485,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0843245,0.0001555936,0.000300538,0.00104594,0.0009042874,0.0008765942,0.000781614,0.00009896222,0.001211477],"category_scores_gemma":[0.06162749,0.00009916981,0.0001034973,0.002807675,0.0001351208,0.001453366,0.00006932524,0.0008820886,0.00008394738],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003257608,"about_ca_system_score_gemma":0.003042244,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005197736,"about_ca_topic_score_gemma":0.0001240535,"domain_scores_codex":[0.9591155,0.03538527,0.001766222,0.000520398,0.002743136,0.0004694718],"domain_scores_gemma":[0.9836898,0.009815085,0.0006854397,0.0009203319,0.004486836,0.0004025509],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001339524,0.0003967136,0.000638766,0.000005846054,0.00002894188,2.587471e-7,0.2860321,0.003286086,0.005484136,0.03123828,0.1503981,0.5223569],"study_design_scores_gemma":[0.0003007249,0.0003297446,0.009680836,0.00002708523,0.00001029319,0.00005736183,0.5165219,0.001215086,0.003987325,0.1467645,0.3209326,0.0001726322],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.6080451,0.0006264746,0.3324538,0.04815754,0.008068267,0.001009897,0.000004380924,0.00002312729,0.001611424],"genre_scores_gemma":[0.04158712,0.0001657322,0.9433522,0.0008075286,0.00128098,0.00001756862,0.000004555573,0.0000234294,0.01276085],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6108984,"threshold_uncertainty_score":0.9997016,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W27308458","doi":"10.46743/2160-3715/2007.1645","title":"On Becoming a Qualitative Researcher: The Value of Reflexivity","year":2015,"lang":"en","type":"article","venue":"The Qualitative Report","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":600,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Reflexivity; Qualitative research; Value (mathematics); Phenomenon; Epistemology; Narrative; Process (computing); Field (mathematics); Task (project management); Sociology; Engineering ethics; Psychology; Computer science; Social science; Management","retraction":null,"screen_n_in":null,"score":{"opus":0.8577058564611163,"gpt":0.7533929997569319,"spread":0.1043128567041844,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1583719,0.0001453919,0.0003410934,0.0001406072,0.0002858688,0.00009284744,0.000867712,0.0000484711,0.0004256122],"category_scores_gemma":[0.04114017,0.00006280229,0.0001490899,0.0009077045,0.0007221018,0.0002649611,0.000186502,0.000349973,0.0002999714],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001403049,"about_ca_system_score_gemma":0.000818942,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002826963,"about_ca_topic_score_gemma":0.0001303268,"domain_scores_codex":[0.9800298,0.01320838,0.00128175,0.0003956799,0.004813217,0.000271215],"domain_scores_gemma":[0.9688401,0.02655005,0.00112723,0.001341426,0.002012302,0.0001288461],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002983671,0.0001079782,0.0001673635,0.000005303441,0.00006343748,0.00001269612,0.7163416,0.0002682975,0.000162214,0.2596619,0.02050715,0.002403654],"study_design_scores_gemma":[0.0003115668,0.0004169391,0.001905798,0.00002107744,0.000009306118,0.00001844057,0.4572823,0.0008735577,0.0009206796,0.5290682,0.009077872,0.00009427477],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8832726,0.00007323032,0.004030298,0.02279365,0.0002587237,0.0004821563,0.00002472486,0.0000210571,0.08904348],"genre_scores_gemma":[0.9817423,0.000003296546,0.0008657159,0.0005722548,0.0000570078,0.00005293678,0.000004967465,0.000009749294,0.01669178],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2694062,"threshold_uncertainty_score":0.9669367,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2108155649","doi":"10.1177/1049732309344612","title":"Strengths and Challenges in the Use of Interpretive Description: Reflections Arising From a Study of the Moral Experience of Health Professionals in Humanitarian Work","year":2009,"lang":"en","type":"article","venue":"Qualitative Health Research","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":533,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Interpretation (philosophy); Qualitative research; Engineering ethics; Context (archaeology); Epistemology; Strengths and weaknesses; Discipline; Sociology; Management science; Psychology; Social psychology; Social science; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.9440937988956182,"gpt":0.7470193126754651,"spread":0.1970744862201531,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0329422,0.00008190715,0.0003738023,0.0004659493,0.0002958218,0.0000398802,0.0004314124,0.00003621757,0.00003203442],"category_scores_gemma":[0.007472688,0.00004540645,0.00002878372,0.001852337,0.0002966588,0.000338556,0.0001004254,0.000482352,7.862784e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001637752,"about_ca_system_score_gemma":0.0007497794,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004208372,"about_ca_topic_score_gemma":0.01425924,"domain_scores_codex":[0.9740418,0.02140572,0.001454414,0.0003464308,0.00246538,0.0002862909],"domain_scores_gemma":[0.9842762,0.01400948,0.0005677862,0.0004511223,0.0006468113,0.00004860111],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001175134,0.001577457,0.01201196,0.00001925361,0.000004015115,1.759126e-7,0.9539602,0.00002555837,0.00001010134,0.004118193,0.000121438,0.02803408],"study_design_scores_gemma":[0.0003585732,0.0009436284,0.384531,0.0003644975,3.818982e-7,9.02968e-8,0.6066943,0.0001235954,0.00000699029,0.006877849,0.00007592575,0.00002310489],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9721007,0.00126848,0.00006435574,0.02522451,0.00005691721,0.001151187,0.00001408443,0.000001808018,0.000117988],"genre_scores_gemma":[0.9986405,0.0003472842,0.0006038705,0.0002988122,0.000008870548,0.00006822425,0.000001110877,0.00000274471,0.00002854947],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.372519,"threshold_uncertainty_score":0.9957895,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2068298920","doi":"10.1111/j.1754-7121.2009.00070_1.x","title":"Policy analytical capacity and evidence‐based policy‐making: Lessons from Canada","year":2009,"lang":"en","type":"article","venue":"Canadian Public Administration","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":524,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Policy making; Political science; Government (linguistics); Order (exchange); Public administration; Welfare economics; Policy analysis; Humanities; Economics; Philosophy; Finance","retraction":null,"screen_n_in":null,"score":{"opus":0.3480418204072303,"gpt":0.4978244758265805,"spread":0.1497826554193502,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001169275,0.0001620285,0.0002092981,0.0006587813,0.0003923007,0.000840639,0.0004144879,0.0001111343,0.0006022605],"category_scores_gemma":[0.006447338,0.0001453782,0.00004432704,0.001265454,0.00009774936,0.000843627,0.00001296596,0.000168163,0.00002219487],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001122935,"about_ca_system_score_gemma":0.05825957,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.7075905,"about_ca_topic_score_gemma":0.9965984,"domain_scores_codex":[0.9971567,0.0001960172,0.0005102351,0.0004591687,0.001231391,0.0004464469],"domain_scores_gemma":[0.9973442,0.0005056598,0.0001839317,0.0005069042,0.0004414338,0.00101786],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.00002159462,0.00003921888,0.03782581,0.00000387823,0.00001700157,0.00002594114,0.0001388863,0.000032759,0.00006368592,0.8423034,0.02028378,0.09924406],"study_design_scores_gemma":[0.0004362662,0.0002413926,0.7013828,0.00005294602,0.00001908792,0.00001663164,0.0004065396,0.05523138,0.000157273,0.03090543,0.210704,0.0004461766],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2408331,0.00003880498,0.0007269546,0.7516578,0.0001247319,0.0001661274,0.000118506,0.00001876487,0.006315254],"genre_scores_gemma":[0.9765114,0.000005603116,0.0003203772,0.02232193,0.0003523451,0.000006738623,0.00004341569,0.00000533535,0.0004328226],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.811398,"threshold_uncertainty_score":0.9470792,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2002747675","doi":"10.1177/1075547004267491","title":"New Evidence on Instrumental, Conceptual, and Symbolic Utilization of University Research in Government Agencies","year":2004,"lang":"en","type":"article","venue":"Science Communication","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":465,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval","funders":"","keywords":"Government (linguistics); The Symbolic; Conceptual framework; Sociology; Public relations; Political science; Psychology; Social science; Linguistics","retraction":null,"screen_n_in":null,"score":{"opus":0.6781789006118617,"gpt":0.5680393100705534,"spread":0.1101395905413083,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01118702,0.00005540158,0.00009867051,0.0003361324,0.0003686054,0.0001023855,0.001229603,0.00002823711,0.00008949516],"category_scores_gemma":[0.001457223,0.00004779472,0.0000132821,0.002580762,0.001183931,0.001281356,0.0004153363,0.0001383764,0.00003162616],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006810701,"about_ca_system_score_gemma":0.0009360348,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003972214,"about_ca_topic_score_gemma":0.0003952194,"domain_scores_codex":[0.9961839,0.0004255126,0.0002861025,0.0002664355,0.002679866,0.0001581627],"domain_scores_gemma":[0.9978806,0.0007920936,0.0001523084,0.0007859009,0.0003113687,0.00007771138],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0002074038,0.0002477929,0.1384046,0.00001268469,0.000004219347,0.00000103963,0.03409092,0.004315949,0.02990657,0.5459726,0.0004852541,0.246351],"study_design_scores_gemma":[0.001107127,0.00050294,0.844837,0.000521945,0.000003837556,0.000002128731,0.08191963,0.004190298,0.03499659,0.02920696,0.002532356,0.0001792106],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9892183,0.0002300102,0.0004445141,0.002483821,0.00004201645,0.0002146893,0.000001698296,0.000006382967,0.007358568],"genre_scores_gemma":[0.9967055,0.001373178,0.001680658,0.00006049419,0.000003678643,0.000001551055,7.606683e-7,0.000001612592,0.0001725269],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7064324,"threshold_uncertainty_score":0.4362243,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1605140183","doi":"10.21225/d53s33","title":"How to Conduct Surveys: A Step-by-Step Guide","year":2010,"lang":"en","type":"article","venue":"Canadian Journal of University Continuing Education","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":405,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Psychology; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.07307313749290577,"gpt":0.37441708903417,"spread":0.3013439515412643,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005651499,0.00008117913,0.0001642551,0.0007123547,0.0001647101,0.000402001,0.0006306921,0.00006219152,0.0006184204],"category_scores_gemma":[0.002299651,0.00007607167,0.00007218146,0.0005463226,0.00005347948,0.0007977762,0.00001450375,0.0002386833,0.00004930961],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002277944,"about_ca_system_score_gemma":0.003795699,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004118046,"about_ca_topic_score_gemma":0.1102982,"domain_scores_codex":[0.9985127,0.0003717488,0.000272671,0.0001535771,0.0005033528,0.0001859197],"domain_scores_gemma":[0.9964969,0.0002809858,0.0003695484,0.0002565139,0.001882897,0.0007131272],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000007070172,0.000017679,0.04211196,0.000001396054,0.00001063523,0.000005974947,0.001270482,0.000007990356,0.00265996,0.0007853992,0.7369218,0.2161996],"study_design_scores_gemma":[0.0002130427,0.00004953849,0.04527151,0.00002312001,0.00001196724,0.00002029818,0.01365456,0.00006082483,0.0001737329,0.00003582112,0.9403935,0.00009207797],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9581265,0.00004195987,0.01295038,0.0173752,0.003198673,0.000131219,0.00002067372,0.000004591654,0.008150836],"genre_scores_gemma":[0.9500477,0.000003327433,0.005409362,0.000345706,0.0001753256,1.416714e-7,0.000004540051,0.000004774523,0.04400917],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2161075,"threshold_uncertainty_score":0.9059365,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2112238768","doi":"10.1136/jech.2004.031765","title":"Can scientists and policy makers work together?: Table 1","year":2005,"lang":"en","type":"review","venue":"Journal of Epidemiology & Community Health","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":395,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Government of Canada; University of Toronto; Public Health Agency of Canada; University of Ottawa","funders":"","keywords":"Incentive; Work (physics); General partnership; Accountability; Public relations; Science policy; Engineering ethics; Perception; Medicine; Political science; Public administration; Epistemology; Economics; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.630728599622718,"gpt":0.6526695250093589,"spread":0.02194092538664094,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.134772,0.000375773,0.004180781,0.001443853,0.001015546,0.00009662248,0.001956533,0.0003748578,0.0003719518],"category_scores_gemma":[0.03274496,0.0002372595,0.0005096193,0.001870926,0.00050308,0.0002605921,0.0004206594,0.00340531,0.00006357951],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001132912,"about_ca_system_score_gemma":0.009672734,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00162788,"about_ca_topic_score_gemma":0.003386697,"domain_scores_codex":[0.9676666,0.02523799,0.005040588,0.0003512892,0.0009384259,0.0007650607],"domain_scores_gemma":[0.9624819,0.02744063,0.007101494,0.001539771,0.0006698756,0.0007662968],"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":[0.000009434469,0.00006767316,0.002672271,0.0003770856,0.00005694751,0.000001091838,0.000391131,0.00003768023,5.455507e-9,0.001076312,0.03506098,0.9602494],"study_design_scores_gemma":[0.0002431821,0.0003179052,0.002753903,0.003267369,0.00006263758,0.0002679755,0.0003375096,0.00001839858,5.709694e-9,0.005021024,0.9875559,0.0001542052],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002174937,0.9734762,0.0002769264,0.0233729,0.0007387222,0.0003479804,0.0000372219,0.000009099183,0.001523476],"genre_scores_gemma":[0.0002070764,0.9850167,0.005263754,0.008053036,0.0005540162,0.000006660608,0.0000175128,0.00002095016,0.0008602933],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9600952,"threshold_uncertainty_score":0.9988939,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1603616381","doi":"10.1177/160940690700600208","title":"Redefining Case Study","year":2007,"lang":"en","type":"article","venue":"International Journal of Qualitative Methods","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":371,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Case study research; Relation (database); Computer science; Case analysis; Research design; Management science; Sociology; Knowledge management; Engineering; Artificial intelligence; Data mining; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.9205268197323564,"gpt":0.8096859168208992,"spread":0.1108409029114572,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1627998,0.0001023799,0.0002828625,0.0008060886,0.00008828375,0.0001859265,0.0007079623,0.00003275428,0.0009113252],"category_scores_gemma":[0.01338519,0.00006895462,0.0001643524,0.0004471365,0.00007326257,0.0005677063,0.00008943291,0.0002656639,0.00004039308],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001297049,"about_ca_system_score_gemma":0.0001846068,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004470091,"about_ca_topic_score_gemma":0.0000544914,"domain_scores_codex":[0.9887435,0.005564181,0.00203178,0.0001940129,0.003308058,0.0001584036],"domain_scores_gemma":[0.9806876,0.01353604,0.001378247,0.0001745722,0.004106742,0.0001168245],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002731669,0.0004121674,0.01169664,8.458032e-7,0.00031737,0.003129658,0.2575686,0.0002207756,0.0008266115,0.006370225,0.001014292,0.7181697],"study_design_scores_gemma":[0.00174634,0.00086834,0.01040217,0.00002300894,0.00003196187,0.004696809,0.9159033,0.0006379128,0.001355286,0.04498458,0.01918657,0.000163753],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4999714,0.00005634335,0.4923038,0.0008563726,0.001734601,0.00005938899,0.000001952948,0.000003961352,0.005012206],"genre_scores_gemma":[0.7305536,0.000002727368,0.2684369,0.0002910479,0.0002461079,0.0000010397,2.790733e-7,0.000004868859,0.0004633551],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.718006,"threshold_uncertainty_score":0.9978371,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2113515522","doi":"10.3138/cjpe.016.001","title":"Addressing Attribution through Contribution Analysis: Using Performance Measures Sensibly","year":2001,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":342,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"Government of Canada","funders":"","keywords":"Attribution; Accountability; Authorship attribution; Public relations; Administration (probate law); Psychology; Business; Political science; Social psychology; Computer science; Law; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.6796845234554371,"gpt":0.5702891541640307,"spread":0.1093953692914064,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01855108,0.0001717153,0.0004181861,0.001231228,0.0006017481,0.000654636,0.0003546966,0.0001321038,0.001019635],"category_scores_gemma":[0.00228243,0.0001382279,0.0002745675,0.003573975,0.0001165234,0.001845184,0.0000119094,0.0002630332,0.00004416694],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009571845,"about_ca_system_score_gemma":0.004131797,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005635114,"about_ca_topic_score_gemma":0.0101921,"domain_scores_codex":[0.993841,0.0008973504,0.001319936,0.0002705077,0.003234984,0.0004362063],"domain_scores_gemma":[0.9895816,0.0001514445,0.001232262,0.0003376986,0.008321461,0.000375514],"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":[0.00005729624,0.0000558953,0.2139138,0.000003362315,0.0002584292,0.00001239033,0.0008444975,0.1098887,0.0003892263,0.0002682935,0.0003693116,0.6739388],"study_design_scores_gemma":[0.001325101,0.0004245279,0.2091643,0.00009822765,0.001331502,0.0001780687,0.0008806009,0.7357028,0.0007337685,0.00230982,0.04758309,0.0002682058],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8888697,0.0008263701,0.1075659,0.0007287594,0.0006935711,0.0006666715,0.000008013981,0.00001363579,0.0006273757],"genre_scores_gemma":[0.9956023,0.00004838652,0.003818422,0.0001350753,0.0002826333,0.00001613176,0.00004216899,0.000008724095,0.00004621566],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6736705,"threshold_uncertainty_score":0.9998935,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1904300077","doi":"10.1186/s13012-015-0345-7","title":"The concept of mechanism from a realist approach: a scoping review to facilitate its operationalization in public health program evaluation","year":2015,"lang":"en","type":"review","venue":"Implementation Science","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":312,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; Montreal Clinical Research Institute; Université de Montréal","funders":"Institut National Du Cancer; Agence Nationale de la Recherche","keywords":"Operationalization; Referent; CLARITY; Mechanism (biology); Epistemology; Public health; Context (archaeology); Health informatics; Psychological intervention; Sociology; Medicine; Nursing","retraction":null,"screen_n_in":null,"score":{"opus":0.9035080940398004,"gpt":0.7158287156651659,"spread":0.1876793783746344,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0650932,0.0002928807,0.001131461,0.0007524231,0.0005386344,0.0008836623,0.001659941,0.00005908293,0.0005547369],"category_scores_gemma":[0.005530053,0.0001870172,0.0001404076,0.006375848,0.000204605,0.001419562,0.0002789915,0.0001535727,0.00008829177],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001069663,"about_ca_system_score_gemma":0.0184105,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002282597,"about_ca_topic_score_gemma":0.001279329,"domain_scores_codex":[0.9834861,0.002991745,0.003690773,0.001071571,0.0082205,0.000539318],"domain_scores_gemma":[0.9918543,0.0008985871,0.002187899,0.0008591028,0.003900066,0.0003000628],"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":[0.000001071313,0.00003502605,0.000004690248,0.0007991921,0.000006072869,4.643735e-8,0.001454129,0.00006675292,5.359228e-7,0.004920613,0.000918358,0.9917935],"study_design_scores_gemma":[0.0008678273,0.0003362167,0.0001035021,0.03021211,0.00009246726,0.000004319717,0.004969284,0.01240049,0.000006906341,0.002477335,0.9480284,0.0005010712],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000008849111,0.9618493,0.01424151,0.002299013,0.0003904198,0.02033047,0.0003111715,0.00002668543,0.0005426367],"genre_scores_gemma":[0.000348632,0.9879692,0.003888797,0.0007858086,0.00003901722,0.005810968,0.00107671,0.000014786,0.00006612403],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9912924,"threshold_uncertainty_score":0.9871542,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2212762088","doi":"10.1186/s13643-015-0163-7","title":"All in the Family: systematic reviews, rapid reviews, scoping reviews, realist reviews, and more","year":2015,"lang":"en","type":"editorial","venue":"Systematic Reviews","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":303,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Medicine; Systematic review; MEDLINE; Engineering ethics","retraction":null,"screen_n_in":null,"score":{"opus":0.5113523182310088,"gpt":0.5556523779986046,"spread":0.04430005976759577,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow"],"category_scores_codex":[0.4398393,0.002827304,0.02832102,0.001079043,0.0005313233,0.002150909,0.006736779,0.001256099,0.0004378587],"category_scores_gemma":[0.2676933,0.001284166,0.00275351,0.003704539,0.0003338722,0.00133134,0.0007070469,0.002438626,0.01236759],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009389229,"about_ca_system_score_gemma":0.001123142,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001535497,"about_ca_topic_score_gemma":0.0004097344,"domain_scores_codex":[0.8271449,0.1115201,0.04358127,0.00390079,0.01197431,0.001878621],"domain_scores_gemma":[0.9264865,0.02072586,0.03624544,0.01346698,0.002157298,0.0009179615],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000003709925,0.00004164976,0.000004420238,0.4551576,0.00002994467,0.000006238925,0.001345762,3.359481e-7,8.271194e-7,0.00005325998,0.5407882,0.002568061],"study_design_scores_gemma":[0.0003621221,0.00008031388,0.000002769934,0.4437167,0.0008377807,0.00003427639,0.0004266732,0.00007718839,6.695248e-8,0.0001722463,0.5535523,0.0007375491],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000001090059,0.7436713,0.0002756031,0.0007508305,0.152302,0.09976178,0.00005890639,0.00004649392,0.003132067],"genre_scores_gemma":[0.000002388693,0.8733447,0.001470112,0.003628265,0.07945374,0.03706316,0.0004187139,0.0001297322,0.004489147],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.172146,"threshold_uncertainty_score":0.9998628,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2142374214","doi":"10.1177/1075547001022004003","title":"Climbing the Ladder of Research Utilization","year":2001,"lang":"en","type":"article","venue":"Science Communication","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":269,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval","funders":"","keywords":"Climbing; Climb; Transmission (telecommunications); Stage (stratigraphy); Knowledge management; Computer science; Engineering; Biology; Telecommunications; Ecology","retraction":null,"screen_n_in":null,"score":{"opus":0.8407600227914698,"gpt":0.7017700442979139,"spread":0.1389899784935559,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04268249,0.00004096626,0.0000714237,0.0003821314,0.000967512,0.0002577929,0.002718281,0.00002324063,0.0005794945],"category_scores_gemma":[0.004292473,0.00002395165,0.00002558507,0.00508849,0.001452191,0.0008660209,0.0004285129,0.0001687536,0.0001913507],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005913707,"about_ca_system_score_gemma":0.0003865595,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004929955,"about_ca_topic_score_gemma":0.0001090286,"domain_scores_codex":[0.9953181,0.0007044934,0.0004167112,0.0001970386,0.003169695,0.000194012],"domain_scores_gemma":[0.9943008,0.001497952,0.0001924379,0.001755188,0.002212369,0.00004127591],"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":[0.00003485223,0.000204492,0.1722204,0.000004091695,0.000004512776,1.917599e-7,0.01592463,0.001877339,0.01246776,0.2709869,0.004628242,0.5216466],"study_design_scores_gemma":[0.0003056232,0.0001041343,0.5898822,0.00005315683,0.000004393468,0.000006695308,0.02146578,0.1388094,0.007108928,0.1239422,0.1181913,0.0001261961],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8391384,0.0003607298,0.002554921,0.01440329,0.0001766965,0.0003287519,8.165499e-7,0.00002293843,0.1430135],"genre_scores_gemma":[0.9977428,0.0004134464,0.0006848898,0.0001387134,0.0000139667,0.00001377378,0.000001810501,0.000002173257,0.0009883663],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5215204,"threshold_uncertainty_score":0.9857599,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1524663891","doi":"","title":"Evaluation in a nutshell: a practical guide to the evaluation of health promotion programs","year":2013,"lang":"en","type":"book","venue":"ePrints Soton (University of Southampton)","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":244,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Health promotion; Public relations; Promotion (chess); Accountability; Medical education; Computer science; Political science; Public health; Management science; Medicine; Engineering; Nursing","retraction":null,"screen_n_in":null,"score":{"opus":0.3758254524828953,"gpt":0.49534675205837,"spread":0.1195212995754747,"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":[0.05913281,0.0002456042,0.0006432126,0.0007320815,0.0001911902,0.00008898287,0.0009038294,0.0002538182,0.009877515],"category_scores_gemma":[0.002185979,0.0002080888,0.0002169498,0.0007358909,0.0001671001,0.0006902413,0.0003416757,0.0004101628,0.00135864],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001634599,"about_ca_system_score_gemma":0.006782887,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007937238,"about_ca_topic_score_gemma":0.002720998,"domain_scores_codex":[0.9854851,0.00272106,0.001174141,0.0007817469,0.009519723,0.0003182598],"domain_scores_gemma":[0.9916914,0.0003663488,0.001907443,0.001110952,0.004787486,0.0001363774],"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":[0.00006673646,0.0006064242,0.0029774,0.00006940441,0.00009946992,8.198934e-7,0.01626113,0.002388989,0.00001723466,0.001838463,0.03359767,0.9420763],"study_design_scores_gemma":[0.007316909,0.002007474,0.1465573,0.002095594,0.0008766048,0.00001395705,0.0435232,0.3186265,0.00005929978,0.07926556,0.3984602,0.001197365],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.3259217,0.0003777985,0.04744767,0.04816596,0.001739237,0.03392768,0.000115601,0.0001140452,0.5421904],"genre_scores_gemma":[0.5961699,0.0001009401,0.009828543,0.0002162496,0.0001501178,0.0001007023,0.0003288666,0.00005164329,0.393053],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9408789,"threshold_uncertainty_score":0.9994189,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3110959118","doi":"10.1093/jeea/jvaa019","title":"How Much Can We Generalize From Impact Evaluations?","year":2020,"lang":"en","type":"article","venue":"Journal of the European Economic Association","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":232,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Exploit; Set (abstract data type); Sample (material); Government (linguistics); Sample size determination; Econometrics; Computer science; Treatment effect; Economics; Statistics; Mathematics; Computer security","retraction":null,"screen_n_in":null,"score":{"opus":0.1830321189311144,"gpt":0.4270938186999413,"spread":0.2440616997688269,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007320421,0.00010399,0.0002446826,0.00007574658,0.0001531224,0.0005864402,0.0007784068,0.00003212687,0.0008422736],"category_scores_gemma":[0.002802406,0.00006228925,0.0003263739,0.0001664326,0.00001355559,0.0006021761,0.00009173233,0.0002191638,0.0004135106],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009323934,"about_ca_system_score_gemma":0.0003907502,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001555075,"about_ca_topic_score_gemma":0.00004178996,"domain_scores_codex":[0.9962759,0.001696399,0.0007943034,0.0001561187,0.0009417768,0.0001354861],"domain_scores_gemma":[0.9961337,0.0005010912,0.002630359,0.0002393436,0.0003706891,0.000124764],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005804953,0.00004741075,0.4746076,0.000001661335,0.0003965255,0.000002774077,0.006470362,0.03717791,0.003145216,0.000151129,0.3823286,0.09561273],"study_design_scores_gemma":[0.001640077,0.0001571951,0.8663104,0.00001937289,0.0001161455,0.00000512399,0.001518112,0.03936847,0.0007657313,0.00882196,0.08107654,0.0002008946],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8437785,0.00009874553,0.000679066,0.1496118,0.001249911,0.0001002451,0.00005649829,0.000007750177,0.004417547],"genre_scores_gemma":[0.9944611,0.00007317227,0.0002883862,0.001463131,0.00156545,4.468501e-7,0.000003498725,0.00001258333,0.002132191],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3917027,"threshold_uncertainty_score":0.9222306,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1486190448","doi":"10.1111/medu.12091","title":"Rethinking programme evaluation in health professions education: beyond ‘did it work?’","year":2013,"lang":"en","type":"article","venue":"Medical Education","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":217,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Holland Bloorview Kids Rehabilitation Hospital; The Wilson Centre; London Health Sciences Centre; Centre Hospitalier Universitaire Sainte-Justine; SickKids Foundation; University of Toronto","funders":"","keywords":"Parallels; Context (archaeology); Engineering ethics; Curriculum; Field (mathematics); Process (computing); Work (physics); Health professions; Sociology; Program evaluation; Medical education; Management science; Health care; Psychology; Pedagogy; Political science; Medicine; Computer science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.2344919811446988,"gpt":0.5649593503657884,"spread":0.3304673692210895,"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":[0.0197132,0.0001492923,0.0002461922,0.0004800796,0.0002740421,0.0002666949,0.000561383,0.000173115,0.02777538],"category_scores_gemma":[0.01327927,0.0001115862,0.00005725619,0.001874581,0.00009005772,0.0007750383,0.00007520419,0.0004627018,0.001484708],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005610498,"about_ca_system_score_gemma":0.02447969,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005847532,"about_ca_topic_score_gemma":0.0006480052,"domain_scores_codex":[0.9917692,0.001235487,0.001319934,0.0005079057,0.004767813,0.0003996373],"domain_scores_gemma":[0.9966258,0.0005502746,0.0004994622,0.0005935089,0.001342724,0.0003882717],"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":[0.000003358183,0.000600027,0.02735099,0.00001246583,0.000002066635,3.81092e-8,0.006203607,0.00001397351,0.000001740782,0.001750415,0.1787212,0.7853401],"study_design_scores_gemma":[0.000949665,0.0001507055,0.406803,0.001011632,0.0000126931,0.000009404574,0.02274506,0.02668127,0.00001304838,0.2428915,0.2983561,0.0003759445],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.3264521,0.0008345829,0.0008498525,0.6307736,0.00554372,0.003072134,6.404841e-7,0.00006353504,0.03240988],"genre_scores_gemma":[0.9565141,0.00006178032,0.003574868,0.0306138,0.0004396267,0.001866368,0.0001139846,0.00001279787,0.006802675],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7849641,"threshold_uncertainty_score":0.9992927,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4230488502","doi":"10.3138/cjpe.230","title":"Useful Theory of Change Models","year":2015,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":213,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Causality (physics); Theory of change; Development theory; Intervention (counseling); Epistemology; Computer science; Psychological intervention; Management science; Psychology; Sociology; Economics; Philosophy; Physics","retraction":null,"screen_n_in":null,"score":{"opus":0.8815743086593234,"gpt":0.5830366120430426,"spread":0.2985376966162808,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02951679,0.00008291962,0.0002148965,0.000645357,0.00006086782,0.0001321304,0.0004306173,0.00006207573,0.0009688999],"category_scores_gemma":[0.002523195,0.00006067476,0.0001000823,0.0006161088,0.00008981459,0.001054459,0.00001074054,0.0001199315,0.00004725626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002436918,"about_ca_system_score_gemma":0.005279141,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003148608,"about_ca_topic_score_gemma":0.005280161,"domain_scores_codex":[0.995709,0.000712624,0.0008087651,0.0001212273,0.002461384,0.0001870183],"domain_scores_gemma":[0.9930302,0.0001720576,0.0007556617,0.0002630102,0.005218913,0.0005601838],"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":[0.00003011145,0.0000372332,0.01456073,0.000002721423,0.00002123891,0.000002753057,0.00536727,0.004864279,0.000006562896,0.005610535,0.002937779,0.9665588],"study_design_scores_gemma":[0.002441526,0.001567708,0.06029489,0.00009861121,0.0001692313,0.0000638531,0.008755816,0.2816721,0.0002045228,0.5905778,0.05390577,0.0002480593],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9596446,0.002650479,0.00972136,0.003488117,0.002291091,0.001916573,0.00001772277,0.00001230612,0.02025782],"genre_scores_gemma":[0.9969894,0.000007206828,0.002442915,0.0001788082,0.0001888478,0.00003806376,0.0000045625,0.000006667958,0.0001434966],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9663107,"threshold_uncertainty_score":0.9999443,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2143645386","doi":"","title":"How to Change 5000 Schools: A Practical and Positive Approach for Leading Change at Every Level","year":2008,"lang":"en","type":"book","venue":"","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":191,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Government (linguistics); Embodied cognition; Public administration; Political science; Economic growth; Public education; Public relations; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.7024683524713844,"gpt":0.5225452151253678,"spread":0.1799231373460166,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002698859,0.0003979882,0.0006796927,0.0005848215,0.0004127714,0.0005798354,0.0003843614,0.0003815199,0.0003288017],"category_scores_gemma":[0.001455094,0.0002899237,0.0001820452,0.0002603558,0.00008286752,0.001409788,0.0003824795,0.0003674784,0.0004577431],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005109371,"about_ca_system_score_gemma":0.00069513,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001256588,"about_ca_topic_score_gemma":0.0001028872,"domain_scores_codex":[0.9956311,0.0001322564,0.0005226922,0.001055024,0.002200712,0.0004582235],"domain_scores_gemma":[0.9965678,0.001205794,0.0003756633,0.0005391589,0.0009515804,0.000360006],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001807828,0.00007542883,0.0002689314,0.00004560073,0.00007359911,0.000009396464,0.002823205,0.00000138563,0.00001323034,0.007745384,0.9180648,0.07069831],"study_design_scores_gemma":[0.0008033417,0.0005657942,0.00320624,0.0001244536,0.00007305833,0.000136375,0.001175664,0.005148196,0.00005905103,0.001687445,0.9862731,0.0007472981],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0004211693,0.00135504,0.1081515,0.08341528,0.001098671,0.01051607,0.001142539,0.0001525788,0.7937471],"genre_scores_gemma":[0.0009883288,0.0003162985,0.04927365,0.009161486,0.001242161,0.001016544,0.0002203739,0.00004471114,0.9377365],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.1439894,"threshold_uncertainty_score":0.9999553,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2093491439","doi":"10.1037/1082-989x.7.1.126","title":"The role of qualitative research in psychological journals.","year":2002,"lang":"en","type":"article","venue":"Psychological Methods","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":183,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Windsor","funders":"","keywords":"PsycINFO; Qualitative research; Qualitative analysis; Psychology; Content analysis; Applied psychology; Social psychology; MEDLINE; Social science; Sociology","retraction":null,"screen_n_in":null,"score":{"opus":0.9375029604873701,"gpt":0.8324841901757505,"spread":0.1050187703116197,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1253068,0.0001189281,0.0003481349,0.0003604667,0.0002803812,0.0001670491,0.001376161,0.0001720436,0.007345484],"category_scores_gemma":[0.01519728,0.00005497122,0.0001408868,0.002592566,0.0006824816,0.0001605042,0.0001359368,0.0009031989,0.0005211517],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003620322,"about_ca_system_score_gemma":0.000008418303,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004991409,"about_ca_topic_score_gemma":0.000005368148,"domain_scores_codex":[0.9704429,0.02457314,0.00135745,0.0005819649,0.002533609,0.0005109739],"domain_scores_gemma":[0.9730021,0.0252032,0.0002902099,0.0007487409,0.0006247978,0.0001309902],"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":[0.0000738244,0.0004253943,0.003294932,5.34131e-7,0.000005033755,0.000002172525,0.003499191,0.00001271307,0.00132133,0.02033557,0.002358517,0.9686708],"study_design_scores_gemma":[0.0004234338,0.0005919901,0.09659927,0.00001283433,0.000001608849,0.000010714,0.01230831,0.002639046,0.0002737186,0.8330836,0.0539583,0.00009722701],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.50292,0.007284496,0.01766171,0.02131618,0.0008823148,0.0006068209,0.000005238865,0.00003384536,0.4492894],"genre_scores_gemma":[0.9168964,0.0007487345,0.08095983,0.0003371218,0.00007302813,0.00005673225,1.936408e-7,0.000004920134,0.0009230337],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9685736,"threshold_uncertainty_score":0.9935619,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2313407085","doi":"10.1332/174426412x654031","title":"Use of research-based information by school practitioners and determinants of use: a review of empirical research","year":2012,"lang":"en","type":"review","venue":"Evidence & Policy","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":173,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Concordia University; Université de Montréal; Centre de Liaison Sur l'Intervention et la Prévention Psychosociales","funders":"","keywords":"Empirical research; Research policy; Public relations; Knowledge management; Psychology; Medical education; Political science; Medicine; Computer science; Public administration","retraction":null,"screen_n_in":null,"score":{"opus":0.8888262891940105,"gpt":0.7463502074399516,"spread":0.1424760817540589,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03857718,0.0002338429,0.001526719,0.002444659,0.0001449628,0.0002079993,0.000904686,0.0002528263,0.0006437538],"category_scores_gemma":[0.1159315,0.0001592246,0.0002634726,0.004770828,0.001031957,0.005542263,0.000358799,0.0007607568,0.0002492222],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002740257,"about_ca_system_score_gemma":0.006814746,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007159734,"about_ca_topic_score_gemma":0.00002137453,"domain_scores_codex":[0.98466,0.005123359,0.00298491,0.0003578305,0.006384028,0.0004898637],"domain_scores_gemma":[0.976152,0.01519896,0.00226969,0.001257446,0.004814682,0.0003072013],"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":[0.00003025727,0.00009369468,0.004018907,0.0611442,0.00002022969,4.707435e-7,0.0002300438,0.000001137935,0.000003251464,0.0002503328,0.05818309,0.8760244],"study_design_scores_gemma":[0.00007691131,0.0001855992,0.001148847,0.118773,0.00007276798,0.000006652067,0.00005246864,0.00006632446,0.0000217391,0.00008571434,0.8793883,0.0001216055],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0004742343,0.9960839,0.00003021673,0.0013729,0.00006131565,0.00165186,0.0001521337,0.000006621974,0.00016685],"genre_scores_gemma":[0.0010791,0.9974827,0.0005037773,0.0002836942,0.00007111188,0.0001425443,0.00003115755,0.00001295596,0.0003930309],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8759028,"threshold_uncertainty_score":0.9988157,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3110971321","doi":"10.1016/j.tree.2020.11.001","title":"Evaluating Impact Using Time-Series Data","year":2020,"lang":"en","type":"review","venue":"Trends in Ecology & Evolution","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":171,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Cambridge Trust; European Commission; Leverhulme Trust; Natural Environment Research Council; Cambridge Philosophical Society; Arcadia Fund; Bird Studies Canada; Villum Fonden","keywords":"Series (stratigraphy); Time series; Computer science; Statistics; Mathematics; Geology","retraction":null,"screen_n_in":null,"score":{"opus":0.6901198173060419,"gpt":0.6637457293661183,"spread":0.02637408793992357,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.008976687,0.000414935,0.001819312,0.001357916,0.0001891133,0.0001414946,0.001798912,0.000431807,0.008711573],"category_scores_gemma":[0.002709501,0.0003029071,0.0003174543,0.002818702,0.0001354547,0.001037469,0.0008314188,0.0005828172,0.002377114],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001348101,"about_ca_system_score_gemma":0.002148521,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004221383,"about_ca_topic_score_gemma":0.0003939004,"domain_scores_codex":[0.9932717,0.00202337,0.001826489,0.001143394,0.001262233,0.0004728167],"domain_scores_gemma":[0.9958365,0.001109499,0.001224988,0.001528917,0.0001849857,0.0001151391],"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":[0.00001181864,0.00003724811,0.001503434,0.000078735,0.00005080207,0.000004656102,0.00005612045,0.0007927967,8.086735e-7,0.00004499443,0.003571803,0.9938468],"study_design_scores_gemma":[0.0003797324,0.0003610922,0.009845714,0.0006330468,0.0004698985,0.00009420219,0.0000602162,0.2503909,7.274658e-8,0.0008726754,0.7364437,0.0004487049],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0004514815,0.9948406,0.0002729545,0.0002414725,0.001042058,0.0004420939,0.0002643498,0.00006169127,0.002383255],"genre_scores_gemma":[0.0009094991,0.9885375,0.005368002,0.00005180127,0.0005701274,0.00005186789,0.001585903,0.00005695294,0.002868344],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9933981,"threshold_uncertainty_score":0.9999423,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1965824013","doi":"10.1177/1098214013478142","title":"The Case for Participatory Evaluation in an Era of Accountability","year":2013,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":165,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Accountability; Technocracy; Citizen journalism; Participatory evaluation; Context (archaeology); Participatory GIS; Public sector; Sociology; Politics; Public administration; Public relations; Government (linguistics); Social accounting; Political science; Business; Accounting; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.3658990701241563,"gpt":0.5819220676204166,"spread":0.2160229974962604,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06152453,0.00008670531,0.0002680563,0.0001731548,0.0001388021,0.0001395658,0.0002852828,0.0000257447,0.0008524447],"category_scores_gemma":[0.006495482,0.00005166707,0.0001018592,0.0006170449,0.0002068225,0.001221398,0.00001850171,0.0001485512,0.00001455976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002291656,"about_ca_system_score_gemma":0.001046737,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001816407,"about_ca_topic_score_gemma":0.0008789924,"domain_scores_codex":[0.9933106,0.002223165,0.001491578,0.0001501892,0.002655332,0.0001690883],"domain_scores_gemma":[0.9876441,0.002507981,0.00177697,0.0004105217,0.007575229,0.0000851332],"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":[0.00009059246,0.00009918122,0.03286157,0.000001546771,0.00001130706,4.064665e-7,0.002471878,0.009100755,0.0004589886,0.00007705129,0.0002700547,0.9545566],"study_design_scores_gemma":[0.0011462,0.001111472,0.3556952,0.00001653843,0.00007755272,0.00004219375,0.02528007,0.5954453,0.0002970716,0.02062242,0.0001857139,0.00008033651],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9956147,0.0001095748,0.001291594,0.001679054,0.0002966348,0.0008882008,0.000002510991,0.000001802988,0.0001158948],"genre_scores_gemma":[0.9986309,0.0000110999,0.0009842381,0.0001272632,0.00008269704,0.0001510222,0.000002187676,0.00000531798,0.000005267404],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9544764,"threshold_uncertainty_score":0.966358,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2133886430","doi":"10.1002/wcc.73","title":"Participatory methods of integrated assessment—a review","year":2010,"lang":"en","type":"article","venue":"Wiley Interdisciplinary Reviews Climate Change","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":162,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Citizen journalism; Political science; Field (mathematics); Process (computing); Management science; Public relations; Sociology; Engineering ethics; Computer science; Engineering; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.5007791023596583,"gpt":0.6263696570533983,"spread":0.12559055469374,"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":[0.02917668,0.0003739606,0.001436475,0.0003170167,0.0002277131,0.00008861165,0.001256797,0.0001340523,0.01937957],"category_scores_gemma":[0.001277296,0.000234507,0.0005382926,0.001338667,0.0002391236,0.0008409367,0.001149128,0.0006422149,0.001110604],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006425138,"about_ca_system_score_gemma":0.0001345316,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005010729,"about_ca_topic_score_gemma":0.00008612648,"domain_scores_codex":[0.9931013,0.001853183,0.002708857,0.0006942975,0.00113319,0.0005091613],"domain_scores_gemma":[0.9951908,0.0007189489,0.001502201,0.001691998,0.00064924,0.0002468211],"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":[0.00002336225,0.0002689227,0.01522487,0.001359226,0.00002556121,0.000004269289,0.001491436,4.773237e-7,0.004527886,0.002038484,0.03159365,0.9434419],"study_design_scores_gemma":[0.0005218335,0.0004648865,0.02431474,0.009393409,0.0002380768,0.00003934342,0.001119061,0.004672605,0.0006186977,0.004190897,0.9537614,0.0006650678],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.1926612,0.4391285,0.01749379,0.05115327,0.02703023,0.02175086,0.001249047,0.0007262395,0.2488069],"genre_scores_gemma":[0.1964722,0.6643722,0.1162859,0.01306492,0.001437167,0.00588559,0.0005233213,0.0001844982,0.001774191],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9427768,"threshold_uncertainty_score":0.9996672,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2759512360","doi":"10.1186/s13012-017-0646-0","title":"Advancing the literature on designing audit and feedback interventions: identifying theory-informed hypotheses","year":2017,"lang":"en","type":"article","venue":"Implementation Science","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":160,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa; University of British Columbia; Ottawa Hospital; Women's College Hospital; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Psychological intervention; Medicine; Health services research; Audit; Health administration; Health informatics; Health care; Health economics; Public health; Nursing; Applied psychology; Psychology; Accounting","retraction":null,"screen_n_in":null,"score":{"opus":0.3972575760089357,"gpt":0.6178921478400988,"spread":0.2206345718311631,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01488621,0.0001202191,0.0001255742,0.0003311413,0.003717088,0.00560077,0.001360085,0.00002057417,0.0009527329],"category_scores_gemma":[0.006709795,0.00007028599,0.00007424974,0.0005491129,0.0005222167,0.004039162,0.0003055912,0.0001256406,0.0001396999],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009410956,"about_ca_system_score_gemma":0.0002996968,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000143124,"about_ca_topic_score_gemma":0.0002011985,"domain_scores_codex":[0.9966256,0.0001844763,0.000577928,0.0004171871,0.001880081,0.0003147301],"domain_scores_gemma":[0.996318,0.001597889,0.0007047438,0.0007913002,0.0004902575,0.00009776156],"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":[0.00005519129,0.00003395785,0.04717753,0.00003161447,0.00001769347,0.000002706169,0.01692606,0.0001700349,0.005811069,0.0467508,0.003360067,0.8796633],"study_design_scores_gemma":[0.001173443,0.0002076075,0.8374136,0.0005200054,0.00002339265,0.00001822923,0.04916221,0.001745857,0.01544676,0.08728727,0.006711119,0.0002905116],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9385494,0.0002448399,0.03983736,0.008320465,0.00147222,0.001001264,0.00002679081,0.00005156027,0.01049611],"genre_scores_gemma":[0.9957231,0.00007705073,0.002447036,0.000783039,0.00006158539,0.0000435738,0.000002359158,0.000004907555,0.0008573758],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8793728,"threshold_uncertainty_score":0.9999605,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1594257894","doi":"10.3138/cjpe.23.003","title":"Reframing Evaluation: Defining an Indigenous Evaluation Framework","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":159,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Cognitive reframing; Indigenous; Traditional knowledge; Focus group; Common ground; Sociology; Public relations; Political science; Engineering ethics; Psychology; Engineering; Social psychology; Anthropology; Ecology","retraction":null,"screen_n_in":null,"score":{"opus":0.396733003676957,"gpt":0.552227878522765,"spread":0.155494874845808,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07245219,0.0002736312,0.0004493441,0.001522558,0.00118894,0.0006292202,0.0009150127,0.0002739923,0.008595575],"category_scores_gemma":[0.01642953,0.0002358241,0.0002122362,0.001882623,0.0001953106,0.002185793,0.00002159183,0.0005971259,0.0002806355],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001730375,"about_ca_system_score_gemma":0.02561126,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005223277,"about_ca_topic_score_gemma":0.008125919,"domain_scores_codex":[0.9832129,0.0031676,0.001969507,0.0005177734,0.0105287,0.0006034719],"domain_scores_gemma":[0.9810528,0.0006045217,0.00170871,0.0007943672,0.01490828,0.0009313123],"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":[0.00002261393,0.00008495262,0.03465482,0.00000290023,0.00004113437,0.00001091034,0.01349066,0.02055896,0.00001956082,0.0003557827,0.00076893,0.9299888],"study_design_scores_gemma":[0.002924512,0.001940261,0.2955871,0.0001803921,0.0006329354,0.0005999489,0.007993684,0.6049691,0.0001406582,0.06572765,0.0186782,0.0006255984],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9838338,0.002488669,0.003372459,0.0007311968,0.002198379,0.002861754,0.000009603757,0.00002705073,0.004477085],"genre_scores_gemma":[0.9794262,0.00002905847,0.01921723,0.0003187683,0.0005447072,0.0002830508,0.00009365455,0.00002850955,0.00005883276],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9293632,"threshold_uncertainty_score":0.9923107,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["sts"],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2053476798","doi":"10.1016/j.obhdp.2009.11.007","title":"What types of advice do decision-makers prefer?","year":2010,"lang":"en","type":"article","venue":"Organizational Behavior and Human Decision Processes","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":158,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Wilfrid Laurier University; University of Ottawa","funders":"","keywords":"Advice (programming); Situational ethics; Psychology; Decision maker; Perspective (graphical); Autonomy; Interpersonal communication; Decision analysis; Social psychology; Management science; Applied psychology; Computer science; Political science; Engineering; Artificial intelligence","retraction":null,"screen_n_in":null,"score":{"opus":0.05479233121572701,"gpt":0.4294093933916098,"spread":0.3746170621758828,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001502901,0.0002061821,0.0003189763,0.0004394329,0.0004308307,0.0008085694,0.0006722049,0.0001508702,0.0166381],"category_scores_gemma":[0.003453507,0.0001511357,0.00005063038,0.00139286,0.0001900713,0.001659458,0.0002076443,0.0002084624,0.0002314206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000016308,"about_ca_system_score_gemma":0.000444329,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004820073,"about_ca_topic_score_gemma":0.0001990721,"domain_scores_codex":[0.9958792,0.00004185323,0.0009492529,0.0005734662,0.002368971,0.0001872995],"domain_scores_gemma":[0.9937398,0.00170147,0.0003833798,0.0005341902,0.003469952,0.0001711947],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001400468,0.0008063802,0.7151518,0.0000515618,0.00002521915,0.000007017143,0.001952549,0.0001385549,0.01166826,0.03014119,0.007680716,0.2322367],"study_design_scores_gemma":[0.001777648,0.0003094625,0.8145102,0.0002626299,0.0001276666,0.00006620827,0.002545171,0.0001368879,0.00866717,0.1348003,0.0361392,0.0006575148],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9959916,0.0003319652,0.001054461,0.0001671025,0.0007876543,0.0002599924,0.00002314817,0.00004915467,0.001334933],"genre_scores_gemma":[0.9937312,0.0002438065,0.004440813,0.000144956,0.00009569023,0.000021522,0.0000325369,0.00002499046,0.001264472],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2315792,"threshold_uncertainty_score":0.9842608,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1484543936","doi":"10.1177/160940690200100101","title":"Influence of the Research Frame on Qualitatively Derived Health Science Knowledge","year":2002,"lang":"en","type":"article","venue":"International Journal of Qualitative Methods","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":156,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Qualitative research; Field (mathematics); Orientation (vector space); Frame (networking); Epistemology; Process (computing); Body of knowledge; Psychology; Sociology; Knowledge management; Computer science; Social science; Mathematics","retraction":null,"screen_n_in":null,"score":{"opus":0.94641623802093,"gpt":0.8155280055453412,"spread":0.1308882324755888,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1560163,0.0001228648,0.0003737625,0.001281029,0.00032466,0.0001692253,0.003045001,0.00004017407,0.0004656022],"category_scores_gemma":[0.04752349,0.00006929951,0.00019103,0.002275639,0.00167575,0.0008100135,0.0002757085,0.0006362828,0.00008864516],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004422183,"about_ca_system_score_gemma":0.001116344,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002737282,"about_ca_topic_score_gemma":0.000008576645,"domain_scores_codex":[0.9729568,0.01657222,0.002007983,0.0003234356,0.007826641,0.0003129178],"domain_scores_gemma":[0.9634341,0.01936312,0.002007929,0.0004026165,0.01462132,0.0001709218],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0002251896,0.0007489899,0.002317326,0.00001467432,0.0001534983,0.000002562144,0.5638859,0.006461922,0.0289707,0.1726848,0.005317539,0.2192169],"study_design_scores_gemma":[0.002507789,0.002855709,0.2847377,0.0009402322,0.00001409526,0.00005639461,0.1670578,0.009994461,0.07132445,0.4207211,0.03938531,0.0004050021],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9400995,0.0004019848,0.02624756,0.02513707,0.001091978,0.0001981932,0.00001556183,0.000004217569,0.006803984],"genre_scores_gemma":[0.9357807,0.00006275168,0.06252913,0.0007699557,0.00009837679,0.000005013039,1.453341e-7,0.000006247881,0.0007476833],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3968281,"threshold_uncertainty_score":0.9604996,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1971555857","doi":"10.1016/j.polsoc.2010.03.004","title":"Policy analysis and policy work in federal systems: Policy advice and its contribution to evidence-based policy-making in multi-level governance systems","year":2010,"lang":"en","type":"article","venue":"Policy and Society","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":154,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Simon Fraser University","funders":"","keywords":"Work (physics); Policy analysis; Corporate governance; Public administration; Policy studies; Political science; Public policy; Order (exchange); Public economics; Economics; Management; Finance; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.1679909213834969,"gpt":0.489356959801353,"spread":0.3213660384178562,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.005337303,0.0004258259,0.0009090228,0.001795012,0.0005315993,0.001087122,0.0004773378,0.0003734638,0.00001103344],"category_scores_gemma":[0.01051017,0.0003649122,0.0001986358,0.009520574,0.0001951499,0.001013232,0.0002879978,0.0005338478,0.00001841997],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008280825,"about_ca_system_score_gemma":0.002591087,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.06257301,"about_ca_topic_score_gemma":0.008750222,"domain_scores_codex":[0.9949786,0.0005875861,0.001263047,0.0009051101,0.001305834,0.0009598026],"domain_scores_gemma":[0.9962405,0.001606967,0.0005926659,0.0005680615,0.0005081976,0.0004836214],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003327976,0.0003413346,0.6371726,0.0003748611,0.0003174443,0.000008584677,0.02535502,0.01990466,0.00216564,0.2916988,0.0009215054,0.02140669],"study_design_scores_gemma":[0.001811158,0.00007087327,0.8343166,0.0004173537,0.00003291144,0.00001145444,0.001013231,0.158577,0.00003954406,0.0007212744,0.002531844,0.0004567706],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9541388,0.001628399,0.001700755,0.04036373,0.0001766506,0.001167033,0.0002731663,0.00006635095,0.0004851382],"genre_scores_gemma":[0.9909441,0.001067303,0.0004648543,0.005212374,0.001200422,0.0001398276,0.000009167671,0.00002199372,0.0009399285],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2909776,"threshold_uncertainty_score":0.9999499,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2415836739","doi":"","title":"Developmental evaluation exemplars : principles in practice","year":2016,"lang":"en","type":"book","venue":"","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":154,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Indigenous; Sociology; Watson; Management; Library science","retraction":null,"screen_n_in":null,"score":{"opus":0.3816857263393895,"gpt":0.533974536343964,"spread":0.1522888100045744,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.02122089,0.000245253,0.0003413571,0.0006685382,0.0001143168,0.0002243853,0.000662261,0.0002474005,0.02886853],"category_scores_gemma":[0.005534299,0.0001525096,0.00008216701,0.0002566237,0.00005615234,0.001052339,0.0002072747,0.0002449865,0.009748576],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002265458,"about_ca_system_score_gemma":0.007963419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008767757,"about_ca_topic_score_gemma":0.0006375279,"domain_scores_codex":[0.9912925,0.000488564,0.001273131,0.0006732607,0.006015476,0.0002570933],"domain_scores_gemma":[0.9951003,0.002492341,0.0006881874,0.0003831416,0.00124781,0.00008828203],"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":[0.00008313494,0.0001060612,0.002044368,0.000008995416,0.0000593694,0.00001192427,0.001251199,0.0001510061,0.00003068895,0.04601278,0.2745675,0.6756729],"study_design_scores_gemma":[0.000647228,0.0000319511,0.003064384,0.00009813105,0.00002587918,0.00001431224,0.0004956858,0.001285039,0.00002287432,0.02254332,0.9715081,0.0002631075],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0004141703,0.0001058943,0.0004518877,0.003268116,0.0004448799,0.0007619684,0.00001112463,0.00002710247,0.9945149],"genre_scores_gemma":[0.001868953,0.0001340087,0.005150983,0.0009853023,0.0001372529,0.00009877544,0.00004118574,0.00001920295,0.9915643],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.6969405,"threshold_uncertainty_score":0.9976605,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2273826813","doi":"10.7748/nr2009.04.16.3.70.c6947","title":"Rigour in qualitative research: mechanisms for control","year":2009,"lang":"en","type":"article","venue":"Nurse Researcher","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":149,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Brandon University; University of Calgary","funders":"","keywords":"Rigour; Qualitative research; Nursing research; Control (management); Computer science; Management science; Psychology; Engineering ethics; Computational biology; Epistemology; Medicine; Sociology; Nursing; Biology; Artificial intelligence; Engineering; Philosophy; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.7725949983468796,"gpt":0.7359392547724206,"spread":0.036655743574459,"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":[0.06803908,0.0001022722,0.0002502689,0.0008438198,0.0002117323,0.0002546307,0.0007336675,0.0001049549,0.002105182],"category_scores_gemma":[0.01110387,0.00007252496,0.00009155615,0.00139846,0.0001421013,0.0003932369,0.00002412695,0.0004742696,0.000904042],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002017012,"about_ca_system_score_gemma":0.0004825984,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004778868,"about_ca_topic_score_gemma":0.0001650214,"domain_scores_codex":[0.990903,0.003152339,0.0005453948,0.0004751096,0.004222023,0.0007021789],"domain_scores_gemma":[0.9916059,0.005938808,0.00007709261,0.0005154974,0.001666938,0.0001956995],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001298205,0.0007271673,0.0005825724,0.000009915784,0.00002399173,0.0000154186,0.03279816,0.0003548255,0.00539444,0.5537189,0.1102072,0.2948692],"study_design_scores_gemma":[0.002609964,0.0008355054,0.01724797,0.00002586718,0.000001870343,8.826494e-7,0.03098802,0.02469895,0.0007408331,0.8944019,0.02831483,0.0001333995],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.2847793,0.0005188283,0.151776,0.2791411,0.0006864436,0.008013288,0.0001728765,0.0001388934,0.2747733],"genre_scores_gemma":[0.9803482,0.000009240783,0.002808023,0.0006007332,0.00007997052,0.0001739041,0.000005102008,0.000008917328,0.01596596],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6955689,"threshold_uncertainty_score":0.9998739,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1544576951","doi":"","title":"A Review of the Literature on Case Study Research","year":2008,"lang":"en","type":"review","venue":"DOAJ (DOAJ: Directory of Open Access Journals)","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":146,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Computer science; Data science; Information retrieval","retraction":null,"screen_n_in":null,"score":{"opus":0.8931945989899815,"gpt":0.7971990924816132,"spread":0.0959955065083683,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.03449428,0.000609928,0.003809446,0.002636679,0.0007936478,0.002252998,0.01185976,0.0002556166,0.01698733],"category_scores_gemma":[0.007792914,0.0003070157,0.001260478,0.01106434,0.0003002301,0.001693622,0.002946648,0.002381117,0.0001486057],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002419609,"about_ca_system_score_gemma":0.002246968,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002352299,"about_ca_topic_score_gemma":0.00005883504,"domain_scores_codex":[0.9761184,0.009994899,0.004183768,0.001021322,0.00818668,0.0004949287],"domain_scores_gemma":[0.9828843,0.005686148,0.004291669,0.003037626,0.003812019,0.0002881645],"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":[0.00001871954,0.0005807767,0.002248846,0.008437269,0.0001879102,0.0003941258,0.0004263134,0.000003944443,0.000001516004,0.00001944313,0.2737661,0.713915],"study_design_scores_gemma":[0.0002023195,0.00004202621,0.001732869,0.151432,0.0001937118,0.0007816408,0.0002418574,0.000003062548,0.000004304642,0.0004460303,0.8446277,0.0002925013],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001301312,0.9867095,0.000001780538,0.0001700494,0.0009111908,0.004200243,0.0001107092,0.000008794581,0.006586416],"genre_scores_gemma":[0.0009176703,0.9962684,0.00002062075,0.000474614,0.0002325693,0.0002573807,0.000009680725,0.00005219784,0.001766864],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7136225,"threshold_uncertainty_score":0.9999382,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2775174218","doi":"10.3138/cjpe.31122","title":"Theory of Change Analysis: Building Robust Theories of Change","year":2017,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":145,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Theory of change; Computer science; Management science; Risk analysis (engineering); Positive economics; Econometrics; Economics; Business; Management","retraction":null,"screen_n_in":null,"score":{"opus":0.7318312016170765,"gpt":0.5642255157815168,"spread":0.1676056858355597,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02695549,0.0001136358,0.0004325363,0.001245814,0.00031207,0.0002771632,0.0009126268,0.0000778667,0.002079111],"category_scores_gemma":[0.004102443,0.00008389902,0.0002730514,0.0007235947,0.0002827659,0.001225016,0.00002846376,0.0001323784,0.000008144712],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001332133,"about_ca_system_score_gemma":0.001414063,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00131182,"about_ca_topic_score_gemma":0.01937543,"domain_scores_codex":[0.9960204,0.0006297944,0.001003098,0.0001674196,0.001959151,0.00022014],"domain_scores_gemma":[0.9925872,0.0002683639,0.002643926,0.0006208509,0.003619072,0.0002606036],"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":[0.00003053009,0.00003606013,0.1773194,0.000009459533,0.0002196487,0.000002095182,0.004115089,0.0006708325,0.00004289546,0.01083199,0.00009081062,0.8066311],"study_design_scores_gemma":[0.0008044912,0.0005412624,0.9022572,0.0001362424,0.00105172,0.000007164045,0.003570409,0.03193347,0.0005632939,0.05624328,0.002730956,0.0001605033],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9858212,0.001821543,0.004568031,0.002264469,0.001038813,0.001255007,0.00003408555,0.000005312036,0.003191527],"genre_scores_gemma":[0.9968542,0.00002688133,0.002668766,0.00006407727,0.0002457231,0.00006105687,0.000005214231,0.000006776988,0.00006730475],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8064706,"threshold_uncertainty_score":0.9988331,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2120292430","doi":"10.1177/1098214012440030","title":"A New Realistic Evaluation Analysis Method","year":2012,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":144,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Public Health Ontario; University of Toronto","funders":"Health Research Board","keywords":"Coding (social sciences); Computer science; Qualitative research; Qualitative analysis; Narrative; Management science; Context (archaeology); Evaluation methods; Data science; Psychology; Sociology; Social science; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.2974231815816114,"gpt":0.6135208376358036,"spread":0.3160976560541922,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.08044642,0.0001415747,0.000533049,0.001323794,0.0001095023,0.0001471727,0.0003528727,0.00003633859,0.009592068],"category_scores_gemma":[0.008161817,0.0001007493,0.000343252,0.004146119,0.00005274584,0.001195702,0.00003159003,0.0001700161,0.0001588976],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003798557,"about_ca_system_score_gemma":0.001389211,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001176767,"about_ca_topic_score_gemma":0.00005235481,"domain_scores_codex":[0.9862043,0.003609062,0.001451623,0.0002130883,0.008253094,0.0002688432],"domain_scores_gemma":[0.9905058,0.001539608,0.002638974,0.0004734318,0.004504244,0.0003379342],"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":[0.0000564718,0.00005386677,0.03920751,5.019984e-7,0.0002852232,2.033798e-7,0.001772269,0.04308439,0.0001658937,0.0003873015,0.003879573,0.9111068],"study_design_scores_gemma":[0.001017387,0.0004586474,0.5332966,0.00001123998,0.004213882,0.00004073945,0.003279864,0.443103,0.0001597274,0.009809973,0.004422669,0.0001862399],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2729564,0.0004732068,0.7160211,0.002329892,0.0008939134,0.0003456153,0.000003462476,0.00000912737,0.006967263],"genre_scores_gemma":[0.9390099,0.00001810296,0.05985039,0.0003308245,0.0005459119,0.000008960162,0.00001233311,0.000007946536,0.0002156545],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9109206,"threshold_uncertainty_score":0.9913133,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2121899445","doi":"","title":"A Book Review: Anfara, V. A., & Mertz, N. T. (Eds.). (2006). Theoretical frameworks in qualitative research. Thousand Oaks, CA: Sage","year":2008,"lang":"en","type":"article","venue":"DOAJ (DOAJ: Directory of Open Access Journals)","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":139,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Task (project management); Qualitative research; Epistemology; Sociology; Process (computing); Guideline; Engineering ethics; Computer science; Management science; Data science; Engineering; Social science; Philosophy; Political science; Systems engineering; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.4891708145727964,"gpt":0.7036625099608368,"spread":0.2144916953880404,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04556262,0.0004075849,0.001457985,0.001885266,0.0006790431,0.001265841,0.005061719,0.000322991,0.1882254],"category_scores_gemma":[0.009803932,0.000295204,0.0003148552,0.004338606,0.00132988,0.003910657,0.001330645,0.002308569,0.000580524],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002002052,"about_ca_system_score_gemma":0.0009493391,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003583066,"about_ca_topic_score_gemma":0.00008589508,"domain_scores_codex":[0.9815946,0.007271161,0.002784603,0.000978129,0.00647512,0.0008964313],"domain_scores_gemma":[0.9870134,0.008189272,0.001144897,0.001263442,0.001879484,0.000509496],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002037696,0.000348213,0.02545398,0.00006886714,0.00005823528,0.00008972354,0.00277355,0.00009935012,0.0004263459,0.001070349,0.9511336,0.01827403],"study_design_scores_gemma":[0.001164968,0.00006772091,0.1331394,0.0033414,0.00004732955,0.00004969329,0.001521538,0.0007845294,0.001441495,0.06719663,0.7904986,0.0007466822],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.07694244,0.8781567,0.001459615,0.01015463,0.0008416768,0.00195656,0.0001250677,0.00003697226,0.03032633],"genre_scores_gemma":[0.2777932,0.6989488,0.0004668041,0.01442772,0.0002159991,0.0002207215,0.00002139148,0.00005213545,0.007853219],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.2008508,"threshold_uncertainty_score":0.9999931,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W256235996","doi":"10.1093/oxfordhb/9780199548453.003.0015","title":"The Politics of Policy Evaluation","year":2009,"lang":"en","type":"book-chapter","venue":"Oxford University Press eBooks","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":136,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute on Governance","funders":"","keywords":"Politics; Political science; Policy analysis; Public administration; Public policy; Key (lock); Computer science; Law; Computer security","retraction":null,"screen_n_in":null,"score":{"opus":0.1710313777369994,"gpt":0.4025742096289181,"spread":0.2315428318919187,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002363105,0.0001921762,0.000280849,0.0003279778,0.0004330761,0.0001003617,0.001029676,0.0002026313,0.0002104488],"category_scores_gemma":[0.0002721261,0.0001413523,0.0002090319,0.00003032554,0.0002433431,0.0001212313,0.0002011434,0.0002331184,0.000007631521],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002836279,"about_ca_system_score_gemma":0.001133563,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000295116,"about_ca_topic_score_gemma":0.00003528867,"domain_scores_codex":[0.9963846,0.0001538205,0.0004503349,0.0003048079,0.002497086,0.0002094052],"domain_scores_gemma":[0.9964733,0.0006336524,0.000679924,0.0008608393,0.001255591,0.00009672617],"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":[0.00002107107,0.000004363067,0.000004285795,0.000002625831,0.00002624627,9.953581e-7,0.0001669817,0.00006302742,0.000002236275,0.7998837,0.002892311,0.1969322],"study_design_scores_gemma":[0.0003235132,0.00007232685,0.0001184432,0.00002813933,0.00008390522,0.000001406338,0.0001383002,0.001446014,0.00004133289,0.07433396,0.9232799,0.0001328],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0001637202,0.00008514596,0.0001205187,0.0003727131,0.0001568644,0.00040551,0.00005175664,0.00001832498,0.9986255],"genre_scores_gemma":[0.006312576,0.0002718945,0.00008371516,0.0001036021,0.0001378508,4.846332e-7,0.00001216808,0.00001013848,0.9930676],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.9203876,"threshold_uncertainty_score":0.5764179,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2778556126","doi":"10.15171/ijhpm.2017.142","title":"The Qualitative Descriptive Approach in International Comparative Studies: Using Online Qualitative Surveys","year":2017,"lang":"en","type":"article","venue":"International Journal of Health Policy and Management","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":136,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Vancouver Coastal Health; Vancouver Coastal Health Research Institute; University of British Columbia","funders":"","keywords":"Depiction; Context (archaeology); Realm; Data science; Qualitative research; Data collection; Computer science; Proposition; Qualitative property; Descriptive statistics; Sociology; Management science; Political science; Epistemology; Social science","retraction":null,"screen_n_in":null,"score":{"opus":0.8202046577575901,"gpt":0.737857736916085,"spread":0.08234692084150508,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03256046,0.000138834,0.0003869212,0.0006081332,0.0005726189,0.0004823018,0.001314433,0.0000254936,0.00001362569],"category_scores_gemma":[0.002257394,0.00008523451,0.00008420157,0.000157654,0.0003409441,0.0008489029,0.0003839852,0.0002166793,0.000004892475],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006740511,"about_ca_system_score_gemma":0.0003616229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007375666,"about_ca_topic_score_gemma":0.000686954,"domain_scores_codex":[0.9931738,0.002812237,0.001559403,0.0002300529,0.001997855,0.0002266185],"domain_scores_gemma":[0.9938699,0.001695008,0.002413151,0.0002380094,0.001698883,0.00008503793],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0004333109,0.0004997107,0.001430053,0.00003055152,0.001431859,0.000024586,0.6537822,0.002446381,0.000004257479,0.2019723,0.00391154,0.1340332],"study_design_scores_gemma":[0.002166874,0.0002322269,0.04659219,0.0002151182,0.00001175944,0.00002634315,0.8426906,0.01006999,0.000003963226,0.07174715,0.02609667,0.0001470634],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6359779,0.003400669,0.07016329,0.2640708,0.007167954,0.001343961,0.00019944,0.00001356923,0.01766235],"genre_scores_gemma":[0.9879096,0.003135017,0.006741957,0.001150936,0.0003174872,0.00001004686,0.000005334474,0.000005189926,0.0007244219],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3519317,"threshold_uncertainty_score":0.9961826,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2620297369","doi":"10.1007/978-3-319-56129-5","title":"Understanding and Investigating Response Processes in Validation Research","year":2017,"lang":"en","type":"book","venue":"Social indicators research series","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":131,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychology; Test (biology); Data science; Computer science; Biology","retraction":null,"screen_n_in":null,"score":{"opus":0.8421524619491882,"gpt":0.6459391458506124,"spread":0.1962133160985758,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","scholarly_communication","research_integrity"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.09630429,0.0002595665,0.0005415409,0.004855587,0.004596794,0.002871471,0.001516902,0.0006312368,0.0004518363],"category_scores_gemma":[0.05306182,0.0002260365,0.00006114967,0.002321634,0.004409959,0.001549734,0.0009750838,0.002572735,0.0001850875],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001893862,"about_ca_system_score_gemma":0.01427922,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006141275,"about_ca_topic_score_gemma":0.0009325898,"domain_scores_codex":[0.9824128,0.00521476,0.00094874,0.0009613399,0.009536144,0.0009262478],"domain_scores_gemma":[0.9875684,0.009066406,0.0005899268,0.0006854331,0.001820222,0.0002696204],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002143973,0.0001997572,0.07754075,0.001403768,0.0001725162,0.0001027894,0.1770522,0.00001511591,0.0005198719,0.2502068,0.4680711,0.0225713],"study_design_scores_gemma":[0.0004595196,0.0003940171,0.01304909,0.0005582505,0.000006229721,0.000002954194,0.02553471,0.00002444161,0.0002890538,0.6822923,0.2770057,0.0003837284],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1970157,0.0007794315,0.00003690436,0.02517884,0.0004491682,0.00262036,0.0001337562,0.00009478271,0.7736911],"genre_scores_gemma":[0.5101663,0.0006919696,0.0001003471,0.0000407818,0.0005509938,0.0002028906,0.00005852355,0.00005685037,0.4881314],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.4320855,"threshold_uncertainty_score":0.9997284,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2163053247","doi":"10.1097/01.aids.0000343760.70078.89","title":"Evaluation design for large-scale HIV prevention programmes: the case of Avahan, the India AIDS initiative","year":2008,"lang":"en","type":"review","venue":"AIDS","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":126,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; Centre hospitalier universitaire de Québec","funders":"","keywords":"Monitoring and evaluation; Program evaluation; Government (linguistics); Scale (ratio); Bespoke; Medicine; Environmental health; Business; Economic growth; Political science; Geography; Public administration; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.4563612954648097,"gpt":0.5529884177901366,"spread":0.09662712232532694,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03200427,0.0003183978,0.0009006106,0.0003011474,0.0005730175,0.0001714897,0.000904703,0.0002226098,0.0006076883],"category_scores_gemma":[0.003212434,0.0001437164,0.0006963463,0.001438212,0.0001558811,0.0003620493,0.0001208471,0.0003027705,0.0001470959],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001501094,"about_ca_system_score_gemma":0.002175738,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005247908,"about_ca_topic_score_gemma":0.00003640959,"domain_scores_codex":[0.9907405,0.004719506,0.001514947,0.0005107265,0.00219961,0.0003146978],"domain_scores_gemma":[0.9902488,0.005388651,0.001590551,0.0009761104,0.001737355,0.00005851227],"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":[0.000007573218,0.00009899541,0.00001882462,0.0001769528,0.0001220928,0.00000227483,0.001894907,0.00003878294,3.832022e-8,0.000548913,0.007991008,0.9890996],"study_design_scores_gemma":[0.0006694964,0.0007680908,0.00002749752,0.0008700993,0.001299587,0.0001866652,0.003634826,0.007812673,0.000006687135,0.00736526,0.9771262,0.0002329254],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00007935634,0.914721,0.07380412,0.0001377453,0.0005971811,0.009006983,0.0001789033,0.00002386727,0.001450887],"genre_scores_gemma":[0.001692794,0.9862549,0.005728955,0.0001461806,0.0002607279,0.004341638,0.000232495,0.00003718402,0.001305176],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9888667,"threshold_uncertainty_score":0.9967553,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2548378115","doi":"10.1177/0739456x16675930","title":"Evaluation Theory and Practice: Comparing Program Evaluation and Evaluation in Planning","year":2016,"lang":"en","type":"article","venue":"Journal of Planning Education and Research","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":123,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"Program evaluation; Evaluation methods; Outcome (game theory); Management science; Impact evaluation; Theory of change; Process management; Political science; Computer science; Business; Engineering; Public administration; Economics; Management; Medicine","retraction":null,"screen_n_in":null,"score":{"opus":0.6508698838348513,"gpt":0.7143752125896877,"spread":0.06350532875483639,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2270684,0.0001001195,0.0002066872,0.001244786,0.0002647672,0.0004352937,0.0001609255,0.0000825408,0.0002052202],"category_scores_gemma":[0.03962864,0.00006119938,0.00002160638,0.0005898369,0.0001543567,0.001419228,0.00006866404,0.0003286509,0.000006344019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003129304,"about_ca_system_score_gemma":0.002221317,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000939084,"about_ca_topic_score_gemma":0.000008265448,"domain_scores_codex":[0.9835698,0.0083188,0.000885952,0.0003343568,0.006641762,0.0002493528],"domain_scores_gemma":[0.9860083,0.006449745,0.0005583122,0.0001768829,0.006651406,0.0001553418],"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":[0.0002255056,0.0001629819,0.09130771,0.000008905349,0.00001858783,7.379317e-7,0.008856738,0.0004814627,0.0005880463,0.0005374738,0.0009843089,0.8968275],"study_design_scores_gemma":[0.002702998,0.0004727411,0.8192401,0.000729217,0.0001092634,0.0001792341,0.03278574,0.09441547,0.00008058509,0.04055252,0.008591017,0.000141108],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9829497,0.005780747,0.0003217213,0.004335065,0.0003807462,0.001030886,4.817376e-7,0.000005034085,0.005195671],"genre_scores_gemma":[0.9980395,0.0002144447,0.001236262,0.00006585731,0.000184137,0.0001086261,0.000002765039,0.000006683579,0.0001416953],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8966864,"threshold_uncertainty_score":0.968461,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2800401326","doi":"10.1177/1098214018765698","title":"Outcomes and Impacts of Development Interventions","year":2018,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":120,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Royal Roads University","funders":"International Fund for Agricultural Development; Consortium of International Agricultural Research Centers; Centre for International Forestry Research; Canada Research Chairs; United Nations Development Programme","keywords":"CLARITY; Psychological intervention; Consistency (knowledge bases); Outcome (game theory); Accountability; Confusion; Management science; Psychology; Computer science; Sociology; Political science; Economics","retraction":null,"screen_n_in":null,"score":{"opus":0.2737634216383244,"gpt":0.5787112819025034,"spread":0.304947860264179,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01181891,0.00006126443,0.0002483221,0.0003475341,0.000067655,0.00004331749,0.0001496351,0.00001190632,0.001016921],"category_scores_gemma":[0.00220232,0.00003933239,0.00009056251,0.0004205721,0.0002053173,0.0004622133,0.00003324825,0.00005454061,0.00003021261],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005629722,"about_ca_system_score_gemma":0.0005211114,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007969195,"about_ca_topic_score_gemma":0.00003772479,"domain_scores_codex":[0.9966794,0.000327317,0.001015414,0.00009055112,0.00180506,0.00008226833],"domain_scores_gemma":[0.9954387,0.0003576854,0.001524976,0.0001417035,0.002461234,0.00007569888],"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":[0.00004291228,0.00006814596,0.2518061,0.000002770231,0.00005665754,2.202018e-7,0.002524343,0.00004020466,0.00048378,0.0002323389,0.0004708661,0.7442716],"study_design_scores_gemma":[0.0005413622,0.0009014718,0.9900609,0.00006337133,0.00004447654,0.00001469398,0.002282281,0.002044455,0.001224651,0.001731304,0.001038453,0.00005263404],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.990976,0.0000636114,0.006919491,0.00106276,0.0002295233,0.0000828946,8.281481e-7,0.00000171875,0.0006631245],"genre_scores_gemma":[0.9925541,0.00001350645,0.007161065,0.0001566256,0.00004617428,0.000001714813,6.766589e-7,0.000002849355,0.00006332814],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.744219,"threshold_uncertainty_score":0.9998963,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1975564755","doi":"10.1177/1098214013477235","title":"Understanding Dimensions of Organizational Evaluation Capacity","year":2013,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":118,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Ottawa; École Nationale d'Administration Publique","funders":"Australian Government","keywords":"Dimension (graph theory); Capacity building; Government (linguistics); Knowledge management; Organization development; Business; Organizational learning; Process management; Computer science; Economics; Economic growth","retraction":null,"screen_n_in":null,"score":{"opus":0.5129125946126178,"gpt":0.4844682783398096,"spread":0.02844431627280825,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01910018,0.0001094667,0.0003316766,0.0005714151,0.0001347644,0.00009478243,0.0002615037,0.00003310406,0.008698439],"category_scores_gemma":[0.006523244,0.00007107978,0.0001120218,0.001496452,0.0002100906,0.0009792264,0.00003289757,0.0001383088,0.0001209087],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004922193,"about_ca_system_score_gemma":0.001124391,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003710673,"about_ca_topic_score_gemma":0.00001639104,"domain_scores_codex":[0.990752,0.001275434,0.001320222,0.0001766352,0.006333304,0.0001423445],"domain_scores_gemma":[0.9866315,0.001033897,0.002368042,0.0002809939,0.009570603,0.0001150139],"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":[0.0001624636,0.0007334481,0.2735724,0.00001296341,0.0004161826,0.000001011216,0.01086504,0.1709167,0.04356529,0.0174337,0.02258581,0.4597349],"study_design_scores_gemma":[0.001769225,0.0009835371,0.2806074,0.00006787218,0.0002852064,0.00004969687,0.01304665,0.497262,0.001848736,0.2035649,0.0002909997,0.0002237907],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9566416,0.00004203936,0.03805484,0.002062328,0.0004543359,0.0004082503,0.000003051229,0.000004784384,0.002328777],"genre_scores_gemma":[0.995521,0.00001925611,0.004124605,0.0001777771,0.00009882332,0.000009666234,0.00000676589,0.000008409908,0.00003371686],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4595112,"threshold_uncertainty_score":0.9922078,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1713678554","doi":"10.56645/jmde.v1i1.144","title":"Unpacking the Participatory Process","year":2004,"lang":"en","type":"article","venue":"Journal of MultiDisciplinary Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":116,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Unpacking; Citizen journalism; Process (computing); Participatory action research; Sociology; Computer science; Linguistics; Anthropology; World Wide Web","retraction":null,"screen_n_in":null,"score":{"opus":0.4185578280274691,"gpt":0.5807891987238559,"spread":0.1622313706963868,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02926393,0.0001121422,0.0002086713,0.000244679,0.0003855488,0.0001604463,0.0006156685,0.00005013071,0.0005789404],"category_scores_gemma":[0.002714318,0.000057794,0.0001518386,0.0006487271,0.00009244587,0.001040633,0.00006436052,0.0002526594,0.0001387654],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002293693,"about_ca_system_score_gemma":0.001299178,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001785504,"about_ca_topic_score_gemma":0.00002441436,"domain_scores_codex":[0.9930677,0.0005869774,0.001180173,0.000174227,0.004807885,0.0001829753],"domain_scores_gemma":[0.9950879,0.0005598975,0.001237641,0.0003490257,0.00266955,0.0000960523],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.0002061366,0.0003843575,0.0425607,0.00000975351,0.00005739502,0.000009186549,0.03173842,0.8148253,0.002278961,0.001044969,0.0006851172,0.1061997],"study_design_scores_gemma":[0.004535934,0.0007419103,0.5798056,0.0001686388,0.000226649,0.0001467432,0.02624557,0.1564118,0.005251966,0.2249998,0.001193845,0.0002715633],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9832875,0.0002503344,0.004428559,0.009536241,0.0008645875,0.0003006312,0.000001021185,0.00000715259,0.001323989],"genre_scores_gemma":[0.9988037,0.00001722458,0.0006321574,0.0001296857,0.0003247077,0.00001664283,0.00000120677,0.000007606176,0.00006708209],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6584135,"threshold_uncertainty_score":0.999577,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2324664192","doi":"10.1177/1356389014564248","title":"The institutionalization of evaluation matters: Updating the <i>International Atlas of Evaluation</i> 10 years later","year":2015,"lang":"en","type":"article","venue":"Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":114,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval","funders":"","keywords":"Institutionalisation; Political science; Regional science; Public administration; Geography; Law","retraction":null,"screen_n_in":null,"score":{"opus":0.2102689892953537,"gpt":0.4898436435508309,"spread":0.2795746542554772,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07178859,0.0001495262,0.0001990999,0.0002427507,0.000268019,0.0002438684,0.0006609633,0.00008097247,0.004859765],"category_scores_gemma":[0.01087038,0.00009052535,0.000118416,0.0008691557,0.0002237658,0.0008932847,0.0001351706,0.0001155614,0.0003690316],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004202178,"about_ca_system_score_gemma":0.002098589,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003811751,"about_ca_topic_score_gemma":0.0001409408,"domain_scores_codex":[0.9822151,0.002071504,0.001502384,0.0003835365,0.01363366,0.000193791],"domain_scores_gemma":[0.986153,0.001104237,0.001332236,0.0007328261,0.01061667,0.00006101394],"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":[0.0002743169,0.0002132381,0.01464059,0.000009817303,0.0001903928,2.556244e-7,0.006394628,0.4119725,0.002095879,0.0141995,0.1377852,0.4122236],"study_design_scores_gemma":[0.001961516,0.00007998525,0.05222847,0.00003409219,0.0001868774,0.000004835899,0.002446987,0.8618255,0.00158068,0.03365036,0.04586974,0.0001310026],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9080491,0.001107223,0.003734284,0.02608272,0.004665275,0.003868282,0.00004980385,0.0000384776,0.05240481],"genre_scores_gemma":[0.9978589,0.00002755371,0.0004092325,0.0004820907,0.000224025,0.0002346401,0.0002569448,0.00001100381,0.0004956533],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4498529,"threshold_uncertainty_score":0.9974615,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W2015742271","doi":"10.1016/s0277-9536(01)00247-7","title":"Evaluating the participatory process in a community-based heart health project","year":2002,"lang":"en","type":"article","venue":"Social Science & Medicine","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":114,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Vancouver Coastal Health; University of Victoria; Ministry of Health","funders":"","keywords":"Participatory action research; Agency (philosophy); Fieldnotes; Public relations; Project team; Community-based participatory research; Citizen journalism; Community project; Population; Sociology; Psychology; Medical education; Business; Medicine; Political science; Knowledge management; Environmental health; Computer science","retraction":null,"screen_n_in":null,"score":{"opus":0.701204145841031,"gpt":0.6632814748752291,"spread":0.03792267096580182,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.06204068,0.0001126375,0.0002866656,0.0003314156,0.002237485,0.0001109263,0.0011149,0.00003224736,0.0007887106],"category_scores_gemma":[0.004721736,0.00006067673,0.00003879612,0.004490064,0.001707417,0.0003775093,0.0000565225,0.0004682602,0.00005804396],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002339856,"about_ca_system_score_gemma":0.001419204,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004579047,"about_ca_topic_score_gemma":0.0003486217,"domain_scores_codex":[0.9920365,0.001874066,0.0007325378,0.0003105884,0.004522403,0.000523918],"domain_scores_gemma":[0.9976481,0.001044925,0.0003004934,0.0004254287,0.0004672852,0.0001137864],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00004352453,0.0004705483,0.1457466,0.00004731876,0.000003802639,0.000001254186,0.6240122,0.001012003,0.001139032,0.001463904,0.0184911,0.2075688],"study_design_scores_gemma":[0.004942551,0.003823951,0.2972554,0.0004279847,0.00002404638,0.000007457931,0.1803302,0.4677801,0.0002845346,0.01476417,0.02976443,0.0005951241],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9303197,0.0001022773,0.00007182144,0.06125131,0.0002790977,0.0005604363,0.000001333548,0.00002842516,0.00738564],"genre_scores_gemma":[0.9862035,0.000003514923,0.00004060571,0.01337696,0.0001593913,0.0000875699,0.000001097015,0.000004460087,0.0001229112],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4667681,"threshold_uncertainty_score":0.9990615,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2102960304","doi":"10.1177/1098214012464037","title":"Arguments for a Common Set of Principles for Collaborative Inquiry in Evaluation","year":2013,"lang":"en","type":"article","venue":"American Journal of Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":113,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; Carleton University; University of Ottawa","funders":"","keywords":"Logic model; Set (abstract data type); Context (archaeology); Field (mathematics); Stakeholder; Program evaluation; Management science; Engineering ethics; Sociology; Computer science; Epistemology; Knowledge management; Public relations; Political science; Social science; Public administration; Engineering","retraction":null,"screen_n_in":null,"score":{"opus":0.3925397717083799,"gpt":0.5662179449091042,"spread":0.1736781732007242,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02986495,0.0001181926,0.0004737499,0.0005855241,0.00006999497,0.00008531129,0.0002877161,0.00003708544,0.0005051129],"category_scores_gemma":[0.006137033,0.00008853722,0.0001253126,0.0009930803,0.0001350084,0.0008650051,0.00002447766,0.00008149468,0.00001168903],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002925213,"about_ca_system_score_gemma":0.001271717,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002315656,"about_ca_topic_score_gemma":0.00007633395,"domain_scores_codex":[0.993849,0.001068436,0.001714679,0.0002019802,0.002994235,0.0001716621],"domain_scores_gemma":[0.9852949,0.00197222,0.003016125,0.000248715,0.009396606,0.00007141837],"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":[0.0004938131,0.0001903483,0.04199456,0.00001398281,0.00008054135,9.910812e-8,0.01138916,0.05161992,0.002236897,0.0005150211,0.003115026,0.8883506],"study_design_scores_gemma":[0.005472619,0.003473912,0.2617393,0.0001106951,0.0001705561,0.000004547064,0.03310417,0.6547024,0.002050935,0.03649376,0.002490915,0.0001861505],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9855835,0.000102092,0.009325097,0.00177112,0.0004096548,0.002485854,0.00001987165,0.000002198318,0.0003005954],"genre_scores_gemma":[0.9892596,0.00001828076,0.009960947,0.0001531373,0.000131214,0.0004107251,0.00002621216,0.00000874278,0.00003114015],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8881645,"threshold_uncertainty_score":0.9989582,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}