{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":892,"total_is_capped":false,"direct_labels_cover":52,"predictions_cover":892,"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","author_layer_release":"2026-06-26"},"query_hash":"466ddf5883b4","filters":{"venue":"Canadian Journal of Program Evaluation"}},"results":[{"id":"W1663638353","doi":"10.3138/cjpe.30.1.108","title":"Robert K. Yin. (2014). <i>Case Study Research Design and Methods (5th ed.).</i>","year":2015,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Qualitative Comparative Analysis Research","field":"Social Sciences","cited_by":406,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Psychology; Philosophy; Sociology","authors":[{"name":"Trista Hollweck","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6829255352418476,"gpt":0.6779845172067633,"spread":0.004941018035084288,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.029781,0.002018087,0.001048459,0.00745489,0.003653242,0.004405613,0.002244778,0.002689517,0.0219285],"category_scores_gemma":[0.05177889,0.002115208,0.0008713582,0.006992853,0.005416553,0.009186066,0.002653795,0.008071247,0.01344532],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005103575,"about_ca_system_score_gemma":0.01438703,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03330822,"about_ca_topic_score_gemma":0.06936149,"domain_scores_codex":[0.9892762,0.005781197,0.001155578,0.0005411503,0.003009275,0.0002366282],"domain_scores_gemma":[0.9348931,0.04314958,0.002836506,0.001602436,0.01632929,0.001189145],"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":[0.00004991242,0.0000457421,0.000866619,0.00288227,0.00002770126,0.0001503625,0.003055526,0.0002543624,0.0003800475,0.02432662,0.6488004,0.3191604],"study_design_scores_gemma":[0.0000472024,0.0000692937,0.003843266,0.01115862,0.0001005593,0.0003729652,0.004105941,0.0002857108,0.001111351,0.04359932,0.9351793,0.0001265108],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.002307758,0.4594231,0.186167,0.2111774,0.013715,0.00212161,0.003586668,0.002850197,0.1186513],"genre_scores_gemma":[0.03602777,0.4862839,0.3806584,0.02150985,0.002991993,0.006189463,0.002186295,0.001427443,0.06272488],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.03330822,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"opus","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"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","authors":[{"name":"John Mayne","is_ca":true}],"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2501988,0.002366789,0.001465762,0.009304908,0.005556213,0.02194463,0.004056383,0.004386266,0.005542464],"category_scores_gemma":[0.5074871,0.0006854335,0.001222845,0.008964634,0.01608508,0.02803998,0.01388507,0.009504279,0.001142507],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007487682,"about_ca_system_score_gemma":0.01189447,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001871119,"about_ca_topic_score_gemma":0.001441552,"domain_scores_codex":[0.5601743,0.3661699,0.00970088,0.005340076,0.05454807,0.004066752],"domain_scores_gemma":[0.3836576,0.4559083,0.04896079,0.04298695,0.06314822,0.005338111],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005304755,0.0007259019,0.04637861,0.001847096,0.0004623037,0.0002591189,0.03509289,0.004942377,0.001774646,0.4606157,0.01762081,0.4297501],"study_design_scores_gemma":[0.0001664698,0.0009136579,0.02325374,0.00356083,0.0003885338,0.0004174144,0.03337291,0.04519353,0.01555561,0.8025494,0.07419185,0.0004361568],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09221153,0.00242573,0.6746214,0.07397915,0.002323177,0.001373435,0.0002996597,0.001126257,0.1516397],"genre_scores_gemma":[0.8616789,0.0009639693,0.1294415,0.001966472,0.0006381311,0.000983318,0.0001093175,0.0002376686,0.003980828],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7498012,"threshold_uncertainty_score":0.9246384,"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","authors":[{"name":"John Mayne","is_ca":false}],"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00942474,0.00136629,0.0009386336,0.002382661,0.001178016,0.00420648,0.002685595,0.003131139,0.01036225],"category_scores_gemma":[0.0218691,0.0004531274,0.001447932,0.00223167,0.004318601,0.005028626,0.00225605,0.004231288,0.001465584],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004854846,"about_ca_system_score_gemma":0.003999347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004107931,"about_ca_topic_score_gemma":0.0036031,"domain_scores_codex":[0.9921045,0.005060092,0.0002965251,0.0005924361,0.001528049,0.0004184574],"domain_scores_gemma":[0.9843253,0.0115339,0.0007811371,0.001023229,0.001949314,0.0003871947],"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":[0.000008294068,0.00003603455,0.0002568214,0.00007005314,0.00001578741,0.00002267181,0.0001961153,0.01401701,0.00002709545,0.9782569,0.001968609,0.005124595],"study_design_scores_gemma":[0.00002505634,0.00003061254,0.0001243155,0.00009358687,0.00001451181,0.00002077784,0.0001120192,0.05279203,0.00008422742,0.9344803,0.01221156,0.00001100668],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.009907967,0.001712816,0.8585519,0.01659132,0.0003414743,0.0004836648,0.00071992,0.0002903501,0.1114005],"genre_scores_gemma":[0.6758206,0.003231082,0.2915302,0.002801924,0.0003847582,0.003342184,0.0008921719,0.0001599006,0.02183715],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.01036225,"threshold_uncertainty_score":0.04984343,"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","authors":[{"name":"Joan LaFrance","is_ca":false},{"name":"Richard Nichols","is_ca":false}],"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.3747783,0.001557079,0.001969939,0.01079155,0.01388484,0.02329494,0.006419277,0.005635799,0.00175719],"category_scores_gemma":[0.1820413,0.0009820202,0.001635619,0.005210307,0.05415861,0.01864165,0.01919183,0.009775716,0.0003761933],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.04280109,"about_ca_system_score_gemma":0.07545284,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02529006,"about_ca_topic_score_gemma":0.02750646,"domain_scores_codex":[0.5482283,0.3949724,0.01538564,0.006328452,0.02971142,0.005373669],"domain_scores_gemma":[0.7037707,0.2032652,0.009856616,0.01120318,0.06604079,0.005863505],"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":[0.00006387779,0.0001996708,0.002282284,0.002049425,0.0001218555,0.0002691736,0.1094218,0.00253095,0.000559381,0.7893184,0.005334386,0.08784875],"study_design_scores_gemma":[0.000115467,0.000394034,0.003515326,0.01128692,0.0002830648,0.000423618,0.1280249,0.01112096,0.001880194,0.6758608,0.1668267,0.0002679646],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03198801,0.01176455,0.6235296,0.1099531,0.001322717,0.007447339,0.0001216114,0.0004571814,0.213416],"genre_scores_gemma":[0.5588014,0.002952785,0.4223841,0.006304051,0.0003189197,0.004826165,0.00007735299,0.0001222876,0.004212955],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3747783,"threshold_uncertainty_score":0,"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":"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","authors":[{"name":"John Mayne","is_ca":false}],"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1084694,0.002346028,0.004492797,0.010102,0.003352775,0.01468983,0.006293539,0.004998704,0.007232175],"category_scores_gemma":[0.1973201,0.001457148,0.004341368,0.007173288,0.01979185,0.01634277,0.007101709,0.01033812,0.001083904],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01340907,"about_ca_system_score_gemma":0.01178513,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003745678,"about_ca_topic_score_gemma":0.00324727,"domain_scores_codex":[0.8728059,0.1035499,0.003953888,0.006292882,0.01207273,0.001324673],"domain_scores_gemma":[0.656586,0.2915028,0.0104848,0.02353567,0.01611905,0.001771698],"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":[0.00003776334,0.0001049695,0.001520911,0.0007335225,0.0003193384,0.0000520096,0.001096368,0.02325902,0.00008197864,0.9404117,0.002155964,0.03022646],"study_design_scores_gemma":[0.00003153388,0.00006049106,0.0004331452,0.0005009061,0.0000540279,0.00001689514,0.0004266233,0.0363926,0.0001435221,0.9575073,0.004401489,0.00003148466],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.006003807,0.002020127,0.9592777,0.01218908,0.0003603019,0.0006774827,0.0002304116,0.0003110449,0.01893001],"genre_scores_gemma":[0.2702892,0.002102942,0.720706,0.00158615,0.0003486913,0.003418203,0.0004119287,0.0001773691,0.0009595199],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1084694,"threshold_uncertainty_score":0.5736481,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W47101223","doi":"10.3138/cjpe.0015.003","title":"Predictors of Educator’s Valuing of Systematic Inquiry in Schools","year":2001,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Educational Assessment and Improvement","field":"Decision Sciences","cited_by":116,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Toronto; University of Ottawa","funders":"","keywords":"Psychology; Variance (accounting); Relevance (law); Exploratory research; Medical education; Value (mathematics); Medicine; Sociology; Social science","authors":[{"name":"J. Bradley Cousins","is_ca":true},{"name":"Cheryl Walker","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3838212522965732,"gpt":0.506814129178217,"spread":0.1229928768816438,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01096568,0.0001397941,0.0002652697,0.001555557,0.0006980893,0.002240427,0.0004219212,0.0007267608,0.002672177],"category_scores_gemma":[0.05897043,0.0002986111,0.0002997241,0.0009045549,0.001186313,0.0009722494,0.001244044,0.001316664,0.0003379522],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009148808,"about_ca_system_score_gemma":0.001587975,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002023143,"about_ca_topic_score_gemma":0.003198913,"domain_scores_codex":[0.9929919,0.003448092,0.0008494226,0.0003037679,0.001874199,0.0005325785],"domain_scores_gemma":[0.8952266,0.05177227,0.02860709,0.003098314,0.01244104,0.008854679],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00008498404,0.0001552528,0.9811998,0.00008524428,0.00002625855,0.0001184935,0.009723896,0.00007293861,0.0004947356,0.0001247879,0.0001860797,0.007727428],"study_design_scores_gemma":[0.0000156177,0.0003211644,0.9630386,0.0001591079,0.00002600029,0.0003383119,0.03259303,0.0006337345,0.0006177339,0.0003157001,0.001913669,0.00002736314],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.998548,0.00008412321,0.0001827384,0.0002136734,0.000005013862,0.00001340436,0.00002581118,0.000004274537,0.0009230237],"genre_scores_gemma":[0.9995219,0.00006759311,0.0001776983,0.00003416757,0.000003513508,0.000008552326,0.00002533562,0.000001634687,0.0001596882],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01096568,"threshold_uncertainty_score":0.05799282,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366450241","doi":"10.3138/cjpe.024.007","title":"Richard A. Krueger and Mary Anne Casey. (2009). <i>Focus Groups: A Practical Guide for Applied Research</i> (4th ed.). Thousand Oaks, CA: Sage. 205 pages, plus preface, acknowledgements, references, and indices.","year":2009,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Focus Groups and Qualitative Methods","field":"Social Sciences","cited_by":114,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"SAGE; Psychology; Focus (optics); Sociology; Library science; Gerontology; Computer science; Medicine","authors":[{"name":"Robert G. Brunger","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1941600388254136,"gpt":0.4936611830970598,"spread":0.2995011442716462,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008913502,0.0008998857,0.0006525559,0.003058212,0.002912466,0.002650559,0.001915308,0.002661467,0.03183458],"category_scores_gemma":[0.03847759,0.001312968,0.0005610997,0.002413908,0.001853803,0.006993175,0.002036226,0.007032067,0.0242021],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001479656,"about_ca_system_score_gemma":0.005406746,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02449945,"about_ca_topic_score_gemma":0.07457152,"domain_scores_codex":[0.9965259,0.001224317,0.0002868778,0.0002471692,0.001608596,0.0001070524],"domain_scores_gemma":[0.9789047,0.009948602,0.0009228933,0.0005156863,0.008809623,0.0008985061],"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":[0.0000434136,0.00003925991,0.0003938139,0.0009708312,0.000006613577,0.0000932122,0.001712683,0.00005617859,0.0003536421,0.003623582,0.7649599,0.2277468],"study_design_scores_gemma":[0.00003961624,0.00009415314,0.003653345,0.004560464,0.00003439936,0.000681225,0.003768296,0.0001668789,0.001160837,0.01499474,0.970742,0.0001040413],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.004099682,0.4056312,0.1092291,0.3331401,0.02678734,0.001846278,0.002981445,0.002410341,0.1138746],"genre_scores_gemma":[0.0338752,0.5154502,0.1866544,0.0627459,0.003895704,0.002110624,0.001656886,0.001539386,0.1920717],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.03183458,"threshold_uncertainty_score":0.1064972,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366725272","doi":"10.3138/cjpe.19.006","title":"Integrating Evaluative Inquiry into the Organizational Culture: A Review and Synthesis of the Knowledge Base","year":2004,"lang":"en","type":"review","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":112,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Ottawa","funders":"","keywords":"Knowledge management; Organizational culture; Organizational learning; Knowledge base; Conceptual framework; Empirical research; Sociology; Work (physics); Management science; Epistemology; Political science; Computer science; Public relations; Social science; Engineering","authors":[{"name":"J. Bradley Cousins","is_ca":false},{"name":"Swee C. Goh","is_ca":false},{"name":"Shannon Clark","is_ca":true},{"name":"Linda E. Lee","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3839810064574307,"gpt":0.5592321866859272,"spread":0.1752511802284965,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02876894,0.001065346,0.003349199,0.0222238,0.00165537,0.007738108,0.002218887,0.002386074,0.00228851],"category_scores_gemma":[0.05061668,0.0009303041,0.0009372753,0.03247645,0.004943326,0.006564529,0.002433405,0.002180593,0.0004883645],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007004973,"about_ca_system_score_gemma":0.02271881,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01451784,"about_ca_topic_score_gemma":0.03136452,"domain_scores_codex":[0.9869003,0.00662685,0.002037454,0.0009251973,0.003205585,0.0003047224],"domain_scores_gemma":[0.8633461,0.1188051,0.004817211,0.00159136,0.01081783,0.0006223974],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0000498778,0.000101274,0.001739559,0.09200201,0.0003267407,0.0002279693,0.005080316,0.0005637713,0.0003943204,0.01003502,0.004812562,0.8846666],"study_design_scores_gemma":[0.00004231523,0.0002414873,0.01518909,0.4662342,0.001534305,0.001426686,0.02176013,0.0009242818,0.001468892,0.02409323,0.4669188,0.0001666223],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0005974466,0.9958228,0.001025896,0.001255592,0.0001141284,0.00001881604,0.00001585614,0.000005158822,0.001144299],"genre_scores_gemma":[0.0117624,0.9857348,0.001761461,0.0004069414,0.0001308943,0.00006545414,0.0000223243,0.000004987275,0.0001106727],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.02876894,"threshold_uncertainty_score":0.1521465,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1826075361","doi":"10.3138/cjpe.22.007","title":"Participatory Impact Pathways Analysis: A Practical Application of Program Theory in Research-for-Development","year":2007,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":104,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Logic model; Citizen journalism; Food security; Computer science; Poverty; Narrative; Theory of change; Management science; Knowledge management; Process management; Sociology; Business; Engineering; Economics; Economic growth; Social science","authors":[{"name":"Boru Douthwaite","is_ca":false},{"name":"Sophie Alvarez","is_ca":false},{"name":"Simón Cook","is_ca":false},{"name":"Rick Davies","is_ca":false},{"name":"P. Reeja George","is_ca":false},{"name":"John J. Howell","is_ca":false},{"name":"Ronald Mackay","is_ca":true},{"name":"Jorge Rubiano","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.815960147008348,"gpt":0.6979548155552535,"spread":0.1180053314530944,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05444057,0.001413311,0.0009123429,0.007600152,0.00579289,0.008182487,0.002828105,0.002042426,0.01352304],"category_scores_gemma":[0.07623443,0.0008961734,0.001643293,0.008722563,0.008116969,0.012006,0.00779637,0.002644419,0.0008681473],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008496668,"about_ca_system_score_gemma":0.0150187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00625913,"about_ca_topic_score_gemma":0.007040315,"domain_scores_codex":[0.9564533,0.03812346,0.0009957644,0.001366571,0.00245731,0.0006035994],"domain_scores_gemma":[0.925275,0.0616666,0.002302702,0.005291001,0.004701562,0.0007631269],"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":[0.00009988418,0.00019928,0.002969801,0.0008878586,0.0001099094,0.0004323146,0.0193293,0.02515955,0.0004455977,0.7879646,0.003943196,0.1584587],"study_design_scores_gemma":[0.00008732102,0.0001221957,0.000567964,0.0006397751,0.00006245517,0.0002294738,0.01450076,0.05577021,0.0009986577,0.8849876,0.04197846,0.00005514315],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.009603402,0.0001256318,0.9651693,0.003145667,0.00006421066,0.001738979,0.0003440969,0.0003303292,0.01947837],"genre_scores_gemma":[0.1620219,0.0001885197,0.8325685,0.0001412858,0.00001499952,0.002919939,0.00022581,0.00007640878,0.001842572],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.05444057,"threshold_uncertainty_score":0.2879127,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4255004666","doi":"10.3138/cjpe.30.3.05","title":"Decolonizing and Indigenizing Evaluation Practice in Africa: Toward African Relational Evaluation Approaches","year":2016,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":103,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Indigenization; Decolonization; Sociology; African culture; Metaphor; Epistemology; African philosophy; Political science; Gender studies; Anthropology; Linguistics; Philosophy","authors":[{"name":"Bagele Chilisa","is_ca":false},{"name":"Thenjiwe Emily Major","is_ca":false},{"name":"Michael Gaotlhobogwe","is_ca":false},{"name":"Hildah L. Mokgolodi","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6556148384806202,"gpt":0.5116189627666728,"spread":0.1439958757139473,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06342129,0.000691445,0.000635786,0.005040349,0.008657922,0.01720361,0.00191351,0.002848372,0.00313949],"category_scores_gemma":[0.03943826,0.0005077822,0.0004613839,0.003317958,0.04850417,0.01722864,0.0154625,0.006204399,0.0002887083],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01479562,"about_ca_system_score_gemma":0.01591616,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00528395,"about_ca_topic_score_gemma":0.00637418,"domain_scores_codex":[0.9534067,0.039169,0.001109646,0.001406521,0.003236525,0.001671546],"domain_scores_gemma":[0.961737,0.0267473,0.003033087,0.002420347,0.004903047,0.001159205],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002049405,0.00003106973,0.0007934612,0.000163582,0.000007773087,0.00008637406,0.04340147,0.0003290862,0.0001736751,0.9345021,0.0008001955,0.01969079],"study_design_scores_gemma":[0.00004821129,0.00008828009,0.001706736,0.003017824,0.0000486728,0.0003855015,0.137583,0.004551932,0.001733406,0.7534108,0.09737368,0.00005194541],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1449884,0.01241398,0.3128234,0.1287057,0.0005786783,0.0006553577,0.00004573797,0.0001513257,0.3996374],"genre_scores_gemma":[0.9542293,0.002430357,0.03622692,0.002449288,0.00006750049,0.0002940976,0.00001035659,0.00004629045,0.004246],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.06342129,"threshold_uncertainty_score":0.3354079,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366448482","doi":"10.3138/cjpe.020.005","title":"The Delphi Technique as a Method for Increasing Inclusion in the Evaluation Process","year":2005,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":80,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Delphi method; Delphi; Compromise; Inclusion (mineral); Process (computing); Set (abstract data type); Evaluation methods; Psychology; Economic Justice; Representation (politics); Public relations; Management science; Sociology; Engineering ethics; Computer science; Social psychology; Political science; Engineering; Social science; Artificial intelligence; Law","authors":[{"name":"Christina A. Christie","is_ca":false},{"name":"Eric Barela","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3689523270921288,"gpt":0.6178187315297932,"spread":0.2488664044376644,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3784713,0.002256457,0.002065032,0.01004077,0.007172154,0.006095967,0.003204141,0.002380268,0.01040698],"category_scores_gemma":[0.3295668,0.001876949,0.001315588,0.006516421,0.009790544,0.006586414,0.01576302,0.006411629,0.002252146],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004406443,"about_ca_system_score_gemma":0.01314928,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001280438,"about_ca_topic_score_gemma":0.002470285,"domain_scores_codex":[0.34213,0.6042072,0.01645138,0.003658949,0.03163965,0.001912832],"domain_scores_gemma":[0.5753041,0.3326742,0.009167216,0.02245212,0.05754546,0.002856868],"domain_codex":"methods","domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00120687,0.0008302667,0.002169311,0.005599879,0.0002046603,0.0008176289,0.2907682,0.002910637,0.01730672,0.07881551,0.01720865,0.5821616],"study_design_scores_gemma":[0.002313374,0.006126159,0.01368456,0.01712961,0.0004158072,0.002469053,0.2202808,0.03464665,0.03266539,0.2772962,0.3914678,0.001504485],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.03981146,0.0005735795,0.8618709,0.004977835,0.0008772229,0.05976003,0.0001942406,0.0005953332,0.03133936],"genre_scores_gemma":[0.07192934,0.0003604925,0.8722822,0.000637926,0.0001371416,0.05204704,0.00005236096,0.0001583515,0.002395155],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3784713,"threshold_uncertainty_score":0.7664556,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2289811664","doi":"10.3138/cjpe.20.001","title":"Evaluability Assessment of a Survivors of Torture Program: Lessons Learned","year":2005,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Interpreting and Communication in Healthcare","field":"Health Professions","cited_by":71,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Torture; Accountability; Logic model; Program evaluation; Process (computing); Political science; Psychology; Public relations; Public administration; Computer science; Human rights; Law","authors":[{"name":"Wilfreda E. Thurston","is_ca":true},{"name":"Anila Ramaliu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3722469645419353,"gpt":0.6063679190213934,"spread":0.2341209544794581,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08597425,0.0005663986,0.000676713,0.002719712,0.002050648,0.003377628,0.001773159,0.0008466928,0.00171037],"category_scores_gemma":[0.1176846,0.0002435004,0.0009137553,0.001307209,0.001451685,0.004222832,0.003314207,0.002141378,0.0001598822],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009187624,"about_ca_system_score_gemma":0.01688498,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01611422,"about_ca_topic_score_gemma":0.01939568,"domain_scores_codex":[0.9430974,0.04075195,0.002983178,0.001054294,0.01021653,0.001896669],"domain_scores_gemma":[0.8577762,0.07911028,0.006825319,0.005012904,0.04718276,0.00409257],"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":[0.0006487241,0.005231861,0.1178483,0.00186194,0.0001761051,0.0004701357,0.05050876,0.004880302,0.002059978,0.00867826,0.006581613,0.8010541],"study_design_scores_gemma":[0.000728245,0.01448669,0.5310482,0.008185885,0.000501894,0.001282879,0.2289081,0.0744729,0.02695925,0.04710023,0.06582727,0.0004984282],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8887905,0.001195443,0.06423614,0.01474534,0.0001884144,0.005193666,0.000178679,0.0003455205,0.02512634],"genre_scores_gemma":[0.9419165,0.000349928,0.05456213,0.0004055598,0.00003636349,0.001295997,0.0001172726,0.00003210019,0.001284232],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08597425,"threshold_uncertainty_score":0.4546807,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W79165327","doi":"10.3138/cjpe.018.001","title":"They’re Happy, but Did They Make a Difference? Applying Kirkpatrick’s Framework to the Evaluation of a National Leadership Program","year":2003,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Human Resource Development and Performance Evaluation","field":"Psychology","cited_by":59,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Formative assessment; Process (computing); Program evaluation; Leadership development; Psychology; Investment (military); Management science; Process management; Medical education; Political science; Business; Computer science; Public relations; Pedagogy; Engineering; Public administration; Medicine","authors":[{"name":"Scott McLean","is_ca":true},{"name":"Gwenna Moss","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3557958721915049,"gpt":0.4343449716571408,"spread":0.07854909946563587,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1305081,0.0009376201,0.0012669,0.005773663,0.01015267,0.01140008,0.003644972,0.002713944,0.003201783],"category_scores_gemma":[0.1000214,0.0005541301,0.001165583,0.003113857,0.01695278,0.009591647,0.009304842,0.004469697,0.0003942728],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03357432,"about_ca_system_score_gemma":0.05238308,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0228134,"about_ca_topic_score_gemma":0.03786518,"domain_scores_codex":[0.7296565,0.2363615,0.004463406,0.002491489,0.02227394,0.004753201],"domain_scores_gemma":[0.9307827,0.04484203,0.004200274,0.001919788,0.01556579,0.002689229],"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":[0.000605369,0.003362342,0.0244253,0.003665946,0.000262879,0.001223537,0.1706737,0.005394479,0.00124147,0.3319835,0.02185132,0.43531],"study_design_scores_gemma":[0.0008187901,0.003583083,0.04144494,0.01157049,0.0003938804,0.001446205,0.5405111,0.0273939,0.006959056,0.236831,0.1284771,0.0005705239],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.3986435,0.005319128,0.1570966,0.1064034,0.0009275795,0.01370627,0.0002821436,0.0002061674,0.3174151],"genre_scores_gemma":[0.9087812,0.0009091906,0.08025724,0.002278438,0.00003031607,0.004652699,0.00005309676,0.00003664174,0.003001176],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1305081,"threshold_uncertainty_score":0.6902012,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2188387579","doi":"10.3138/cjpe.019.002","title":"Reporting on Outcomes: Setting Performance Expectations and Telling Performance Stories","year":2004,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":57,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"Government of Canada","funders":"","keywords":"Outcome (game theory); Chart; Public relations; Actuarial science; Business; Psychology; Political science; Economics; Microeconomics; Statistics","authors":[{"name":"John Mayne","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4090562215533984,"gpt":0.5347074822731473,"spread":0.125651260719749,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1349927,0.001200937,0.0005231036,0.005044462,0.004015363,0.0168375,0.002463764,0.00278672,0.00431065],"category_scores_gemma":[0.2903183,0.0006777737,0.000599569,0.003035377,0.008964244,0.02110801,0.007623678,0.004336548,0.001083323],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0105935,"about_ca_system_score_gemma":0.01361899,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0081196,"about_ca_topic_score_gemma":0.005195632,"domain_scores_codex":[0.8623679,0.1056275,0.009063789,0.002035272,0.01860372,0.00230192],"domain_scores_gemma":[0.5885613,0.2771844,0.04809887,0.0242396,0.05332229,0.008593654],"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":[0.0007454038,0.0005456429,0.04235562,0.001421687,0.0001023254,0.0007180657,0.06965319,0.0112829,0.001589133,0.3031508,0.06579766,0.5026376],"study_design_scores_gemma":[0.000239328,0.002027341,0.04055053,0.006001134,0.0002479261,0.0008490882,0.1119483,0.05691677,0.02824856,0.4262147,0.3257308,0.001025604],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1519569,0.001750576,0.5173674,0.1404638,0.001242352,0.002143795,0.002242789,0.003244074,0.1795882],"genre_scores_gemma":[0.799867,0.001302649,0.1901894,0.002151098,0.0003009189,0.001395942,0.0008092981,0.0002645376,0.003719117],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1349927,"threshold_uncertainty_score":0.713918,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4362681378","doi":"10.3138/cjpe.18.001","title":"Evaluation and Research: Differences and Similarities","year":2003,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":57,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Generalization; Similarity (geometry); Relevance (law); Causality (physics); Psychology; Management science; Computer science; Knowledge management; Epistemology; Artificial intelligence; Political science","authors":[{"name":"Miri Levin‐Rozalis","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7713106731453503,"gpt":0.617815743068949,"spread":0.1534949300764012,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1230626,0.0006308706,0.002239551,0.01320505,0.003014016,0.02680122,0.002140673,0.00512289,0.002320044],"category_scores_gemma":[0.134208,0.000998002,0.0009227682,0.01083957,0.05526543,0.02114664,0.01195232,0.006009002,0.0007117789],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01231155,"about_ca_system_score_gemma":0.01036201,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002903111,"about_ca_topic_score_gemma":0.002277982,"domain_scores_codex":[0.7925403,0.1469838,0.01403254,0.006481189,0.03738457,0.002577648],"domain_scores_gemma":[0.7568157,0.1918096,0.007609081,0.01476076,0.02484207,0.004162726],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001054288,0.0000955439,0.004842022,0.001920454,0.0001513335,0.0002119237,0.03028168,0.0005145104,0.0003077063,0.7832203,0.003368783,0.1749803],"study_design_scores_gemma":[0.00009476535,0.0002516948,0.01385329,0.006699387,0.0001145257,0.001184279,0.03246838,0.001955837,0.000593629,0.7892709,0.1533725,0.0001408182],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.05236108,0.369868,0.1683347,0.1236251,0.004369985,0.000952032,0.0001629603,0.0005296073,0.2797965],"genre_scores_gemma":[0.8347474,0.05263366,0.08834735,0.01184738,0.002857121,0.001561834,0.0001193724,0.0002777705,0.00760809],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8769374,"threshold_uncertainty_score":0.6508248,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2996731041","doi":"10.3138/cjpe.68004","title":"Revisiting Contribution Analysis","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":48,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Operationalization; Set (abstract data type); Epistemology; Positive economics; Causal analysis; Narrative; Key (lock); Sociology; Economics; Computer science; Econometrics; Philosophy","authors":[{"name":"John Mayne","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2744377398669307,"gpt":0.5383881020311411,"spread":0.2639503621642104,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1447931,0.002320757,0.001868114,0.01158522,0.008483631,0.01547138,0.007409976,0.005425608,0.01921032],"category_scores_gemma":[0.2524429,0.001014044,0.003270308,0.009527015,0.02731362,0.03116143,0.01638599,0.01054615,0.003145122],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01366427,"about_ca_system_score_gemma":0.01564497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005424893,"about_ca_topic_score_gemma":0.004099478,"domain_scores_codex":[0.8375855,0.1138888,0.006426147,0.01139001,0.02726659,0.003442959],"domain_scores_gemma":[0.6595421,0.252746,0.01150673,0.02687477,0.04694675,0.00238366],"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":[0.00004779848,0.00003466893,0.001469404,0.0006378937,0.0001078848,0.0001320725,0.008657114,0.0009049669,0.0000830348,0.932396,0.00846075,0.04706848],"study_design_scores_gemma":[0.00002960015,0.0000300999,0.0007217259,0.001521508,0.00007537082,0.0001285475,0.004065052,0.004394604,0.0004661818,0.9052663,0.08326541,0.00003554532],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01758275,0.007140194,0.7351795,0.06402627,0.005026862,0.0009416489,0.0005884747,0.000612558,0.1689017],"genre_scores_gemma":[0.656426,0.004501286,0.2982156,0.01104109,0.002952911,0.002055189,0.0006152268,0.0008889288,0.0233037],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1447931,"threshold_uncertainty_score":0.7657484,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2163998785","doi":"10.3138/cjpe.0023.002","title":"Organizational Capacity To Do and Use Evaluation: Results of a Pan-Canadian Survey of Evaluators","year":2009,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":48,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"National Research Council Canada; Atlantic Canada Opportunities Agency; University of Ottawa","funders":"","keywords":"Respondent; Psychology; Stakeholder; Knowledge management; Organizational learning; Context (archaeology); Descriptive statistics; Exploratory research; Social psychology; Applied psychology; Public relations; Sociology; Political science; Computer science; Social science; Geography","authors":[{"name":"J. Bradley Cousins","is_ca":true},{"name":"Catherine Elliott","is_ca":true},{"name":"Courtney Amo","is_ca":true},{"name":"Isabelle Bourgeois","is_ca":true},{"name":"Jill Anne Chouinard","is_ca":true},{"name":"Swee C. Goh","is_ca":true},{"name":"Robert Lahey","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4439444040034558,"gpt":0.5021568546325215,"spread":0.05821245062906572,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008102149,0.0002669178,0.0003081682,0.002756606,0.003372483,0.001843363,0.0009595311,0.0003657483,0.001780723],"category_scores_gemma":[0.01954207,0.0003302573,0.0003625996,0.004270488,0.001425409,0.0006997061,0.001709688,0.0007356384,0.0002770408],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01543395,"about_ca_system_score_gemma":0.02788763,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9583757,"about_ca_topic_score_gemma":0.9743662,"domain_scores_codex":[0.9955382,0.0007834628,0.0003623042,0.0003016442,0.002096335,0.0009180834],"domain_scores_gemma":[0.9695155,0.004106346,0.003643461,0.0008197826,0.0184449,0.00346998],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001014621,0.00009902543,0.9433569,0.0001586556,0.00002806462,0.00007688154,0.03081155,0.0001004559,0.0003241127,0.0002069326,0.002242894,0.02249311],"study_design_scores_gemma":[0.00000409918,0.00004659153,0.9728845,0.00006241631,0.000008596255,0.00002518115,0.02385719,0.0001454379,0.0001183802,0.00002088716,0.002806386,0.00002035806],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9942972,0.0003338875,0.0001478588,0.0003471571,0.000005852742,0.0001034457,0.0008640868,0.00001173896,0.003888765],"genre_scores_gemma":[0.9976941,0.000415489,0.0002612509,0.0000946216,0.000002430256,0.00004690313,0.0004300957,0.000007193946,0.001047941],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9918978,"threshold_uncertainty_score":0.1119816,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2161163161","doi":"","title":"Knowlton, L.W. & Phillips, C.C. (2013). The Logic Model Guidebook: Better Strategies for Great Results (2nd Edition), Thousand Oaks, CA: Sage.","year":2012,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Intelligent Tutoring Systems and Adaptive Learning","field":"Computer Science","cited_by":47,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"SAGE; Physics; Nuclear physics","authors":[{"name":"Sandra Sellick","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1031034195254785,"gpt":0.3427688452820291,"spread":0.2396654257565506,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004340908,0.001067573,0.0006103852,0.004191337,0.001482081,0.004181718,0.001967262,0.001833437,0.0518028],"category_scores_gemma":[0.0180797,0.0009440801,0.0006725726,0.002889442,0.001813213,0.007807839,0.001244492,0.003922414,0.0303305],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001732047,"about_ca_system_score_gemma":0.004696794,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01769212,"about_ca_topic_score_gemma":0.04569193,"domain_scores_codex":[0.9989085,0.0002165132,0.0001370141,0.00008972723,0.0005970036,0.00005112035],"domain_scores_gemma":[0.9882863,0.006330257,0.0005307983,0.000558498,0.003884505,0.0004095926],"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":[0.00003748181,0.0000461027,0.0005495345,0.0007391749,0.00001275268,0.00007027383,0.0006413682,0.0001623003,0.0005166056,0.00933746,0.5875308,0.4003562],"study_design_scores_gemma":[0.00001851672,0.00006611487,0.004184878,0.002087218,0.00006523509,0.0004384828,0.0008276664,0.0002931737,0.001875394,0.03871215,0.9513782,0.00005308089],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.005660751,0.3394636,0.1509762,0.08083143,0.009777884,0.0005301615,0.01115577,0.004713616,0.3968906],"genre_scores_gemma":[0.05898078,0.3691204,0.1965906,0.01384417,0.002456221,0.0009349662,0.007444132,0.002541128,0.3480875],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.0518028,"threshold_uncertainty_score":0.1732976,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1490602178","doi":"10.3138/cjpe.0028.005","title":"A Professional Grounding and History of the Development and Formal Use of Evaluator Competencies","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":44,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Professionalization; Viewpoints; Credentialing; Engineering ethics; Value (mathematics); Professional association; Field (mathematics); Foundation (evidence); Professional development; Psychology; Political science; Sociology; Pedagogy; Public relations; Engineering; Computer science; Social science; Law","authors":[{"name":"Yuanjing Wilcox","is_ca":false},{"name":"Jean A. King","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3985154956302613,"gpt":0.4680725057001939,"spread":0.06955701006993253,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03484179,0.0003886809,0.0004695023,0.006663883,0.004495952,0.01019698,0.001490651,0.003329294,0.005558582],"category_scores_gemma":[0.04482992,0.0005540142,0.0003530675,0.004432496,0.02061859,0.008838331,0.005377891,0.007634829,0.0007908993],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01818502,"about_ca_system_score_gemma":0.02251426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01935649,"about_ca_topic_score_gemma":0.0219609,"domain_scores_codex":[0.9823982,0.008765443,0.001359579,0.001732117,0.0047727,0.0009719714],"domain_scores_gemma":[0.9059036,0.0635317,0.003778006,0.005517366,0.01829577,0.002973464],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.00006180116,0.00009406816,0.003080295,0.00109351,0.00001230946,0.0001878199,0.01480077,0.0003329229,0.0005045905,0.5214534,0.03279405,0.4255845],"study_design_scores_gemma":[0.000009665611,0.00008827356,0.006317279,0.004712007,0.00001181761,0.0005055464,0.00522771,0.0005125223,0.001570483,0.05422748,0.9267505,0.00006675527],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.02979035,0.3561368,0.04838993,0.2634225,0.009983225,0.0002134647,0.0002908113,0.0002591069,0.291514],"genre_scores_gemma":[0.6628987,0.1903663,0.03880705,0.03368633,0.006352142,0.0003698505,0.0002686396,0.0004917656,0.06675916],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03484179,"threshold_uncertainty_score":0.1842632,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W166307397","doi":"10.3138/cjpe.24.001","title":"Building an Evaluative Culture: The Key to Effective Evaluation and Results Management","year":2009,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Human Resource Development and Performance Evaluation","field":"Psychology","cited_by":44,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"Rogers Communications (Canada)","funders":"","keywords":"Key (lock); Organizational culture; Knowledge management; Order (exchange); Empirical research; Business; Sociology; Computer science; Public relations; Psychology; Process management; Epistemology; Political science; Computer security","authors":[{"name":"John Mayne","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0790348188226643,"gpt":0.4286523090145266,"spread":0.3496174901918623,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4083444,0.001264998,0.002784829,0.008436148,0.02180662,0.04972962,0.004028688,0.007805128,0.003534705],"category_scores_gemma":[0.3721286,0.001959202,0.001306142,0.005080859,0.06008661,0.02564043,0.02529769,0.02411318,0.001933962],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02014814,"about_ca_system_score_gemma":0.0875332,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005350514,"about_ca_topic_score_gemma":0.003651452,"domain_scores_codex":[0.5287412,0.3425262,0.02534611,0.01291271,0.08188657,0.008587223],"domain_scores_gemma":[0.3377831,0.3703811,0.04917474,0.06458131,0.1296778,0.04840185],"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":[0.0001738148,0.00133031,0.01348296,0.002402638,0.0004199494,0.0003738313,0.1037996,0.001979186,0.002487469,0.4339129,0.06092662,0.3787107],"study_design_scores_gemma":[0.0002192113,0.0008037887,0.02065365,0.01334621,0.0002619516,0.0007544166,0.1033916,0.00357049,0.004384517,0.5706697,0.2811348,0.000809624],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.03677788,0.009692837,0.2567933,0.5640141,0.003884705,0.002004259,0.00009315101,0.001225976,0.1255138],"genre_scores_gemma":[0.6578277,0.006155869,0.2789524,0.04445632,0.002632025,0.002602261,0.0001129921,0.0006898282,0.006570551],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4083444,"threshold_uncertainty_score":0.7296166,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2732819684","doi":"10.3138/cjpe.20.006","title":"Evaluator Competencies in University-Based Evaluation Training Programs","year":2005,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":44,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Credentialing; Accreditation; Licensure; Certificate; Taxonomy (biology); Competence (human resources); Computer science; Medical education; Psychology; Medicine","authors":[{"name":"Laurie Stevahn","is_ca":false},{"name":"Jean A. King","is_ca":false},{"name":"Gail Ghere","is_ca":false},{"name":"Jane Minnema","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4101080033795569,"gpt":0.5034090358417598,"spread":0.09330103246220289,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08040453,0.0004263094,0.0005821008,0.003849098,0.003010608,0.00664692,0.00200669,0.001812212,0.005537992],"category_scores_gemma":[0.1060068,0.0005074252,0.0003740812,0.001865434,0.004696558,0.006581387,0.008502889,0.003306116,0.0009428078],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007107804,"about_ca_system_score_gemma":0.0285188,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00317037,"about_ca_topic_score_gemma":0.00543895,"domain_scores_codex":[0.9176633,0.06173225,0.005109314,0.001711172,0.01125878,0.002525131],"domain_scores_gemma":[0.8736761,0.06451932,0.008238139,0.008433678,0.03449486,0.01063786],"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":[0.0003020623,0.001398705,0.07055115,0.001735651,0.0000470727,0.0002467972,0.03212453,0.005016357,0.001919307,0.2650756,0.02259483,0.5989879],"study_design_scores_gemma":[0.0003505109,0.002616148,0.1640207,0.01071468,0.0001215314,0.001817152,0.06441143,0.04918636,0.02128867,0.3690838,0.315919,0.0004702157],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2773012,0.004267698,0.4884109,0.02456981,0.0007547644,0.003616709,0.000315224,0.001427205,0.1993366],"genre_scores_gemma":[0.8170344,0.001121936,0.1700388,0.001308499,0.00007797637,0.001584074,0.0001944074,0.0001146894,0.008525131],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.08040453,"threshold_uncertainty_score":0.4252248,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2274776964","doi":"10.3138/cjpe.022.003","title":"Developing a Tool to Measure Knowledge Exchange Outcomes","year":2007,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":41,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Waterloo","funders":"","keywords":"USable; Knowledge management; Measure (data warehouse); Process (computing); Computer science; Best practice; Reliability (semiconductor); Point (geometry); Process management; Business; World Wide Web; Data mining","authors":[{"name":"Kelly Skinner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8485447477899826,"gpt":0.719432757191686,"spread":0.1291119905982966,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.05321929,0.000826271,0.001117886,0.01123282,0.001300211,0.003442466,0.001590063,0.001278865,0.005604278],"category_scores_gemma":[0.1302316,0.0004749022,0.001938392,0.006879717,0.001044366,0.005172258,0.003747246,0.00193441,0.001631276],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003184498,"about_ca_system_score_gemma":0.005311815,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001120457,"about_ca_topic_score_gemma":0.001107404,"domain_scores_codex":[0.9584577,0.01884922,0.007548389,0.001268491,0.01274073,0.001135516],"domain_scores_gemma":[0.8351029,0.1105197,0.01598569,0.004806349,0.03178036,0.00180506],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0005749362,0.002042865,0.1973537,0.001941677,0.0003232473,0.00009049097,0.006583644,0.002340874,0.001761162,0.01023039,0.01259019,0.7641668],"study_design_scores_gemma":[0.001105036,0.009350845,0.7623624,0.004851826,0.0009598624,0.000656968,0.02151482,0.03845103,0.03328725,0.03729963,0.08946447,0.000695846],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6067127,0.0008142761,0.2760143,0.003594269,0.0007518466,0.02881121,0.00916543,0.003600551,0.07053541],"genre_scores_gemma":[0.5374244,0.000439104,0.4153733,0.0004265592,0.00007375181,0.03731911,0.00499807,0.0001680981,0.003777594],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9467807,"threshold_uncertainty_score":0.2814538,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2994626447","doi":"10.3138/cjpe.68444","title":"Increasing Cultural Competence in Support of Indigenous-Led Evaluation: A Necessary Step toward Indigenous-Led Evaluation","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":41,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Indigenous; Aotearoa; Competence (human resources); Traditional knowledge; Sociology; Political science; Psychology; Gender studies; Social psychology","authors":[{"name":"Nan Wehipeihana","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2345971341053808,"gpt":0.4965004891018339,"spread":0.2619033549964531,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3049125,0.0009244723,0.001701245,0.003500422,0.01597472,0.0220129,0.005766422,0.00518769,0.008296595],"category_scores_gemma":[0.3056246,0.001139059,0.001327863,0.002262598,0.02414819,0.01958111,0.03403426,0.01757581,0.00201259],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02819594,"about_ca_system_score_gemma":0.1785841,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03773409,"about_ca_topic_score_gemma":0.06172543,"domain_scores_codex":[0.6921414,0.229683,0.01138681,0.007740245,0.04740827,0.01164036],"domain_scores_gemma":[0.4520986,0.2641355,0.01823475,0.03846952,0.1714335,0.05562811],"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":[0.0001990297,0.00202115,0.01574977,0.003423309,0.0002438677,0.000879227,0.3734418,0.001234488,0.004584706,0.1034836,0.05446995,0.4402689],"study_design_scores_gemma":[0.0003362156,0.0006535451,0.02712517,0.01398591,0.0001753475,0.001167669,0.2338835,0.002574485,0.004625417,0.13085,0.584097,0.0005256581],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.05537521,0.006183899,0.1246591,0.6068597,0.00270026,0.005261755,0.00006341627,0.001011819,0.1978849],"genre_scores_gemma":[0.7194207,0.003547235,0.2018105,0.0554814,0.0008343394,0.0041026,0.00008899379,0.0005026979,0.01421153],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9718041,"threshold_uncertainty_score":0.8571666,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366721381","doi":"10.3138/cjpe.15.006","title":"Logic Models in Primary Care Reform: Navigating the Evaluation","year":2000,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"Dalhousie University","funders":"","keywords":"Primary care; Logic model; Nova scotia; Strengths and weaknesses; Quality (philosophy); Political science; Public administration; Process management; Business; Economic growth; Medicine; Psychology; Sociology; Economics; Family medicine","authors":[{"name":"Margo Rowan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4326891670577493,"gpt":0.5423916433560159,"spread":0.1097024762982666,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1106344,0.001470863,0.002330919,0.006006106,0.003911968,0.01743137,0.003593554,0.003900321,0.004961796],"category_scores_gemma":[0.2325443,0.001236795,0.001552818,0.005197979,0.0201925,0.02351141,0.007228509,0.007977723,0.0004679305],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02851271,"about_ca_system_score_gemma":0.03126024,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0313589,"about_ca_topic_score_gemma":0.02985687,"domain_scores_codex":[0.8376287,0.1490206,0.00250709,0.001702791,0.0076831,0.001457714],"domain_scores_gemma":[0.5665433,0.4003608,0.007638621,0.005654497,0.01716385,0.0026388],"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":[0.0001160553,0.0001244583,0.001519607,0.0005606399,0.00007074548,0.00009797405,0.001541618,0.03258289,0.00007774511,0.9154327,0.003032517,0.04484308],"study_design_scores_gemma":[0.00008100703,0.00005725738,0.0001711922,0.0005787227,0.00003127528,0.00002304015,0.001099263,0.05819997,0.0001613449,0.9340514,0.005511018,0.0000345283],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.035916,0.008623664,0.7875485,0.09559957,0.00033039,0.001063852,0.0003231409,0.0005332466,0.07006165],"genre_scores_gemma":[0.5495127,0.003693009,0.4388435,0.003692182,0.000262088,0.001810632,0.00023284,0.0001231265,0.001829963],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1106344,"threshold_uncertainty_score":0.5850974,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366452622","doi":"10.3138/cjpe.0026.005","title":"L’approche <i>Realist</i> à l’épreuve du réel de l’évaluation des programmes","year":2012,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"Centre Hospitalier de l’Université de Montréal; Université de Montréal","funders":"","keywords":"Epistemology; Context (archaeology); Valuation (finance); Psychological intervention; Outcome (game theory); Sociology; Psychology; GRASP; Mechanism (biology); Philosophy; Computer science; Economics; History","authors":[{"name":"Valéry Ridde","is_ca":true},{"name":"Émilie Robert","is_ca":true},{"name":"Anne Guichard","is_ca":false},{"name":"Pierre Blaise","is_ca":false},{"name":"Josefien van Olmen","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4946108818232891,"gpt":0.5147343735641988,"spread":0.02012349174090972,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3288297,0.001969506,0.002962323,0.006850967,0.004127048,0.02746774,0.009255858,0.01191222,0.009080827],"category_scores_gemma":[0.2482206,0.001298192,0.003803274,0.003934316,0.03007935,0.01966181,0.01086768,0.01600724,0.00187536],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02224774,"about_ca_system_score_gemma":0.02126961,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006955909,"about_ca_topic_score_gemma":0.008750296,"domain_scores_codex":[0.4982219,0.4232413,0.01298963,0.00967039,0.05312068,0.002756135],"domain_scores_gemma":[0.5944596,0.3088922,0.01462785,0.03682429,0.04290823,0.002287871],"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":[0.000288677,0.0003729081,0.001355286,0.004632104,0.0004514108,0.0001660416,0.008559674,0.006529707,0.0006950184,0.8758397,0.01015514,0.0909544],"study_design_scores_gemma":[0.0006814172,0.0007875515,0.002344279,0.01162401,0.000343409,0.0004861616,0.006311676,0.01929317,0.003749595,0.7206769,0.2334381,0.0002637798],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01700069,0.0183964,0.5717394,0.2949669,0.003344427,0.00201972,0.0003340612,0.0007630296,0.09143543],"genre_scores_gemma":[0.4868591,0.005336117,0.4684272,0.02566255,0.001835773,0.003997089,0.0001925537,0.0002908836,0.007398676],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3288297,"threshold_uncertainty_score":0.8276725,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366384242","doi":"10.3138/cjpe.0023.008","title":"Insights into Evaluation Capacity Building: Motivations, Strategies, Outcomes, and Lessons Learned","year":2009,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":37,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Variety (cybernetics); Capacity building; Accountability; Face (sociological concept); Public relations; Business; Psychology; Process management; Knowledge management; Political science; Sociology; Computer science","authors":[{"name":"Hallie Preskill","is_ca":false},{"name":"Shanelle Boyle","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.536890060585536,"gpt":0.5401212105282374,"spread":0.00323114994270135,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0448657,0.0004585156,0.0003515358,0.002540866,0.004068457,0.008686027,0.001675771,0.001613503,0.002136057],"category_scores_gemma":[0.06958214,0.000457893,0.0003145345,0.001850447,0.006569311,0.005928765,0.005283885,0.003455112,0.0002475416],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008676663,"about_ca_system_score_gemma":0.01327978,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004128863,"about_ca_topic_score_gemma":0.007088826,"domain_scores_codex":[0.9630778,0.0283275,0.0008742556,0.0006598614,0.003665907,0.003394587],"domain_scores_gemma":[0.8884618,0.0873106,0.005376652,0.002501184,0.008393232,0.007956474],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001715904,0.001697446,0.08395433,0.001091387,0.00004558303,0.002109223,0.5711557,0.001218258,0.002468066,0.07529641,0.006906688,0.2538853],"study_design_scores_gemma":[0.00005199953,0.0002672499,0.04749249,0.001670787,0.00003126374,0.0009946779,0.8443321,0.005308205,0.0039025,0.05685182,0.03897184,0.0001251517],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8884088,0.001231642,0.02275439,0.03540253,0.00006909833,0.0005669168,0.00005766102,0.0001283102,0.05138073],"genre_scores_gemma":[0.9935714,0.0003087676,0.004770644,0.0003820651,0.00001038001,0.0001035925,0.00001406839,0.00001427983,0.0008247052],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0448657,"threshold_uncertainty_score":0.2372753,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W193071","doi":"10.3138/cjpe.17.006","title":"Developing an Evaluation Habit of Mind","year":2002,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":36,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"University of Toronto","funders":"","keywords":"Habit; Accountability; Sustainability; Psychology; Political science; Sociology; Social psychology; Law","authors":[{"name":"Steven T. Katz","is_ca":false},{"name":"Stephanie Sutherland","is_ca":false},{"name":"Lorna Earl","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6840143153944287,"gpt":0.5727226976961833,"spread":0.1112916176982454,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07910418,0.0004474728,0.0004660106,0.00247539,0.004176546,0.01212385,0.001980125,0.003948609,0.00470713],"category_scores_gemma":[0.1078602,0.0006595159,0.0006228182,0.001006887,0.02825219,0.0169794,0.01514692,0.008858819,0.001386694],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007548277,"about_ca_system_score_gemma":0.01490839,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001953429,"about_ca_topic_score_gemma":0.002230284,"domain_scores_codex":[0.9477211,0.03653224,0.001594579,0.003740587,0.008647246,0.001764283],"domain_scores_gemma":[0.871371,0.0653531,0.0063982,0.0190937,0.02860787,0.009176132],"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":[0.00009413223,0.0004272937,0.01917924,0.0003211434,0.00006116434,0.0003606635,0.06946992,0.001112925,0.003241668,0.6965797,0.0283983,0.1807539],"study_design_scores_gemma":[0.00005996494,0.0003861699,0.006032216,0.00133722,0.00004012364,0.0007600681,0.03256794,0.00704607,0.0056685,0.5475942,0.3983451,0.0001622674],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1029025,0.001563633,0.4996344,0.1705142,0.002047436,0.0008370568,0.00006815411,0.001859233,0.2205735],"genre_scores_gemma":[0.8385572,0.0004680429,0.1285183,0.01685193,0.0003156981,0.0007509436,0.0000576809,0.0003836421,0.01409664],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.07910418,"threshold_uncertainty_score":0.4183479,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3099471317","doi":"10.3138/cjpe.70004","title":"Sustainability Analysis of Intervention Benefits: A Theory of Change Approach","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Sustainability; Intervention (counseling); Spell; Sustainability organizations; Work (physics); Environmental economics; Business; Environmental resource management; Economics; Psychology; Engineering; Sociology; Ecology","authors":[{"name":"John Mayne","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6214875283452821,"gpt":0.5413536083546764,"spread":0.0801339199906057,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03599814,0.002627802,0.002473149,0.009978209,0.002351272,0.008729459,0.003766994,0.004051335,0.01465907],"category_scores_gemma":[0.05351125,0.0009190356,0.00382555,0.005364242,0.0141157,0.006311732,0.004594982,0.006593397,0.0006266288],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01811122,"about_ca_system_score_gemma":0.01359689,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006232088,"about_ca_topic_score_gemma":0.006074107,"domain_scores_codex":[0.9538644,0.03785364,0.0008054721,0.001735981,0.004735942,0.001004614],"domain_scores_gemma":[0.9216486,0.06955884,0.002347301,0.001834507,0.003865835,0.00074476],"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":[0.0000708402,0.0003202189,0.001677806,0.001030121,0.0003863454,0.00007913912,0.0009191333,0.04348283,0.0001371335,0.9190918,0.002415804,0.03038885],"study_design_scores_gemma":[0.00007739758,0.0003450925,0.001313271,0.001259021,0.0002844528,0.00006261359,0.001838294,0.07944666,0.0004203078,0.8989553,0.0159404,0.00005719472],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01467828,0.003521379,0.8407967,0.02297356,0.0008817923,0.0015373,0.0004297069,0.0002419983,0.1149393],"genre_scores_gemma":[0.6995466,0.003581203,0.2838722,0.002691414,0.0004104333,0.0035474,0.0002120432,0.0001617427,0.00597688],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03599814,"threshold_uncertainty_score":0.1903787,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366449629","doi":"10.3138/cjpe.0021.006","title":"Studies Are Not Enough: The Necessary Transformation of Evaluation","year":2007,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"GRASP; Public sector; Business; Knowledge management; Profit (economics); Public relations; Not for profit; Political science; Economics; Computer science","authors":[{"name":"John Mayne","is_ca":false},{"name":"Ray C. Rist","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.55640313541793,"gpt":0.5856812041578379,"spread":0.0292780687399079,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.353658,0.000822185,0.00256768,0.005222904,0.008942256,0.02903541,0.004399808,0.01067907,0.003920452],"category_scores_gemma":[0.4488707,0.001151497,0.001260362,0.0033596,0.07728098,0.04439994,0.01873572,0.02637104,0.0007883733],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01847551,"about_ca_system_score_gemma":0.065648,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007013655,"about_ca_topic_score_gemma":0.007928526,"domain_scores_codex":[0.6066589,0.3052572,0.02256578,0.006812635,0.05378854,0.004916838],"domain_scores_gemma":[0.2291865,0.6286788,0.01387563,0.05303248,0.06294917,0.01227745],"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":[0.0001230188,0.0002863307,0.002890466,0.001875647,0.00007767027,0.0002956429,0.02781392,0.0005019939,0.0003370977,0.7152585,0.08077608,0.1697636],"study_design_scores_gemma":[0.0001437118,0.0001769016,0.001776298,0.006042819,0.00006623693,0.0003442288,0.03014342,0.0008729195,0.0005523886,0.7162625,0.2435249,0.00009366114],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.003914469,0.01624174,0.0166572,0.9329439,0.005094341,0.0002547494,0.0000519847,0.0001205609,0.02472104],"genre_scores_gemma":[0.6092291,0.01696663,0.05618177,0.3018836,0.007619272,0.001798696,0.0001198341,0.0002738036,0.005927234],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.353658,"threshold_uncertainty_score":0.7970548,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2484173420","doi":"10.3138/cjpe.0026.006","title":"Logic Analysis: Testing Program Theory to Better Evaluate Complex Interventions","year":2012,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"School Health and Nursing Education","field":"Health Professions","cited_by":33,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":false},"ca_institutions":"Université de Sherbrooke; Université de Montréal","funders":"Canadian Institutes of Health Research; U.S. Public Health Service","keywords":"Psychological intervention; Computer science; Logic model; Risk analysis (engineering); Action (physics); Management science; Psychology; Medicine; Sociology; Engineering","authors":[{"name":"Lynda Rey","is_ca":true},{"name":"Astrid Brousselle","is_ca":true},{"name":"Nicole Dedobbeleer","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5926332520068205,"gpt":0.6085805738558242,"spread":0.01594732184900371,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09226405,0.00192109,0.001548738,0.004592285,0.001721316,0.005196677,0.002693986,0.001460701,0.01251306],"category_scores_gemma":[0.2323447,0.0006965997,0.002245934,0.003006057,0.005577484,0.01070297,0.003553049,0.003372174,0.0007605032],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008428888,"about_ca_system_score_gemma":0.0140744,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004604562,"about_ca_topic_score_gemma":0.004116864,"domain_scores_codex":[0.9091084,0.07468042,0.002843628,0.003078641,0.009253605,0.001035268],"domain_scores_gemma":[0.6580298,0.312827,0.008331132,0.008367836,0.01131564,0.001128634],"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":[0.001852756,0.002580944,0.02002324,0.007029234,0.001289948,0.0002171,0.006717181,0.04838037,0.003437095,0.3616907,0.0100435,0.5367379],"study_design_scores_gemma":[0.002426717,0.007819811,0.01079369,0.004976672,0.001318911,0.0002669004,0.006518464,0.2686407,0.0148766,0.6487314,0.03326952,0.0003606158],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.09333058,0.001327772,0.840848,0.006185607,0.0004023616,0.007617693,0.0009669819,0.001468863,0.04785215],"genre_scores_gemma":[0.2949881,0.0007553725,0.6933873,0.001264323,0.00007795649,0.007764854,0.0004149264,0.0001923407,0.00115482],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.09226405,"threshold_uncertainty_score":0.4879448,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2919316438","doi":"10.3138/cjpe.52946","title":"Using Actor-Based Theories of Change to Conduct Robust Evaluation in Complex Settings","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":33,"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; Democracy; Management science; Computer science; Psychological intervention; Process management; Sociology; Operations research; Business; Risk analysis (engineering); Political science; Economics; Psychology; Management; Mathematics; Law","authors":[{"name":"Andrew Koleros","is_ca":false},{"name":"John Mayne","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8242816801062659,"gpt":0.5913763997790368,"spread":0.2329052803272291,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.233201,0.001888669,0.002214649,0.005849046,0.00359972,0.01090224,0.00396316,0.002936662,0.008679924],"category_scores_gemma":[0.3376943,0.001128682,0.002363793,0.003538313,0.01541652,0.01287195,0.01007703,0.00586785,0.0008877838],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01501502,"about_ca_system_score_gemma":0.02193115,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007963577,"about_ca_topic_score_gemma":0.01045918,"domain_scores_codex":[0.6669112,0.3042743,0.005982872,0.007918364,0.0132142,0.001699157],"domain_scores_gemma":[0.5476487,0.3812912,0.01871288,0.02954252,0.02019583,0.002608827],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004883928,0.0008977547,0.01957136,0.004107799,0.001235016,0.0001829579,0.01448085,0.1167756,0.001055548,0.6436063,0.003888265,0.1937101],"study_design_scores_gemma":[0.0004825768,0.001189597,0.006556317,0.004536991,0.00042796,0.00009241816,0.009385366,0.2167087,0.002898569,0.7329442,0.02458816,0.0001891511],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02483411,0.0008145169,0.9341268,0.004614874,0.0002419871,0.004779272,0.0002538536,0.0003560214,0.0299785],"genre_scores_gemma":[0.4009103,0.000467129,0.5892372,0.0005823236,0.0000439135,0.007612561,0.0001612478,0.0001001208,0.0008852998],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.766799,"threshold_uncertainty_score":0.9455997,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1594443358","doi":"10.3138/cjpe.027.002","title":"Research on Evaluation: A Needs Assessment","year":2012,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":31,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Context (archaeology); Psychology; Survey research; Engineering ethics; Management science; Sociology; Applied psychology; Engineering","authors":[{"name":"Michael Szanyi","is_ca":false},{"name":"Tarek Azzam","is_ca":false},{"name":"Matthew Galen","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7833440326357193,"gpt":0.6849524617018541,"spread":0.09839157093386519,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2424101,0.0005782541,0.002133234,0.01421515,0.007725369,0.01689139,0.002232369,0.008007633,0.004570825],"category_scores_gemma":[0.3148864,0.0009542456,0.0007612598,0.01034929,0.01385131,0.02312457,0.006887286,0.006038045,0.0007703134],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01546148,"about_ca_system_score_gemma":0.04524801,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002836159,"about_ca_topic_score_gemma":0.003263121,"domain_scores_codex":[0.8170547,0.1347093,0.01542053,0.002018375,0.02760731,0.003189682],"domain_scores_gemma":[0.3551454,0.5328668,0.01328593,0.008982779,0.07583314,0.01388605],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0002476373,0.001494848,0.02473318,0.03894914,0.00009910335,0.0005446845,0.06102716,0.001143767,0.001048071,0.157124,0.03532137,0.678267],"study_design_scores_gemma":[0.0001902154,0.002218191,0.02728864,0.1558491,0.0001699864,0.001826124,0.35981,0.004048598,0.001561663,0.1329394,0.3138379,0.000260215],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.06721896,0.1360958,0.01994883,0.7186894,0.002660146,0.003279918,0.0003002881,0.0001681555,0.05163841],"genre_scores_gemma":[0.8117876,0.09199062,0.05104283,0.03611031,0.002436712,0.004247384,0.0002991828,0.00006097148,0.002024386],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9845385,"threshold_uncertainty_score":0.9342433,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W242703065","doi":"10.3138/cjpe.0014.005","title":"Organizational Development and Evaluation","year":2000,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Organization development; Organizational learning; Organizational effectiveness; Organizational culture; Organizational behavior and human resources; Knowledge management; Facilitation; Organizational commitment; Organizational performance; Organizational studies; Organizational engineering; Process (computing); Psychology; Process management; Business; Computer science; Public relations; Political science; Social psychology","authors":[{"name":"Michael Quinn Patton","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3470470019193243,"gpt":0.5167326550286309,"spread":0.1696856531093066,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1403667,0.0008725005,0.001384978,0.007180443,0.003628397,0.01272222,0.002295408,0.001531736,0.02815723],"category_scores_gemma":[0.1670165,0.0004347784,0.0007266285,0.007281725,0.005095726,0.00467689,0.0066659,0.002350188,0.007309508],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01165,"about_ca_system_score_gemma":0.03737891,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005260071,"about_ca_topic_score_gemma":0.003667716,"domain_scores_codex":[0.7979097,0.1508173,0.01158717,0.00394011,0.03271307,0.003032513],"domain_scores_gemma":[0.791372,0.07033464,0.008957419,0.02432299,0.09888131,0.006131719],"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":[0.0002260958,0.0003390887,0.004980696,0.002056487,0.00008359922,0.00005946299,0.005179083,0.001145751,0.0003589478,0.2714963,0.07280471,0.6412699],"study_design_scores_gemma":[0.0001690398,0.0007180449,0.01192502,0.006363649,0.0001054561,0.0001748434,0.01164261,0.003017473,0.002210275,0.08480655,0.8787739,0.00009318677],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.01948,0.01040871,0.1900172,0.02630502,0.003391458,0.02470445,0.004281546,0.001215728,0.7201958],"genre_scores_gemma":[0.4419363,0.008756287,0.3485641,0.005584746,0.0008977194,0.04045519,0.004445456,0.001095245,0.1482649],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1403667,"threshold_uncertainty_score":0.7423388,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1720965121","doi":"10.3138/cjpe.28.001","title":"Measuring Organizational Evaluation Capacity in the Canadian Federal Government","year":2013,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"Saint Paul University; Carleton University; École Nationale d'Administration Publique","funders":"","keywords":"Government (linguistics); Identification (biology); Organization development; Business; Evaluation methods; Organizational effectiveness; Organizational learning; Organizational identification; Public relations; Process management; Organizational commitment; Knowledge management; Political science; Computer science; Engineering","authors":[{"name":"Isabelle Bourgeois","is_ca":true},{"name":"Eleanor Toews","is_ca":true},{"name":"Jane Whynot","is_ca":false},{"name":"Mary Kay Lamarche","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5879919298116747,"gpt":0.4514966114809471,"spread":0.1364953183307276,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01546718,0.0003480742,0.0003550989,0.006664886,0.00700048,0.0037466,0.00188517,0.0005359274,0.002939616],"category_scores_gemma":[0.03102665,0.0003433425,0.0004106862,0.005605545,0.00272967,0.001217842,0.003044218,0.0009840993,0.0002609484],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.1308671,"about_ca_system_score_gemma":0.1819147,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.9756441,"about_ca_topic_score_gemma":0.9826887,"domain_scores_codex":[0.9836665,0.003617933,0.0007666287,0.0007708727,0.008598268,0.00257979],"domain_scores_gemma":[0.9362344,0.006344948,0.003457631,0.001694828,0.04659132,0.005676783],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003448846,0.0007316595,0.7102209,0.0004837431,0.0001680398,0.000135074,0.02636717,0.00653233,0.001785456,0.02722151,0.03297699,0.1930322],"study_design_scores_gemma":[0.00003134134,0.0001413959,0.9282761,0.0003054557,0.00004808122,0.00003252619,0.0206303,0.005471404,0.002029125,0.001169607,0.04175176,0.0001129616],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9143034,0.0006416833,0.003038117,0.002429314,0.00006201737,0.001049755,0.001879372,0.0001649559,0.07643141],"genre_scores_gemma":[0.9919992,0.0001739875,0.004224079,0.0001205856,0.000006196973,0.0002615272,0.000606374,0.00001447153,0.002593557],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9845328,"threshold_uncertainty_score":0.9495119,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2946489075","doi":"10.3138/cjpe.43050","title":"Principles, Approaches, and Methods for Evaluation in Indigenous Contexts: A Grey Literature Scoping Review","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"University of Toronto; Public Health Ontario","funders":"","keywords":"Grey literature; Indigenous; Interpretation (philosophy); Management science; Engineering ethics; Sociology; Psychology; Computer science; MEDLINE; Political science; Engineering; Ecology","authors":[{"name":"Kriti Chandna","is_ca":false},{"name":"Michelle M. Vine","is_ca":false},{"name":"Susan J. Snelling","is_ca":false},{"name":"Rachel Harris","is_ca":true},{"name":"Janet Smylie","is_ca":false},{"name":"Heather Manson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4738564796823058,"gpt":0.5813766564612814,"spread":0.1075201767789757,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2606859,0.0022201,0.004923676,0.04545834,0.007002658,0.01881349,0.004943403,0.004821462,0.003168678],"category_scores_gemma":[0.2884158,0.001852128,0.004334516,0.04250257,0.01242817,0.01498084,0.01065574,0.006124752,0.0009553327],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02859141,"about_ca_system_score_gemma":0.1321431,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03219179,"about_ca_topic_score_gemma":0.04793666,"domain_scores_codex":[0.7905819,0.1350142,0.03679332,0.004119088,0.03082606,0.002665506],"domain_scores_gemma":[0.6816413,0.2392655,0.01381054,0.01019553,0.05333888,0.001748262],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0001140878,0.0001780078,0.002824905,0.1890672,0.0007278605,0.0004904763,0.0463493,0.002347363,0.0008235762,0.1312186,0.0172085,0.6086502],"study_design_scores_gemma":[0.0000683394,0.0001212525,0.002802782,0.6729056,0.0009757883,0.0003440238,0.03418388,0.001492038,0.0008264105,0.09093033,0.195187,0.0001625562],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.006384535,0.7141117,0.1579574,0.05419334,0.002644047,0.02006,0.001241434,0.0002947579,0.04311271],"genre_scores_gemma":[0.06821536,0.53038,0.3628475,0.005254621,0.0004984121,0.02934875,0.0008140798,0.0001408911,0.002500433],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.7393141,"threshold_uncertainty_score":0.9117059,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W18830643","doi":"10.3138/cjpe.026.001","title":"Advancing Empirical Scholarship to Further Develop Evaluation Theory and Practice","year":2011,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Scholarship; Grounded theory; Context (archaeology); Development theory; Field (mathematics); Empirical research; Contingency; Epistemology; Sociology; Intersection (aeronautics); Contingency theory; Management science; Engineering ethics; Qualitative research; Knowledge management; Computer science; Social science; Political science; Economics","authors":[{"name":"Christina A. Christie","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.580695406400756,"gpt":0.5966934622915074,"spread":0.01599805589075143,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09965651,0.001029154,0.001283522,0.009681543,0.004086884,0.01503424,0.003440812,0.005808179,0.01001852],"category_scores_gemma":[0.1708134,0.0006412046,0.0009025412,0.005215012,0.02590112,0.02371424,0.00838335,0.01005554,0.001766211],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01577884,"about_ca_system_score_gemma":0.03858793,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005630156,"about_ca_topic_score_gemma":0.00628848,"domain_scores_codex":[0.9400529,0.04521612,0.003214578,0.001772784,0.008369938,0.001373656],"domain_scores_gemma":[0.6790364,0.2508772,0.007243072,0.02148796,0.03686881,0.004486474],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00001206887,0.0002844561,0.001558637,0.0006785861,0.00002107174,0.0000540847,0.002817515,0.001406099,0.0001305821,0.9263563,0.006817226,0.05986344],"study_design_scores_gemma":[0.00004249383,0.00005650786,0.001287844,0.005855753,0.00001927196,0.00006090886,0.007437041,0.005305101,0.0005385125,0.9094335,0.06993362,0.00002937895],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01891878,0.02430781,0.3886815,0.3614273,0.002494595,0.0009004703,0.0001885604,0.0003456891,0.2027352],"genre_scores_gemma":[0.5642834,0.02651652,0.3748195,0.02135965,0.001990595,0.001660338,0.0003586346,0.000201141,0.008810181],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9003435,"threshold_uncertainty_score":0.5270404,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4367029949","doi":"10.3138/cjpe.25.002","title":"Contribution Analysis Applied: Reflections on Scope and Methodology","year":2010,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Scope (computer science); Perspective (graphical); Relation (database); Elaboration; Management science; Epistemology; Computer science; Engineering ethics; Sociology; Engineering; Data mining; Philosophy","authors":[{"name":"Line Dybdal","is_ca":false},{"name":"Steffen Bohni Nielsen","is_ca":false},{"name":"Sebastian Lemire","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6370922382666173,"gpt":0.6304102886950795,"spread":0.006681949571537782,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2396489,0.001425073,0.002056442,0.008922871,0.008250631,0.0229296,0.00638358,0.007237103,0.00697528],"category_scores_gemma":[0.2104961,0.001380092,0.00231723,0.006800049,0.07435194,0.02993663,0.01655673,0.01467404,0.001399486],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01370394,"about_ca_system_score_gemma":0.01629221,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004086325,"about_ca_topic_score_gemma":0.002201741,"domain_scores_codex":[0.7174971,0.2479287,0.005578427,0.00574485,0.01983655,0.003414361],"domain_scores_gemma":[0.5637535,0.3914856,0.003593495,0.01624748,0.02295645,0.001963435],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00001885318,0.00002441598,0.0003916214,0.0002930654,0.00002065339,0.00004989755,0.01195658,0.0003478204,0.00007822732,0.9544989,0.002017857,0.03030195],"study_design_scores_gemma":[0.00002395699,0.00002893907,0.0003731239,0.001379629,0.00001349357,0.0001216364,0.006638133,0.00212729,0.0004235779,0.92773,0.06110235,0.00003779114],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01005037,0.01089358,0.7624105,0.09103332,0.001911663,0.0007260577,0.00007274887,0.0002705341,0.1226313],"genre_scores_gemma":[0.5678867,0.008063321,0.39301,0.01099438,0.002112228,0.003460421,0.00009370441,0.0008009038,0.01357849],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7603511,"threshold_uncertainty_score":0.9376483,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366451089","doi":"10.3138/cjpe.022.006","title":"Youth Voices: Evaluation of Participatory Action Research","year":2007,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Participatory Visual Research Methods","field":"Social Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"York University; University of Toronto","funders":"","keywords":"Participatory action research; Participatory evaluation; Citizen journalism; Reflexivity; Psychosocial; Action research; Psychology; Sociology; Action (physics); Process (computing); Applied psychology; Community-based participatory research; Public relations; Medical education; Pedagogy; Political science; Social science; Computer science; Medicine; Psychotherapist","authors":[{"name":"Sandra Chen","is_ca":false},{"name":"Blake Poland","is_ca":true},{"name":"Harvey A. Skinner","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9730869743906994,"gpt":0.7857522584583132,"spread":0.1873347159323862,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2672002,0.001035076,0.001086288,0.002691301,0.004802553,0.006776205,0.002570282,0.002076407,0.006357534],"category_scores_gemma":[0.2532825,0.0005961632,0.001231904,0.002709892,0.005189474,0.003904054,0.009032641,0.001676807,0.0005506985],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009852321,"about_ca_system_score_gemma":0.01892163,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004537904,"about_ca_topic_score_gemma":0.005711209,"domain_scores_codex":[0.5482342,0.411456,0.01027687,0.004120916,0.02249447,0.003417577],"domain_scores_gemma":[0.7037481,0.229818,0.01194766,0.01485117,0.03489037,0.004744631],"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":[0.003871825,0.003341974,0.01305333,0.01145688,0.0003854455,0.0007593047,0.2622185,0.00329826,0.00321074,0.04845501,0.008784953,0.6411638],"study_design_scores_gemma":[0.006761469,0.02559626,0.0327822,0.02807043,0.001574958,0.001437206,0.393413,0.01438586,0.03970745,0.06826971,0.3873505,0.0006509658],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5998687,0.004214605,0.160503,0.009534791,0.001207623,0.06067692,0.001182034,0.000754869,0.1620576],"genre_scores_gemma":[0.8506115,0.001146173,0.1180346,0.000690148,0.00008001948,0.02542579,0.0002764583,0.0001239416,0.003611389],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2672002,"threshold_uncertainty_score":0.9036726,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2184444384","doi":"10.3138/cjpe.19.007","title":"The Analysis of Focus Groups in Published Research Articles","year":2004,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Focus Groups and Qualitative Methods","field":"Social Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Focus group; sort; Inter-rater reliability; Coding (social sciences); Focus (optics); Psychology; Computer science; Sociology; Applied psychology; Information retrieval; Social science; Developmental psychology","authors":[{"name":"Geoffrey S. Wiggins","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3988284955321382,"gpt":0.5595611463391841,"spread":0.1607326508070459,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1546099,0.001198385,0.001866907,0.05210246,0.005321173,0.007396265,0.003136856,0.00242735,0.006362868],"category_scores_gemma":[0.4614027,0.001166653,0.001617728,0.03238985,0.004650881,0.007584308,0.005936365,0.00129179,0.001199728],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007733393,"about_ca_system_score_gemma":0.006921876,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001780823,"about_ca_topic_score_gemma":0.00194038,"domain_scores_codex":[0.7910122,0.13353,0.02062658,0.005328418,0.04665621,0.002846528],"domain_scores_gemma":[0.1991892,0.7060473,0.02664297,0.009988546,0.05711108,0.001020873],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.000813626,0.0002683241,0.02831952,0.03007866,0.0007529226,0.00200264,0.6580716,0.0004334344,0.004760421,0.009635782,0.008815827,0.2560472],"study_design_scores_gemma":[0.0003187133,0.001158807,0.09458847,0.04913208,0.001432493,0.003004535,0.6734546,0.002041313,0.01806433,0.01718726,0.1392755,0.0003418393],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7792488,0.03753429,0.1003447,0.007323374,0.003926793,0.01415617,0.005063273,0.0005643169,0.05183832],"genre_scores_gemma":[0.8809252,0.01833592,0.06725807,0.002031289,0.001442591,0.02117928,0.002676394,0.0003554799,0.005795806],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8453901,"threshold_uncertainty_score":0.8176653,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2185731738","doi":"10.3138/cjpe.19.001","title":"The Role of the Evaluator in a Political World","year":2004,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":28,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Politics; Power (physics); Simplicity; Argument (complex analysis); Elite; Ambiguity; Servant; Sociology; Independence (probability theory); Fundamentalism; Environmental ethics; Law; Political science; Political economy; Epistemology; Philosophy","authors":[{"name":"Ernest R. House","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2293825789903462,"gpt":0.5204448598505118,"spread":0.2910622808601656,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05846143,0.0008056386,0.001051628,0.00386141,0.01834312,0.04147071,0.001873774,0.00645505,0.004218417],"category_scores_gemma":[0.05238507,0.0006793234,0.0004240078,0.003586445,0.09133042,0.02313066,0.007611832,0.01221747,0.001063847],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01373143,"about_ca_system_score_gemma":0.0151489,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006378277,"about_ca_topic_score_gemma":0.005940548,"domain_scores_codex":[0.8913752,0.09236956,0.001249258,0.003496347,0.008180236,0.003329415],"domain_scores_gemma":[0.9475804,0.03781402,0.001884446,0.002379418,0.005873042,0.004468746],"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":[0.00001827646,0.00001850806,0.0003142991,0.00006591958,0.000008860087,0.00007220335,0.01876612,0.0001642485,0.00008799622,0.952623,0.01683391,0.01102674],"study_design_scores_gemma":[0.00003218566,0.00004384837,0.0005625318,0.0006796815,0.00001273815,0.0001452598,0.0207702,0.0007656987,0.0003683394,0.6220622,0.3545079,0.0000494805],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.01613638,0.03480453,0.05139019,0.4419791,0.005098042,0.0001088389,0.00005767851,0.0002014911,0.4502237],"genre_scores_gemma":[0.8635861,0.01425931,0.01855861,0.04648434,0.004547737,0.0003532176,0.0000509908,0.0005031586,0.0516566],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.05846143,"threshold_uncertainty_score":0.3091773,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2150927259","doi":"10.3138/cjpe.27.003","title":"The Art of the Nudge: Five Practices for Developmental Evaluators","year":2012,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":27,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"Concordia University","funders":"","keywords":"Servant; Psychology; Adaptation (eye); Servant leadership; Energy (signal processing); Pedagogy; Applied psychology; Public relations; Social psychology; Political science; Computer science","authors":[{"name":"Marc Langlois","is_ca":false},{"name":"Natasha Blanchet‐Cohen","is_ca":true},{"name":"Tanya Beer","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5020626007095538,"gpt":0.5787691116820812,"spread":0.07670651097252745,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.4297931,0.002244257,0.002239099,0.0116874,0.02569194,0.03331185,0.006971737,0.01179988,0.003645247],"category_scores_gemma":[0.3127874,0.002380642,0.001467058,0.004580814,0.1088536,0.03389411,0.03753179,0.03204256,0.001273411],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02328794,"about_ca_system_score_gemma":0.04327685,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008023971,"about_ca_topic_score_gemma":0.01606318,"domain_scores_codex":[0.5017094,0.4539477,0.01140238,0.008283056,0.02013813,0.004519418],"domain_scores_gemma":[0.6267995,0.2435865,0.008638199,0.0584804,0.04515722,0.01733817],"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":[0.000179302,0.0003688445,0.005944911,0.001088095,0.00008546364,0.0006239585,0.3683169,0.0006598383,0.00133723,0.3765441,0.0361382,0.2087132],"study_design_scores_gemma":[0.0001688804,0.0003337359,0.001876555,0.006734121,0.00007297278,0.0009883658,0.1856551,0.00279819,0.001605417,0.4275073,0.3719245,0.0003348077],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.03207399,0.01144289,0.3978959,0.4898207,0.00379965,0.002405654,0.00009515625,0.001539184,0.06092699],"genre_scores_gemma":[0.3642545,0.002241286,0.6040189,0.0187626,0.0004304163,0.004165328,0.00004034556,0.0005106954,0.005575996],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.4297931,"threshold_uncertainty_score":0.7031666,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1493255413","doi":"10.3138/cjpe.027.003","title":"Capturing the Complexity of Evaluations of Health Promotion Interventions: A Scoping Review","year":2012,"lang":"en","type":"review","venue":"Canadian Journal of Program Evaluation","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":27,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"Lloydminster Hospital; University of Calgary","funders":"","keywords":"Psychological intervention; Perspective (graphical); Promotion (chess); Grey literature; Health promotion; Management science; Psychology; Applied psychology; Computer science; MEDLINE; Medicine; Public health; Nursing; Political science; Engineering; Artificial intelligence","authors":[{"name":"Ania Kania","is_ca":true},{"name":"Alka Patel","is_ca":true},{"name":"Amrita Roy","is_ca":true},{"name":"Gregory Yelland","is_ca":true},{"name":"Duyên Thi Kim Nguyêñ","is_ca":true},{"name":"Marja J. Verhoef","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9727255468434597,"gpt":0.7916901593614977,"spread":0.181035387481962,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3281443,0.002069994,0.007253244,0.02118894,0.002195904,0.01047111,0.003153474,0.002789951,0.00198802],"category_scores_gemma":[0.5962346,0.002361517,0.006069713,0.02113451,0.003176657,0.01003801,0.004279173,0.002823628,0.000271761],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01390696,"about_ca_system_score_gemma":0.05718158,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01118477,"about_ca_topic_score_gemma":0.03301184,"domain_scores_codex":[0.6193001,0.252163,0.07159695,0.005115998,0.05023911,0.001584804],"domain_scores_gemma":[0.2697079,0.6151707,0.04126368,0.01541815,0.05728308,0.001156509],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0004112026,0.0002217422,0.007278498,0.4689254,0.01130065,0.0001418584,0.005295319,0.002130482,0.0007876255,0.008621112,0.00419114,0.4906949],"study_design_scores_gemma":[0.0003084735,0.0007494228,0.01144052,0.9145831,0.02057099,0.0002179651,0.004142864,0.002198292,0.00159371,0.01039595,0.03361371,0.0001849761],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.02166447,0.8879426,0.04890873,0.009747835,0.001407358,0.01962988,0.00107179,0.0001374527,0.00948988],"genre_scores_gemma":[0.2338677,0.5744503,0.1494889,0.003508123,0.000578216,0.03589088,0.001276721,0.0000896451,0.0008494728],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6718557,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2043777716","doi":"10.3138/cjpe.29.1.36","title":"Supporting Knowledge Translation Through Evaluation: Evaluator as Knowledge Broker","year":2014,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"McMaster University; Queen's University","funders":"","keywords":"Knowledge translation; Context (archaeology); Knowledge management; Computer science; History","authors":[{"name":"Catherine Donnelly","is_ca":true},{"name":"Lori Letts","is_ca":true},{"name":"Don A. Klinger","is_ca":true},{"name":"Lyn M. Shulha","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4946854783565933,"gpt":0.5959329823406669,"spread":0.1012475039840736,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2860007,0.001044451,0.001779371,0.004360806,0.00657123,0.01763914,0.00295219,0.003316939,0.01002268],"category_scores_gemma":[0.3300284,0.000970153,0.000917877,0.003659431,0.006513211,0.01880264,0.01475577,0.003256354,0.002548755],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009213918,"about_ca_system_score_gemma":0.03893637,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00306716,"about_ca_topic_score_gemma":0.003035498,"domain_scores_codex":[0.5438042,0.4126205,0.01206977,0.006616288,0.01932669,0.005562517],"domain_scores_gemma":[0.5103707,0.3702594,0.02440257,0.03401584,0.04793186,0.01301959],"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":[0.002140456,0.00379795,0.0276984,0.003698169,0.0002642255,0.001717361,0.137223,0.004032015,0.007633,0.04863327,0.0187159,0.7444463],"study_design_scores_gemma":[0.003551085,0.008277378,0.03100457,0.01723344,0.00139716,0.002193218,0.2544269,0.09716104,0.03975472,0.1455947,0.3983962,0.001009545],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.245385,0.00299615,0.5178627,0.05026895,0.001185859,0.01354368,0.0002598553,0.002851227,0.1656467],"genre_scores_gemma":[0.7825405,0.0008122145,0.2046748,0.001929347,0.000206484,0.003806237,0.0001347103,0.0002297484,0.005666079],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2860007,"threshold_uncertainty_score":0.8804883,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366450544","doi":"10.3138/cjpe.021.002","title":"Audit and Evaluation in Public Management: Challenges, Reforms, and Different Roles","year":2006,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Public Policy and Administration Research","field":"Social Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Audit; Public sector; Confusion; Business; New public management; Public relations; Public management; Face (sociological concept); Relation (database); Accounting; Political science; Sociology; Psychology; Computer science","authors":[{"name":"John Mayne","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1978141145687296,"gpt":0.4281585864049609,"spread":0.2303444718362313,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2276103,0.0006336669,0.001597008,0.008470509,0.009990237,0.0314019,0.002599031,0.009422389,0.003320767],"category_scores_gemma":[0.2016524,0.0008065535,0.0009147562,0.008173614,0.04248808,0.02260265,0.01095487,0.01305392,0.000589697],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.04469998,"about_ca_system_score_gemma":0.07229225,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03095755,"about_ca_topic_score_gemma":0.03939107,"domain_scores_codex":[0.6900478,0.236898,0.01124102,0.004493675,0.04451868,0.01280088],"domain_scores_gemma":[0.5261188,0.327488,0.01804115,0.01559534,0.09016714,0.02258959],"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":[0.0002082285,0.0002269176,0.008560669,0.002062896,0.00006013922,0.0001706889,0.004621073,0.002141254,0.0001819568,0.6161731,0.03435851,0.3312346],"study_design_scores_gemma":[0.0002248359,0.0004972552,0.02218882,0.01915706,0.0001163343,0.0007805863,0.02778617,0.005639534,0.00127933,0.5500339,0.3717776,0.0005186136],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.01244301,0.1148283,0.0138491,0.8084091,0.002780027,0.0002193942,0.0000461969,0.0001116981,0.04731316],"genre_scores_gemma":[0.8094403,0.08564563,0.03029044,0.06006172,0.006052375,0.0004533424,0.00006745862,0.0001456312,0.00784301],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2276103,"threshold_uncertainty_score":0.952494,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3105492687","doi":"10.3138/cjpe.69535","title":"Toward Learning from Change Pathways: Reviewing Theory of Change and Its Discontents","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"University of Guelph","funders":"","keywords":"Underpinning; Vagueness; Theory of change; Context (archaeology); Stakeholder; Process (computing); Stakeholder engagement; Epistemology; Management science; Knowledge management; Sociology; Computer science; Political science; Public relations; Economics; Engineering; Artificial intelligence","authors":[{"name":"Steven Lâm","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.918735116591738,"gpt":0.5122009043033375,"spread":0.4065342122884005,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1082902,0.001605559,0.003606569,0.02472061,0.004183078,0.01799041,0.006761186,0.007745811,0.002885476],"category_scores_gemma":[0.1711154,0.0009989961,0.002416411,0.02057079,0.02944988,0.02100604,0.006216215,0.0123211,0.0006602599],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02974603,"about_ca_system_score_gemma":0.04416681,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01612941,"about_ca_topic_score_gemma":0.02640666,"domain_scores_codex":[0.9423748,0.03628026,0.005682079,0.003725677,0.0109311,0.001006074],"domain_scores_gemma":[0.5903162,0.3653243,0.008294511,0.005031014,0.02895881,0.002075195],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00009457614,0.0001810564,0.002211924,0.07663037,0.0005726552,0.0003450548,0.01599032,0.004865854,0.000225291,0.3234645,0.02210297,0.5533155],"study_design_scores_gemma":[0.00004825703,0.00026033,0.003100295,0.2873466,0.000776638,0.0005661036,0.02799563,0.005258761,0.0008548646,0.3572784,0.3163411,0.0001730271],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00230001,0.876936,0.01996716,0.09056303,0.003213774,0.0002031685,0.00006651358,0.00006357135,0.006686924],"genre_scores_gemma":[0.1103215,0.8366564,0.02839342,0.02093102,0.002069573,0.0007611088,0.000136989,0.00006657859,0.0006635361],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8917098,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W281362470","doi":"10.3138/cjpe.23.004","title":"Challenges in Applying Indigenous Evaluation Practices in Mainstream Grant Programs to Indigenous Communities","year":2008,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Indigenous; Mainstream; Government (linguistics); Substance abuse prevention; Competence (human resources); Cultural competence; Traditional knowledge; Political science; Culturally appropriate; Public relations; Substance use; Sociology; Psychology; Medicine; Pedagogy; Gerontology; Social psychology","authors":[{"name":"Jane Gray Grover","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6836246640574786,"gpt":0.5359148392352847,"spread":0.1477098248221939,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5871889,0.0008533945,0.001850746,0.004591893,0.01523102,0.01864122,0.007853328,0.00336508,0.004052752],"category_scores_gemma":[0.5619349,0.00106661,0.001173926,0.003618,0.01838995,0.01121218,0.01625092,0.008165898,0.0006005201],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02873174,"about_ca_system_score_gemma":0.1140151,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0812263,"about_ca_topic_score_gemma":0.1484596,"domain_scores_codex":[0.4068881,0.4883211,0.03041249,0.008151392,0.05722749,0.008999411],"domain_scores_gemma":[0.3466752,0.4179647,0.02119153,0.03719273,0.1648374,0.01213844],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0006706282,0.001965955,0.03797376,0.006525357,0.0007140993,0.0008546591,0.3369581,0.00615229,0.002142387,0.09436827,0.02309181,0.4885827],"study_design_scores_gemma":[0.001013706,0.003058686,0.08268696,0.02514409,0.0005503432,0.000979626,0.472232,0.0143091,0.005763493,0.1981592,0.1952512,0.0008516281],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2833314,0.01474071,0.1950809,0.3381717,0.005143996,0.01797934,0.0003090489,0.0008169795,0.1444258],"genre_scores_gemma":[0.8363304,0.003529008,0.1277442,0.01563806,0.0007229202,0.01051827,0.00007537909,0.0002150096,0.005226808],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9712682,"threshold_uncertainty_score":0.5090696,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2945587103","doi":"10.3138/cjpe.42190","title":"Evaluation Literacy: Perspectives of Internal Evaluators in Non-Government Organizations","year":2019,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Government (linguistics); Context (archaeology); Conversation; Narrative; Literacy; Public relations; Psychology; Knowledge management; Sociology; Political science; Pedagogy; Computer science","authors":[{"name":"Alison Rogers","is_ca":false},{"name":"Alicia McCoy","is_ca":false},{"name":"Leanne M. Kelly","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1121734083055392,"gpt":0.4916620196187995,"spread":0.3794886113132602,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.08560964,0.0006761005,0.0009881698,0.003409746,0.02313668,0.02429298,0.002181028,0.004603373,0.004323727],"category_scores_gemma":[0.1045643,0.0008733827,0.0007780507,0.001990687,0.03832741,0.008457511,0.01564568,0.0112216,0.0004307235],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02266224,"about_ca_system_score_gemma":0.02482313,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02398675,"about_ca_topic_score_gemma":0.02512619,"domain_scores_codex":[0.8732608,0.1060117,0.002540399,0.002100023,0.007519554,0.008567649],"domain_scores_gemma":[0.7886861,0.1559713,0.01193942,0.004383196,0.0221637,0.0168563],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.00002403812,0.0000619148,0.005354456,0.0001008176,0.00000770707,0.000412818,0.9819912,0.00003737804,0.0002020175,0.005535449,0.001491652,0.004780419],"study_design_scores_gemma":[0.000005598311,0.00004606976,0.002273383,0.0004351634,0.000008591164,0.0002316053,0.976384,0.0001419214,0.0002542047,0.001503044,0.01868936,0.00002694953],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8731806,0.002493527,0.004293701,0.06157008,0.0003132943,0.0002206105,0.00004007143,0.00004679373,0.05784132],"genre_scores_gemma":[0.9911031,0.0006962796,0.0007002539,0.003930991,0.00004932343,0.00006912853,0.00001129291,0.00003048847,0.003409307],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08560964,"threshold_uncertainty_score":0.4527525,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4234946968","doi":"10.3138/cjpe.30.3.04","title":"Lessons on Decolonizing Evaluation from Kaupapa Māori Evaluation","year":2016,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false},"ca_institutions":"","funders":"","keywords":"Aotearoa; Indigenous; Sociology; Traditional knowledge; Equity (law); Christian ministry; Community development; Cultural competence; Pedagogy; Public relations; Political science; Gender studies","authors":[{"name":"Fiona Cram","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6218029580620157,"gpt":0.5966281862836175,"spread":0.02517477177839822,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2314901,0.0009680847,0.00120529,0.003629853,0.01550909,0.02112985,0.005403821,0.005331295,0.006625647],"category_scores_gemma":[0.2041686,0.0009880979,0.001322982,0.003233471,0.01936968,0.01707057,0.01839954,0.01370901,0.000797773],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.05708936,"about_ca_system_score_gemma":0.0903508,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.07866545,"about_ca_topic_score_gemma":0.136448,"domain_scores_codex":[0.7989737,0.1616187,0.009412055,0.002840073,0.01806146,0.009094099],"domain_scores_gemma":[0.7042134,0.1957637,0.006761682,0.01287222,0.07081874,0.009570265],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003994504,0.0006061498,0.00597787,0.005902341,0.0001897585,0.001800804,0.3120301,0.00140054,0.001671752,0.1595909,0.07133732,0.4390931],"study_design_scores_gemma":[0.0003910372,0.0009571202,0.01399931,0.01832553,0.0002109912,0.001117709,0.2761606,0.002225866,0.002808961,0.077704,0.6056922,0.0004067186],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.1166804,0.02956298,0.02603352,0.6412457,0.003301119,0.005025596,0.0001641728,0.0004043184,0.1775822],"genre_scores_gemma":[0.8563464,0.01449308,0.04843249,0.04792348,0.0007756723,0.006677993,0.0001493927,0.0004865144,0.02471493],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7685099,"threshold_uncertainty_score":0.9477096,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W4366721395","doi":"10.3138/cjpe.0017.007","title":"Introducing Program Teams to Logic Models: Facilitating the Learning Process","year":2003,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":false},"ca_institutions":"Health Canada","funders":"University of Ottawa","keywords":"Logic model; Process (computing); Key (lock); Computer science; Articulation (sociology); Point (geometry); Facilitation; Knowledge management; Process management; Management science; Psychology; Business; Engineering; Sociology; Political science; Mathematics","authors":[{"name":"Nancy L. Porteous","is_ca":true},{"name":"Barbara J. Sheldrick","is_ca":false},{"name":"Paula Stewart","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.367515472877912,"gpt":0.5381391714975737,"spread":0.1706236986196618,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02129254,0.001336889,0.0005395665,0.001544593,0.002358543,0.006658907,0.003033276,0.002313236,0.0121395],"category_scores_gemma":[0.04848566,0.0007375905,0.0009366276,0.0009986645,0.002606616,0.008690548,0.00987584,0.004530592,0.002823221],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002285394,"about_ca_system_score_gemma":0.005106755,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009591134,"about_ca_topic_score_gemma":0.001627658,"domain_scores_codex":[0.9811933,0.01540243,0.0003172448,0.000801176,0.001598007,0.0006878796],"domain_scores_gemma":[0.941784,0.04708508,0.00166306,0.003204051,0.003235655,0.003028233],"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":[0.0007590596,0.009188925,0.007040577,0.001597233,0.00006533417,0.001433955,0.1259927,0.01463026,0.01221026,0.08552048,0.0603227,0.6812385],"study_design_scores_gemma":[0.001326395,0.004943251,0.006399428,0.004282466,0.0001856007,0.002378731,0.09157533,0.1420897,0.03055693,0.2894886,0.4262471,0.0005264465],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1563346,0.0002582899,0.7568213,0.0147496,0.0003788814,0.002181063,0.0001299468,0.002830452,0.06631594],"genre_scores_gemma":[0.3498583,0.0005125292,0.6360174,0.001554874,0.000139551,0.00206671,0.0002532633,0.0003274181,0.009269984],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02129254,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4366382532","doi":"10.3138/cjpe.31132","title":"Moving Beyond the Buzzword: A Framework for Teaching Culturally Responsive Approaches to Evaluation","year":2017,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Conceptual framework; Competence (human resources); Cultural competence; Sociology; Domain (mathematical analysis); Knowledge management; Computer science; Engineering ethics; Psychology; Pedagogy; Social psychology; Social science; Engineering","authors":[{"name":"Ayesha S. Boyce","is_ca":false},{"name":"Jill Anne Chouinard","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6407724104552023,"gpt":0.5597165071488638,"spread":0.0810559033063385,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07858788,0.001479288,0.001024814,0.005907792,0.008998682,0.01482942,0.004696511,0.004775755,0.005119327],"category_scores_gemma":[0.05447143,0.0009083378,0.001239099,0.003621883,0.03557478,0.01367259,0.008342273,0.01064558,0.001404126],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03190763,"about_ca_system_score_gemma":0.0468986,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.04873698,"about_ca_topic_score_gemma":0.07509736,"domain_scores_codex":[0.9295334,0.06057052,0.001980357,0.001380739,0.004870213,0.001664781],"domain_scores_gemma":[0.9301057,0.04954509,0.001589272,0.003859519,0.01037714,0.004523231],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0000199856,0.0001294323,0.0005295515,0.0002946209,0.000007910486,0.0001350984,0.03143628,0.001097814,0.0004015137,0.9115824,0.009194953,0.04517048],"study_design_scores_gemma":[0.00005320083,0.00008136658,0.0005675491,0.002731179,0.00002495852,0.000235442,0.02687484,0.007203903,0.001248983,0.7183028,0.2425895,0.00008615202],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.00680241,0.002613758,0.8158563,0.06487997,0.000704396,0.001162757,0.00007894784,0.0006918893,0.1072095],"genre_scores_gemma":[0.1853369,0.00226491,0.7975312,0.00393203,0.0001570244,0.001911777,0.00009172709,0.0003028673,0.008471456],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9680924,"threshold_uncertainty_score":0.4156174,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}