{"id":"W4353055537","doi":"10.3138/cjpe.75431","title":"Building Evaluation Culture—The Missing Link","year":2023,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Sociology; Knowledge management; Psychology; Epistemology; Computer science; Philosophy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05735271,0.0001314581,0.000201165,0.0008599142,0.0006382905,0.0009247136,0.000674629,0.00009460787,0.001821118],"category_scores_gemma":[0.01042129,0.00008249324,0.0001671914,0.002353229,0.00010128,0.0007337571,0.00001831358,0.0002957174,0.000303388],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005397811,"about_ca_system_score_gemma":0.006295008,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001775126,"about_ca_topic_score_gemma":0.003225833,"domain_scores_codex":[0.9924225,0.001235902,0.001043907,0.0002276132,0.004710314,0.0003598027],"domain_scores_gemma":[0.9925413,0.0004150999,0.000822545,0.0003825198,0.005448218,0.0003903612],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000004649988,0.000006583839,0.002606763,0.000001903781,0.00001911173,0.000002660111,0.001418607,0.01063354,0.0001402006,0.0005200926,0.01371259,0.9709333],"study_design_scores_gemma":[0.0008529485,0.0002377616,0.06726978,0.00006841013,0.0001699818,0.00004499004,0.001420294,0.6034538,0.0001726793,0.07539648,0.2507598,0.0001531159],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8144254,0.004027716,0.01106253,0.1356717,0.01037508,0.007065487,0.00001801494,0.0001280268,0.01722601],"genre_scores_gemma":[0.9940717,0.00001743101,0.004354266,0.0004222477,0.0007002041,0.00009730976,0.00001814706,0.00001314246,0.0003055332],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9707802,"threshold_uncertainty_score":0.9993384,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5208539717794183,"score_gpt":0.5865932421472865,"score_spread":0.0657392703678682,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}