{"id":"W4224229215","doi":"10.3138/cjpe.71300","title":"Professionalism in Program Evaluators: A Comparison of American and Canadian Evaluators","year":2022,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Professionalization; Credentialing; Autonomy; Exploratory research; Field (mathematics); Perception; Psychology; Sociology; Medical education; Pedagogy; Social science; Political science; Medicine; Law","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0294241,0.0001951614,0.0005863548,0.002555056,0.0004630601,0.0001964199,0.000696274,0.00006576128,0.002869964],"category_scores_gemma":[0.002731338,0.0001696412,0.0001332884,0.002663349,0.0002961439,0.0004385932,0.00005042088,0.000569332,0.000009508004],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001504749,"about_ca_system_score_gemma":0.02082393,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.07402527,"about_ca_topic_score_gemma":0.6235431,"domain_scores_codex":[0.9903772,0.002464139,0.001822781,0.0003521446,0.004455364,0.0005283116],"domain_scores_gemma":[0.994511,0.0003472895,0.001564617,0.0003672021,0.002393102,0.0008167662],"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":[0.00002628317,0.0001480695,0.2960329,0.000005435305,0.0000220352,0.000005689003,0.003719577,0.003297336,0.00001130432,0.0002164152,0.001408275,0.6951067],"study_design_scores_gemma":[0.002243588,0.003247332,0.6424399,0.00009012922,0.0001443454,0.000065859,0.02699221,0.2202131,0.00004146445,0.006233647,0.09793392,0.0003544897],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.991336,0.001190618,0.00001431189,0.003151517,0.0007927755,0.002511343,0.00003121654,0.000008251818,0.0009639408],"genre_scores_gemma":[0.9968247,0.000005841843,0.002283016,0.0002022306,0.00006000406,0.0005197826,0.00002888956,0.00001705814,0.00005841074],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6947522,"threshold_uncertainty_score":0.9994121,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4228532574958362,"score_gpt":0.5923266515598159,"score_spread":0.1694733940639797,"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."}}