{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01432512,0.0002446544,0.0004037948,0.003884672,0.00781746,0.003987478,0.0009571626,0.0007174702,0.001733386],"category_scores_gemma":[0.03061724,0.0003499567,0.0002844477,0.003969399,0.003795539,0.0008623715,0.0027048,0.001255818,0.0001483173],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03138737,"about_ca_system_score_gemma":0.05834857,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.8827384,"about_ca_topic_score_gemma":0.9325259,"domain_scores_codex":[0.9895929,0.003073499,0.0003647212,0.000547299,0.00397883,0.002442619],"domain_scores_gemma":[0.9533486,0.01022167,0.004758782,0.0006765996,0.02184137,0.009153019],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003775824,0.0002535763,0.5302114,0.0002049652,0.00003800859,0.0003446877,0.4217964,0.0001162643,0.0009492318,0.001744897,0.003161002,0.04080201],"study_design_scores_gemma":[0.00002364115,0.0001714864,0.5544269,0.0002936497,0.00002422384,0.0002060138,0.4262628,0.0004240184,0.0004296096,0.0002305349,0.01744051,0.00006668914],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.993701,0.0003751082,0.0001751156,0.0007669152,0.00002002591,0.0000437864,0.00004386197,0.000006134587,0.004868101],"genre_scores_gemma":[0.9982877,0.0003289708,0.0001544255,0.0002051164,0.000004085814,0.00001813848,0.00003514812,0.000005001191,0.0009614222],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9686126,"threshold_uncertainty_score":0.2359044,"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."}}