{"id":"W237532803","doi":"10.3138/cjpe.26.006","title":"The Essential Skills Series in Evaluation: Assessing the Validity of the ESS Participant Workshop Evaluation Questionnaire","year":2011,"lang":"en","type":"article","venue":"Canadian Journal of Program Evaluation","topic":"Evaluation and Performance Assessment","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre de Liaison Sur l'Intervention et la Prévention Psychosociales; Université du Québec à Montréal; Université de Montréal","funders":"","keywords":"Psychology; Medical education; Applied psychology; Medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"high","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1516111,0.0004808201,0.0008418028,0.002235517,0.001069703,0.001341769,0.001446373,0.0007988392,0.001869239],"category_scores_gemma":[0.1929417,0.000486676,0.001288981,0.001879201,0.001581106,0.001736898,0.002859536,0.001089985,0.0002554449],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003283975,"about_ca_system_score_gemma":0.005469129,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002419481,"about_ca_topic_score_gemma":0.003617787,"domain_scores_codex":[0.8855783,0.07987706,0.01417302,0.001491408,0.01731781,0.001562458],"domain_scores_gemma":[0.7400023,0.1800595,0.0235845,0.007167635,0.04473337,0.004452685],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005717937,0.004540256,0.6783242,0.002896255,0.0007728072,0.0004223195,0.04285796,0.002544497,0.004593725,0.003554434,0.006718429,0.2470571],"study_design_scores_gemma":[0.001896695,0.0149781,0.9214504,0.001291443,0.0002610926,0.000431255,0.02028812,0.01059896,0.00590351,0.00191635,0.02076736,0.0002166363],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9739559,0.0002264744,0.01094326,0.0005254866,0.0001008674,0.008986772,0.0005402564,0.00006029265,0.004660718],"genre_scores_gemma":[0.9640502,0.0001196452,0.01690755,0.0001929891,0.00002910672,0.01758312,0.0003847755,0.00002578046,0.0007067688],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8483889,"threshold_uncertainty_score":0,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.518137140465873,"score_gpt":0.548237636298514,"score_spread":0.03010049583264096,"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."}}