{"id":"W3011136232","doi":"10.36834/cmej.69276","title":"Black Ice: Five ways to get a grip on grouped self-assessments of competence for program evaluation","year":2020,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Competence (human resources); Self-assessment; Medical education; Applied psychology; Psychology; Computer science; Medicine; Social psychology","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.592704,0.00539459,0.009126624,0.04297097,0.006761929,0.02076689,0.009421383,0.008525942,0.01085642],"category_scores_gemma":[0.6965485,0.005340441,0.0135496,0.03473647,0.01246526,0.04525319,0.02817674,0.01570741,0.00419372],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00910838,"about_ca_system_score_gemma":0.03629575,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00369941,"about_ca_topic_score_gemma":0.01244109,"domain_scores_codex":[0.4503548,0.3536887,0.1190043,0.007255839,0.06621058,0.003485761],"domain_scores_gemma":[0.1535118,0.604084,0.04439585,0.04905354,0.143116,0.005838842],"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":[0.00197214,0.0005733987,0.02023714,0.074754,0.004852972,0.0001908073,0.02268763,0.0005027857,0.001112379,0.04393219,0.09382042,0.7353642],"study_design_scores_gemma":[0.004571836,0.003049155,0.04346747,0.2906836,0.009811671,0.0007261598,0.03657667,0.006181097,0.004823694,0.1985362,0.3996878,0.001884659],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01591602,0.05218692,0.7090086,0.1009194,0.01453998,0.06853666,0.007161574,0.003277647,0.02845319],"genre_scores_gemma":[0.01788969,0.006073179,0.9184608,0.007875113,0.0005710958,0.04646334,0.00114855,0.0004117001,0.00110659],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.592704,"threshold_uncertainty_score":0.5022686,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0433203498075143,"score_gpt":0.4075794729937572,"score_spread":0.3642591231862429,"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."}}