{"id":"W2167376439","doi":"10.1080/01421590220134088","title":"Evolution of student assessment in McMaster University’s MD Programme","year":2002,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Curriculum; Medical education; Educational measurement; Computer science; Psychology; Medicine; Pedagogy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007408331,0.00007790069,0.0001873359,0.0001962467,0.00002140649,0.000003008184,0.0001206104,0.0001601637,0.01508218],"category_scores_gemma":[0.0004192566,0.00006828288,0.00004008333,0.0005703398,0.0001571991,0.00004941463,0.00004372214,0.0005269112,0.00004369308],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006232407,"about_ca_system_score_gemma":0.0001407177,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005845679,"about_ca_topic_score_gemma":0.00001017469,"domain_scores_codex":[0.9983781,0.00008337377,0.0002922995,0.0001698746,0.0008681226,0.0002082138],"domain_scores_gemma":[0.9995557,0.00001286434,0.00006435023,0.0001693104,0.00008989433,0.0001078475],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001888433,0.003356863,0.8108444,0.00008246613,0.00003964987,0.00004864609,0.001438615,0.0000012856,0.00004702526,0.003258685,0.02489635,0.1559672],"study_design_scores_gemma":[0.007537514,0.0006641712,0.764038,0.0006941807,0.0001263351,0.00005561739,0.01016313,0.04630667,0.00002077875,0.0001716812,0.1699346,0.0002872575],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9514009,0.0001434791,0.001216854,0.007610465,0.0002145161,0.0003434937,2.264223e-7,0.00003792458,0.03903217],"genre_scores_gemma":[0.9906291,0.00002790562,0.001351886,0.0001997023,0.0001511051,0.000008275825,0.000007785984,0.000008990904,0.007615217],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1556799,"threshold_uncertainty_score":0.9858181,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03904979777277599,"score_gpt":0.3470801764318788,"score_spread":0.3080303786591028,"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."}}