{"id":"W4286470750","doi":"10.1002/aet2.10781","title":"Does direct observation influence the quality of workplace‐based assessment documentation?","year":2022,"lang":"en","type":"article","venue":"AEM Education and Training","topic":"Innovations in Medical Education","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Generalizability theory; Documentation; Reliability (semiconductor); Quality (philosophy); Psychology; Statistics; Medicine; Computer science; Mathematics","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":[],"consensus_categories":[],"category_scores_codex":[0.001406608,0.00007063385,0.0001289512,0.0001029075,0.0003268123,0.00002273903,0.00007058822,0.00002233642,0.0003609585],"category_scores_gemma":[0.0004561243,0.00004583628,0.00002754008,0.0005471013,0.00008798917,0.00011979,0.00001992893,0.0002039269,8.174827e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001358545,"about_ca_system_score_gemma":0.0008634568,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000107644,"about_ca_topic_score_gemma":0.000005649089,"domain_scores_codex":[0.9988073,0.0001596361,0.0003758383,0.0001598423,0.0004005869,0.00009678607],"domain_scores_gemma":[0.9991063,0.0002058495,0.0002402097,0.0002079818,0.0002098702,0.00002973863],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00008743112,0.0009908022,0.5861516,0.0002623852,0.00007523399,3.161831e-7,0.0171437,0.0005621593,0.002701547,0.02798253,0.002786315,0.361256],"study_design_scores_gemma":[0.0004918815,0.00006739744,0.9626172,0.00004356804,0.00003960474,0.000002338361,0.02556772,0.0007488271,0.0001936668,0.0007228967,0.009430843,0.00007405084],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.977132,0.00004359254,0.0003362496,0.01924681,0.0006400439,0.0003531192,0.000003346576,0.00002626206,0.002218618],"genre_scores_gemma":[0.9864939,0.000006115752,0.007576413,0.004752171,0.00008036401,0.0003095902,0.0001443374,0.000007853775,0.0006292356],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3764656,"threshold_uncertainty_score":0.3952243,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05679538406070442,"score_gpt":0.4349977433329142,"score_spread":0.3782023592722098,"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."}}