{"id":"W3004481391","doi":"10.1080/10401334.2019.1708365","title":"The Quality of Assessment of Learning (Qual) Score: Validity Evidence for a Scoring System Aimed at Rating Short, Workplace-Based Comments on Trainee Performance","year":2020,"lang":"en","type":"article","venue":"Teaching and Learning in Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":54,"is_retracted":false,"has_abstract":true,"ca_institutions":"Impact; McMaster University","funders":"","keywords":"Quality (philosophy); Psychology; Medical education; Scoring system; Medicine; Applied psychology; Internal medicine","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01530138,0.000169032,0.0005716019,0.0001317719,0.0006208272,0.00000874649,0.0001316384,0.00007933848,0.000006660884],"category_scores_gemma":[0.0266901,0.0001194203,0.00004908717,0.0003300151,0.0002515905,0.00005904585,0.00004106853,0.001795815,2.832169e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002812466,"about_ca_system_score_gemma":0.0001447262,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008263753,"about_ca_topic_score_gemma":0.000003139877,"domain_scores_codex":[0.9966223,0.001136613,0.0009935668,0.0002951109,0.00070165,0.0002507609],"domain_scores_gemma":[0.9943451,0.004719644,0.0004978063,0.0002025694,0.0001538678,0.00008103479],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000940125,0.0001035973,0.894215,0.004922206,0.00004952536,0.000001797662,0.01169173,0.008619671,0.008163476,0.0005151947,0.0002494157,0.07052828],"study_design_scores_gemma":[0.004211961,0.003906451,0.4032071,0.02062525,0.0001421027,0.00000686217,0.02174134,0.5442761,0.0006572499,0.00000324165,0.001002202,0.0002200937],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9815954,0.0001776579,0.01005894,0.007005538,0.0001676055,0.0005993117,4.170144e-7,0.00006353336,0.0003315945],"genre_scores_gemma":[0.9948766,0.0000222931,0.004217218,0.000450023,0.0002430545,0.00005948307,0.00002625231,0.00002134588,0.00008378919],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5356565,"threshold_uncertainty_score":0.9815085,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2124900937935385,"score_gpt":0.4522771206867789,"score_spread":0.2397870268932404,"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."}}