{"id":"W4380762580","doi":"10.4300/jgme-d-22-00783.1","title":"Designing a User-Friendly Context-Specific Assessment Tool for Community-Based Teachers","year":2023,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Rubric; Summative assessment; Competence (human resources); Medical education; Computer science; Context (archaeology); Formative assessment; 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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006892609,0.0001695178,0.0004087843,0.0006743516,0.0003372025,0.00004886594,0.0003019063,0.0001852041,0.000161016],"category_scores_gemma":[0.009322973,0.0001405339,0.0001892258,0.001057,0.0001998639,0.0001878935,0.0000236528,0.001493734,0.00002283207],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005125988,"about_ca_system_score_gemma":0.006988193,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001991283,"about_ca_topic_score_gemma":0.000002063558,"domain_scores_codex":[0.9966702,0.0003449005,0.001072903,0.0001480023,0.001456028,0.0003079121],"domain_scores_gemma":[0.9965528,0.0006647786,0.0005729386,0.0003360276,0.001637536,0.000235944],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001651519,0.001494291,0.002886544,0.0002574711,0.00009021543,0.000009287393,0.001796779,0.00001242061,0.0009393621,0.001544381,0.6637074,0.3270967],"study_design_scores_gemma":[0.01202267,0.00303075,0.06419271,0.003495825,0.0004624577,0.0004170914,0.03485941,0.004473413,0.001976964,0.00260434,0.8719288,0.0005355543],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8039795,0.00009729407,0.1021504,0.08931798,0.003418397,0.0007346673,0.000002406233,0.00007415712,0.0002252529],"genre_scores_gemma":[0.914611,0.00009848135,0.07114004,0.01155657,0.001489769,0.000173884,0.0003095986,0.00005360973,0.0005670066],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3265612,"threshold_uncertainty_score":0.9990219,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.083783754459524,"score_gpt":0.4152524694347127,"score_spread":0.3314687149751887,"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."}}