{"id":"W3161561699","doi":"10.2196/23834","title":"Efficiency, Usability, and Outcomes of Proctored Next-Level Exams for Proficiency Testing in Primary Care Education: Observational Study","year":2021,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Usability; Observational study; Test (biology); Psychological intervention; Medical education; Supervisor; Psychology; Computer science; Medicine; Nursing","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004745034,0.00008950959,0.0002198805,0.0001605805,0.0006198148,0.00007870295,0.0002988599,0.0001841915,0.00001921551],"category_scores_gemma":[0.004424845,0.00007838908,0.00003418691,0.00116633,0.0004711817,0.0006048384,0.0001756266,0.0007175158,0.000002156268],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003753458,"about_ca_system_score_gemma":0.003677395,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001073542,"about_ca_topic_score_gemma":0.0006613922,"domain_scores_codex":[0.9972454,0.000689816,0.0004009317,0.0002575744,0.001020296,0.0003859699],"domain_scores_gemma":[0.9953265,0.00244878,0.00009449576,0.0001325388,0.001921904,0.00007580955],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001803568,0.0007526639,0.7336327,0.0003242354,0.000006576354,4.971563e-7,0.2290911,0.000002379088,0.0001263028,0.008432241,0.0001314665,0.02748183],"study_design_scores_gemma":[0.0002857253,0.0001702583,0.7688185,0.00007576551,0.000002133615,4.78935e-7,0.2276689,0.0001103627,0.0001214262,0.00225029,0.000420288,0.00007586404],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9860379,0.0001885039,0.00007570654,0.0008782741,0.0001014195,0.002030408,0.00003712071,0.00001256139,0.01063811],"genre_scores_gemma":[0.9977601,0.000008947869,0.001180713,0.00004460665,0.00005631602,0.0004890542,0.00003041229,0.000005063891,0.0004247616],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0351858,"threshold_uncertainty_score":0.652354,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3415603520781481,"score_gpt":0.4986200422298747,"score_spread":0.1570596901517266,"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."}}