{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009675757,0.000470517,0.0009635194,0.001719534,0.001294493,0.001409037,0.0008566076,0.001127638,0.001737017],"category_scores_gemma":[0.03836421,0.0006523214,0.001659202,0.001070998,0.0009262541,0.001244967,0.001404925,0.001525264,0.0004607424],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001214131,"about_ca_system_score_gemma":0.001633085,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003758444,"about_ca_topic_score_gemma":0.005922105,"domain_scores_codex":[0.9913819,0.003915907,0.0012522,0.0008197399,0.001926046,0.0007041705],"domain_scores_gemma":[0.9668028,0.0148542,0.00838245,0.001898801,0.004698338,0.003363299],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005172779,0.005478074,0.9655485,0.0003967066,0.0002242904,0.0003421183,0.01123214,0.000114505,0.0004784492,0.00003120175,0.0007785627,0.01485829],"study_design_scores_gemma":[0.00007427068,0.005800279,0.9854802,0.00009854913,0.0001248389,0.0002626208,0.006595134,0.0005235857,0.0002883063,0.00003331352,0.0006737623,0.00004517169],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9989955,0.0001004922,0.0001306667,0.00003891414,0.000005618268,0.0002273088,0.0001341142,0.000006654594,0.0003608099],"genre_scores_gemma":[0.9987746,0.0001138176,0.0004421295,0.00006005717,0.000015033,0.0002604486,0.0001564054,0.000006185303,0.0001713539],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009675757,"threshold_uncertainty_score":0.05117089,"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."}}