{"id":"W4383904828","doi":"10.7759/cureus.41702","title":"The Pass/Fail Effect: A Longitudinal Study of United States Medical Licensing Examination (USMLE) Step 1 Performance Over a Decade","year":2023,"lang":"en","type":"article","venue":"Cureus","topic":"Innovations in Medical Education","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; United States Medical Licensing Examination; Descriptive statistics; Test (biology); Family medicine; Demography; Medical education; Medical school; Statistics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002110626,0.0001308293,0.0002123228,0.000240648,0.0002664611,0.00002113226,0.0001658554,0.0001178803,0.0001085414],"category_scores_gemma":[0.002831121,0.00008794418,0.00003056025,0.001961211,0.000156758,0.00009239985,0.00007540992,0.0003842829,0.00003630845],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001426935,"about_ca_system_score_gemma":0.0001245947,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001394941,"about_ca_topic_score_gemma":0.00008880239,"domain_scores_codex":[0.9978666,0.0001624322,0.0004259159,0.0002119819,0.001065953,0.0002670597],"domain_scores_gemma":[0.9984873,0.0005503429,0.0001468821,0.0003734778,0.0003661515,0.00007583166],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003903355,0.001496365,0.6004501,0.0006001788,0.0004764514,0.00008165785,0.02052154,0.00008290237,0.0004002187,0.0006044126,0.04415145,0.3307444],"study_design_scores_gemma":[0.001753885,0.001107969,0.9346761,0.0002536328,0.0001142159,0.0000223711,0.005177649,0.05104025,0.0001975507,0.00001369342,0.00554314,0.00009952496],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9970617,0.00007261609,0.0001306911,0.001249287,0.0005591267,0.0005972177,0.000001117016,0.0000993436,0.0002289315],"genre_scores_gemma":[0.9988124,0.0001373494,0.00008043674,0.0001967096,0.0002362982,0.00006585215,0.0001251095,0.00002162766,0.0003241804],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.334226,"threshold_uncertainty_score":0.3586259,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02612269425499893,"score_gpt":0.3535088670476444,"score_spread":0.3273861727926455,"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."}}