{"id":"W7147536741","doi":"10.4038/seajme.v19i2.682","title":"Performance of Family Medicine Residents on Difficult Test Questions","year":2025,"lang":"","type":"article","venue":"South-East Asian Journal of Medical Education","topic":"Personality Disorders and Psychopathology","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Test (biology); Personality; Big Five personality traits; MEDLINE; Personality Assessment Inventory","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003632022,0.0003141844,0.0003534346,0.001062035,0.0004645014,0.000509575,0.0003805787,0.0003803286,0.005570887],"category_scores_gemma":[0.02868613,0.000130145,0.0004815329,0.0004676183,0.0004588503,0.0006922201,0.0008234269,0.0004178681,0.0008106356],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007899322,"about_ca_system_score_gemma":0.0008729089,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005051313,"about_ca_topic_score_gemma":0.006807832,"domain_scores_codex":[0.9978541,0.000752261,0.0003477224,0.0002361444,0.0005625132,0.0002472523],"domain_scores_gemma":[0.9785042,0.009099935,0.006342303,0.0004972005,0.003439793,0.002116492],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004879113,0.000157461,0.9714171,0.0001312753,0.0000861623,0.0001588011,0.001676814,0.0003167572,0.0005076408,0.00004523181,0.002008422,0.02300644],"study_design_scores_gemma":[0.00001272864,0.0006718906,0.9959493,0.00004895017,0.00001903861,0.0002303939,0.001004294,0.0003112031,0.0004951172,0.0000443172,0.001202651,0.00000998544],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9960685,0.0002494862,0.0002070709,0.000241583,0.00002778492,0.00006708415,0.000598507,0.00002012753,0.002519923],"genre_scores_gemma":[0.998487,0.0001046834,0.0004415867,0.00005087928,0.00001675726,0.0000394909,0.0004231685,0.000003679526,0.0004327206],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005570887,"threshold_uncertainty_score":0.01920819,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02415642348990075,"score_gpt":0.3759198439739978,"score_spread":0.351763420484097,"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."}}