{"id":"W2884774198","doi":"10.1080/0142159x.2018.1481281","title":"Twelve tips for developing key-feature questions (KFQ) for effective assessment of clinical reasoning","year":2018,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; University of Toronto","funders":"","keywords":"Key (lock); Feature (linguistics); Psychology; Computer science; Medical education; Medicine; Philosophy; Linguistics","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":[0.004332645,0.0001932005,0.0008310226,0.00005936642,0.0001181957,0.00001153897,0.0001578933,0.0005754054,0.0002826502],"category_scores_gemma":[0.2667919,0.0001434954,0.0004043938,0.0001632486,0.0004576262,0.00003258488,0.00006874829,0.0006198884,0.00001323287],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009186829,"about_ca_system_score_gemma":0.0009283542,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002785825,"about_ca_topic_score_gemma":0.00002021435,"domain_scores_codex":[0.9975922,0.0001948273,0.0007776859,0.0004796559,0.000567939,0.0003877128],"domain_scores_gemma":[0.9809565,0.01725206,0.0002733976,0.0003805904,0.0006009619,0.0005365146],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001129088,0.001900393,0.4714592,0.0003266709,0.0008313658,0.00002807808,0.000507535,7.92311e-7,0.00003300881,0.01885036,0.1574947,0.3474388],"study_design_scores_gemma":[0.0197446,0.007379255,0.6637546,0.01556628,0.001305833,0.0001033118,0.0002774826,0.04070842,0.0003715604,0.003979719,0.2461174,0.0006915464],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4902566,0.0004367989,0.4662194,0.02390434,0.002622298,0.003666149,0.0000412273,0.0002456495,0.01260753],"genre_scores_gemma":[0.8599719,0.00006012734,0.1301373,0.001777666,0.004198289,0.0005466919,0.0001267194,0.00005811886,0.003123141],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3697153,"threshold_uncertainty_score":0.7393842,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05833185039754752,"score_gpt":0.4903663543018182,"score_spread":0.4320345039042707,"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."}}