{"id":"W2768578567","doi":"10.1080/0142159x.2017.1403015","title":"Creating a test blueprint for a progress testing program: A paired-comparisons approach","year":2017,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; University of British Columbia","funders":"University of British Columbia","keywords":"Blueprint; Test (biology); Computer science; Psychology; Medical physics; Medical education; Medicine; Engineering","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1656505,0.002188191,0.002729431,0.006698093,0.004391738,0.004657371,0.004704035,0.001841727,0.01339071],"category_scores_gemma":[0.3685876,0.001389918,0.002896789,0.004705205,0.004465112,0.005627892,0.005909872,0.005451159,0.00208729],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004022979,"about_ca_system_score_gemma":0.003531187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001057514,"about_ca_topic_score_gemma":0.001797814,"domain_scores_codex":[0.7612903,0.1858495,0.009164204,0.01335177,0.02768292,0.002661263],"domain_scores_gemma":[0.5873462,0.3128355,0.01371825,0.03244686,0.05066659,0.002986622],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.02674446,0.01169222,0.02834159,0.004699718,0.003129175,0.0006208564,0.0359685,0.01395748,0.02881431,0.06154423,0.0176803,0.7668071],"study_design_scores_gemma":[0.009880857,0.09163769,0.1473634,0.002906673,0.003700021,0.0009791964,0.04403123,0.2150786,0.1260978,0.2582358,0.09814199,0.001946732],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1738695,0.0002976185,0.7637087,0.000702765,0.001460076,0.02971975,0.001334723,0.002695482,0.02621132],"genre_scores_gemma":[0.2396643,0.00007479674,0.7210002,0.0002277397,0.0001499298,0.03597506,0.0004873008,0.0006529101,0.001767661],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.1656505,"threshold_uncertainty_score":0.8760538,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1127059462167929,"score_gpt":0.4215779581477646,"score_spread":0.3088720119309717,"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."}}