{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002134166,0.0002145608,0.0004172009,0.00009227005,0.0007405254,0.0001384203,0.000455499,0.0003047088,0.0002755049],"category_scores_gemma":[0.0620822,0.0001694833,0.0001071414,0.0002178643,0.0007069331,0.0001017303,0.0001526184,0.0008266544,0.00002512889],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001327159,"about_ca_system_score_gemma":0.000508138,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005577058,"about_ca_topic_score_gemma":0.000002739035,"domain_scores_codex":[0.9974863,0.00004126609,0.0005540655,0.0004885345,0.0008978027,0.0005320293],"domain_scores_gemma":[0.9979705,0.0001385774,0.0003203552,0.0008737923,0.0003848652,0.0003119557],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002574754,0.002732655,0.5099751,0.0002853873,0.00004944269,0.00001140074,0.0005053807,1.969291e-7,0.00001297406,0.0003446916,0.01159959,0.4744574],"study_design_scores_gemma":[0.004918792,0.001005311,0.2389321,0.001515707,0.0002968589,0.0002825392,0.001545417,0.6769711,0.00006624276,0.0003261309,0.07361122,0.0005286225],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8656584,0.0001946949,0.01928245,0.02980149,0.0006379965,0.005231343,0.000003175592,0.001014374,0.07817606],"genre_scores_gemma":[0.7804297,0.000001201729,0.2136706,0.0003524111,0.001395472,0.002260715,0.00005515545,0.00004745721,0.00178735],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6769709,"threshold_uncertainty_score":0.9458182,"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."}}