{"id":"W2061945763","doi":"10.1111/j.1365-2923.2009.03567.x","title":"Cross‐institutional progress testing: feasibility and value to a new medical school","year":2010,"lang":"en","type":"article","venue":"Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Medical education; Quality assurance; Medical school; Test (biology); Quality (philosophy); Medicine; Graduate medical education; Psychology; Family medicine; Accreditation","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","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001886482,0.0001919918,0.0002564283,0.0002332088,0.000232615,0.00007292614,0.0002656556,0.0004824513,0.007314174],"category_scores_gemma":[0.1354489,0.0001648054,0.00004194248,0.0009664472,0.0007498438,0.0001657679,0.0001046322,0.001303781,0.0002134311],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002348341,"about_ca_system_score_gemma":0.02492433,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002514682,"about_ca_topic_score_gemma":0.00003841721,"domain_scores_codex":[0.9964111,0.00004980763,0.0006031308,0.0005498407,0.002033127,0.00035298],"domain_scores_gemma":[0.99642,0.0001480814,0.0001107047,0.0005812883,0.0007875154,0.001952414],"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":[0.00006622891,0.001336028,0.3008949,0.0001434576,0.0000165745,0.0000092546,0.0002779485,3.961768e-7,0.0001659775,0.005376143,0.09023195,0.6014811],"study_design_scores_gemma":[0.001670213,0.0002284783,0.7823301,0.0006005348,0.00005008869,0.001009888,0.00009223139,0.001887113,0.0001316103,0.001603372,0.2101241,0.0002722974],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8895123,0.0001593736,0.001342781,0.1024597,0.003220799,0.0007695657,0.000001351083,0.0001371185,0.002397034],"genre_scores_gemma":[0.8436145,0.00001283715,0.08663318,0.0617374,0.005430748,0.0002497541,0.0001579627,0.00003634644,0.002127273],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6012088,"threshold_uncertainty_score":0.9935933,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02835180537102516,"score_gpt":0.413599715015092,"score_spread":0.3852479096440668,"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."}}