{"id":"W2159075585","doi":"10.3109/0142159x.2011.565831","title":"Performance in assessment: Consensus statement and recommendations from the Ottawa conference","year":2011,"lang":"en","type":"article","venue":"Medical Teacher","topic":"Innovations in Medical Education","field":"Medicine","cited_by":172,"is_retracted":false,"has_abstract":true,"ca_institutions":"Medical Council of Canada","funders":"Association for the Study of Medical Education","keywords":"Statement (logic); Consensus conference; MEDLINE; Library science; Medicine; Political science; Family medicine; Medical education; Computer science; Law","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001350138,0.00008635387,0.0001406985,0.00004842499,0.00007195323,0.000008572899,0.000105809,0.0000986864,0.01216329],"category_scores_gemma":[0.000943641,0.00005661793,0.00001324315,0.0001872996,0.0002669094,0.00003712205,0.00004398431,0.0005967813,0.00003033452],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008271118,"about_ca_system_score_gemma":0.0002873951,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003042131,"about_ca_topic_score_gemma":0.00007633412,"domain_scores_codex":[0.9987582,0.000118008,0.0003877629,0.0001783875,0.0003906263,0.0001669888],"domain_scores_gemma":[0.9994785,0.00006327601,0.00006821956,0.0002250832,0.00007477977,0.00009013315],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001665647,0.0003601998,0.8423083,0.00001699485,0.0000308056,0.000008964429,0.003469142,3.615706e-8,0.000008226242,0.002035135,0.03394836,0.1177972],"study_design_scores_gemma":[0.001476803,0.0001088572,0.9026608,0.0002874348,0.00003851032,0.0000183445,0.003283023,0.01934574,0.00001793844,0.000550992,0.07209825,0.0001132916],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9226234,0.0000428772,0.0003171623,0.04434396,0.0002652314,0.000239902,0.000001774978,0.00002561817,0.03214006],"genre_scores_gemma":[0.9923909,0.00009516622,0.004304709,0.002136421,0.0001191268,0.00006239999,0.00006439615,0.000008729362,0.0008181996],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1176839,"threshold_uncertainty_score":0.9887397,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06685978808781556,"score_gpt":0.3700311942472204,"score_spread":0.3031714061594048,"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."}}