{"id":"W3037108940","doi":"10.1016/j.jcjo.2020.05.005","title":"Evaluation of ophthalmology residents' self-assessments and peer assessments in simulated surgery","year":2020,"lang":"en","type":"article","venue":"Canadian Journal of Ophthalmology","topic":"Intraocular Surgery and Lenses","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"Kingston Health Sciences Centre; Queen's University","funders":"Queen's University","keywords":"Ophthalmology; Medicine; Optometry; Psychology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":false,"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":[0.003088804,0.0001664343,0.0007355298,0.0006006903,0.00004636736,0.00001384682,0.0001101801,0.0002356821,0.001794259],"category_scores_gemma":[0.001692958,0.0001625241,0.0001277501,0.0003580257,0.0001071481,0.0001550197,0.00001671672,0.0004066254,0.000005408834],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002424242,"about_ca_system_score_gemma":0.004183481,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001820821,"about_ca_topic_score_gemma":0.00002364829,"domain_scores_codex":[0.9971972,0.0007932937,0.0008238499,0.0002160324,0.0006018023,0.0003677865],"domain_scores_gemma":[0.9971874,0.000357291,0.0004177245,0.0001508965,0.001286246,0.0006003926],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002634026,0.0000941486,0.9281963,0.00009306793,0.0003505828,0.06902605,0.0003888128,0.0002766481,0.0004230041,0.00002716706,0.0004044398,0.000456335],"study_design_scores_gemma":[0.001758099,0.0009284123,0.8436266,0.000161725,0.0003872114,0.1504413,0.0002549936,0.001358689,0.00053932,0.0003000693,0.00008711615,0.00015651],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.992815,0.0007882407,0.000004496359,0.001019315,0.0004575927,0.0002165317,0.000008410906,0.000002585886,0.004687895],"genre_scores_gemma":[0.999424,0.000007187971,0.0002496168,0.0001027468,0.00009776178,0.000002118568,0.00001365033,0.00002129241,0.00008164706],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08456975,"threshold_uncertainty_score":0.9991182,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09118053369396946,"score_gpt":0.3767672411276636,"score_spread":0.2855867074336941,"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."}}