{"id":"W2085655721","doi":"10.3402/meo.v20.25819","title":"Extended score interval in the assessment of basic surgical skills","year":2015,"lang":"en","type":"article","venue":"Medical Education Online","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre for Family Medicine","funders":"","keywords":"Ceiling effect; Intraclass correlation; Confidence interval; Physical therapy; Medicine; Knot tying; Reliability (semiconductor); Surgery; Psychometrics; Clinical psychology; Internal medicine","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00112865,0.00008197974,0.0002328048,0.0000948981,0.00001393569,0.000006298468,0.0001254903,0.00008879286,0.0014647],"category_scores_gemma":[0.001703873,0.00004922391,0.00006638519,0.0003407353,0.0001069803,0.00003529785,0.00002361082,0.0003259516,0.0000085213],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006306883,"about_ca_system_score_gemma":0.001642076,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003073083,"about_ca_topic_score_gemma":0.00002236198,"domain_scores_codex":[0.9983061,0.0001567229,0.0004133606,0.0001479191,0.0008428444,0.0001330556],"domain_scores_gemma":[0.9989899,0.0002963475,0.00007141741,0.0001899179,0.0001501511,0.0003022132],"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":[0.000205157,0.02471672,0.3252671,0.00009481019,0.00004786545,0.0001535837,0.002872472,0.00003090788,0.000004948699,0.01010599,0.003385574,0.6331149],"study_design_scores_gemma":[0.007147167,0.0003368455,0.8718747,0.0004390134,0.00003544444,0.0001575251,0.002942997,0.009069179,0.000009774686,0.0006097106,0.1072709,0.0001067617],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9699172,0.000115142,0.000105215,0.01799454,0.000398337,0.0002074994,0.000002714026,0.00001974864,0.01123954],"genre_scores_gemma":[0.9953259,0.00001832887,0.0007005013,0.003083028,0.0003855491,0.00001430941,0.0001822891,0.000007070559,0.0002830149],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6330081,"threshold_uncertainty_score":0.9994481,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05806162525443016,"score_gpt":0.4367311720885782,"score_spread":0.3786695468341481,"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."}}