{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01066506,0.0006544848,0.0004946542,0.001430783,0.0002105785,0.0007892704,0.0007825002,0.00055522,0.00519662],"category_scores_gemma":[0.02463788,0.0001956014,0.0008016099,0.001236473,0.0004529942,0.000896818,0.001499277,0.0008318793,0.001675473],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003368349,"about_ca_system_score_gemma":0.000545062,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004459989,"about_ca_topic_score_gemma":0.0006072512,"domain_scores_codex":[0.9890458,0.004772781,0.001207185,0.0006534805,0.00397757,0.0003430726],"domain_scores_gemma":[0.9821892,0.009260037,0.003065632,0.0008769456,0.003938804,0.0006693947],"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":[0.007466769,0.0009670073,0.3706282,0.001104504,0.0003249314,0.0003221711,0.00168042,0.003604814,0.0126277,0.001885836,0.006513082,0.5928745],"study_design_scores_gemma":[0.0005955042,0.009614172,0.9475114,0.0006092743,0.0001382331,0.001895686,0.000660271,0.007156393,0.007971489,0.00219706,0.02151528,0.000135303],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9073014,0.004199932,0.06173956,0.0004103873,0.0009334059,0.001993607,0.002053361,0.0006696432,0.02069872],"genre_scores_gemma":[0.9149501,0.001262218,0.07472356,0.000113912,0.0001893428,0.002458232,0.001884893,0.00009672409,0.004320976],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01066506,"threshold_uncertainty_score":0.05640292,"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."}}