{"id":"W2979396087","doi":"10.1002/bjs.11359","title":"Development and evaluation of the General Surgery Objective Structured Assessment of Technical Skill (GOSATS)","year":2019,"lang":"en","type":"article","venue":"British journal of surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"St. Michael's Hospital; Mount Sinai Hospital; University of Toronto","funders":"Royal College of Physicians and Surgeons of Canada","keywords":"Medicine; Summative assessment; Cronbach's alpha; Reliability (semiconductor); Specialty; Medical physics; Certification; Formative assessment; Medical education; Family medicine; Psychometrics; Clinical psychology; Psychology; Mathematics education","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.002939,0.00007481937,0.0004971865,0.0001261523,0.00003292675,0.00001115807,0.0000395627,0.00007211772,0.0002243913],"category_scores_gemma":[0.000585373,0.00005911597,0.0002126391,0.0001921654,0.00006261147,0.00008118023,0.00002650536,0.0002033589,2.350292e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009657836,"about_ca_system_score_gemma":0.0008244836,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001055627,"about_ca_topic_score_gemma":0.000006251232,"domain_scores_codex":[0.9979023,0.0001758115,0.0007738805,0.00009920217,0.0009359423,0.0001128411],"domain_scores_gemma":[0.9979286,0.0007057177,0.0005767139,0.00008511309,0.0006297731,0.00007405119],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00010377,0.0001353997,0.7309485,0.00007426207,0.0001738134,0.00003824473,0.0001135046,0.0002841217,0.002544996,0.00003912983,0.0001514759,0.2653928],"study_design_scores_gemma":[0.001039081,0.0000150911,0.9943219,0.0007407422,0.00008499062,0.0008001906,0.0001286992,0.0002795206,0.002184434,0.00006147926,0.0002836629,0.00006023108],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9980756,0.0004813594,0.00002114419,0.00005536505,0.0003540262,0.0001590062,0.000003063452,0.000003751188,0.000846709],"genre_scores_gemma":[0.9988459,0.00007155756,0.0009198986,0.00005951609,0.00006786082,0.000001548055,0.000004169205,0.000008818723,0.00002072519],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2653325,"threshold_uncertainty_score":0.2456928,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05304400114586332,"score_gpt":0.3363322992370958,"score_spread":0.2832882980912325,"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."}}