{"id":"W4406973300","doi":"10.1016/j.bjps.2025.01.040","title":"Technical skills assessment during simulated cleft lip repair","year":2025,"lang":"en","type":"article","venue":"Journal of Plastic Reconstructive & Aesthetic Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Psychology; Orthodontics; Computer science; Dentistry; 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":[],"consensus_categories":[],"category_scores_codex":[0.0008926542,0.0002250328,0.0008813105,0.000697414,0.0001125836,0.00003101838,0.00008283005,0.0001813229,0.0004166338],"category_scores_gemma":[0.001628503,0.0001789568,0.0008356888,0.0005499406,0.000222876,0.0001340157,0.00003651995,0.0006803831,0.00001049871],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002810711,"about_ca_system_score_gemma":0.0006454976,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000216699,"about_ca_topic_score_gemma":5.136981e-7,"domain_scores_codex":[0.9975229,0.0001705861,0.001268887,0.0002546594,0.0004634454,0.0003195427],"domain_scores_gemma":[0.9950336,0.003480837,0.0005417796,0.0002165117,0.0004708213,0.0002564062],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001023826,0.0005751365,0.9383258,0.0001029384,0.0003900584,0.001022671,0.00005216392,0.001259972,0.0005365275,0.000446389,0.0003732653,0.05589126],"study_design_scores_gemma":[0.004283843,0.0001807103,0.982004,0.001864838,0.0003717262,0.005659312,0.0002233432,0.001884477,0.000229518,0.0005965675,0.002464982,0.0002366436],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9914783,0.00007644234,0.001064036,0.0002599419,0.001464199,0.0001543729,0.000002329433,0.00008909993,0.005411277],"genre_scores_gemma":[0.9978356,0.00004302672,0.001720199,0.0001089408,0.0001105198,0.000001906991,0.000002324776,0.00002032534,0.0001571785],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05565462,"threshold_uncertainty_score":0.7297648,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01290829797433353,"score_gpt":0.2973943502243801,"score_spread":0.2844860522500466,"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."}}