{"id":"W4386072775","doi":"10.1007/s00464-023-10311-7","title":"New generation evaluations: video-based surgical assessments","year":2023,"lang":"en","type":"review","venue":"Surgical Endoscopy","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Upload; Medicine; Resource (disambiguation); Quality (philosophy); The Internet; Grading (engineering); Process (computing); Multimedia; Strengths and weaknesses; Computer science; World Wide Web; Engineering","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001009413,0.0006678285,0.002440699,0.000468049,0.0001724675,0.0001215949,0.0002067796,0.000711121,0.006483861],"category_scores_gemma":[0.0003388711,0.0005085485,0.001260063,0.001356932,0.00006960118,0.00009887069,0.00007124442,0.0008803312,0.001944645],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003318192,"about_ca_system_score_gemma":0.001741781,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003846809,"about_ca_topic_score_gemma":0.000004601554,"domain_scores_codex":[0.9951171,0.0005861632,0.001401694,0.0009417714,0.001343522,0.0006097854],"domain_scores_gemma":[0.996058,0.002009059,0.0003789722,0.0006164471,0.000165722,0.0007717825],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001515549,0.0001397024,0.00001360931,0.001847306,0.0003019164,0.001129048,0.000005367657,0.0001956945,8.271719e-8,0.0006629392,0.00093795,0.9946148],"study_design_scores_gemma":[0.01690172,0.0001857281,0.000003625585,0.005060161,0.001236494,0.0001028963,0.000002612461,0.008407068,0.000002235392,0.00003286044,0.9676445,0.0004201183],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00005096829,0.9847434,0.0002901658,0.0001578531,0.001054466,0.001764794,0.00006127689,0.0006276178,0.01124947],"genre_scores_gemma":[0.00008651165,0.9852962,0.001300656,0.0001168893,0.002528233,0.0002485811,0.003975765,0.0001841353,0.006262986],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9941947,"threshold_uncertainty_score":0.9997366,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2576376882106292,"score_gpt":0.5105531471863018,"score_spread":0.2529154589756726,"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."}}