{"id":"W4413970718","doi":"10.1007/s00464-025-12076-7","title":"Artificial intelligence for automatic FLS credentialing: highlighting and addressing current limitations","year":2025,"lang":"en","type":"article","venue":"Surgical Endoscopy","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; University Health Network","funders":"European Association for Endoscopic Surgery and other Interventional Techniques; Agence Nationale de la Recherche; Bpifrance","keywords":"Credentialing; Current (fluid); Artificial intelligence; Computer science; Psychology; Engineering ethics; Cognitive science; Medicine; Engineering; Medical education; Electrical 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01544164,0.0007079192,0.0009251173,0.003266032,0.0009859096,0.008445484,0.004682596,0.002184902,0.01035597],"category_scores_gemma":[0.05521439,0.0003968714,0.0006438902,0.001859185,0.002108718,0.007728869,0.003103537,0.003341187,0.004395668],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001362605,"about_ca_system_score_gemma":0.003294698,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004586839,"about_ca_topic_score_gemma":0.005659772,"domain_scores_codex":[0.9872994,0.005796004,0.0008065309,0.001419058,0.004213588,0.0004653817],"domain_scores_gemma":[0.9323446,0.04644921,0.001860419,0.008742496,0.009489814,0.00111342],"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":[0.0002942042,0.0003243341,0.009319027,0.001118516,0.000121162,0.00007645482,0.0003977268,0.01224897,0.001991159,0.06058076,0.02182573,0.891702],"study_design_scores_gemma":[0.00007664008,0.0003099545,0.01131748,0.002385281,0.0001170875,0.0006830211,0.001508937,0.5939691,0.00748191,0.2267985,0.1551475,0.0002044488],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05851846,0.0261337,0.7317704,0.06581412,0.001679614,0.0004416897,0.001815495,0.005218477,0.108608],"genre_scores_gemma":[0.6885607,0.006201453,0.293971,0.003120351,0.0008457642,0.0001626341,0.001145454,0.0001985837,0.005794066],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01544164,"threshold_uncertainty_score":0.0816642,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1718808348192686,"score_gpt":0.4199129874157317,"score_spread":0.2480321525964631,"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."}}