{"id":"W2104891090","doi":"10.2106/jbjs.m.01491","title":"The Validity of Level of Evidence Ratings of Articles Submitted to JBJS","year":2015,"lang":"en","type":"article","venue":"Journal of Bone and Joint Surgery","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hospital for Sick Children","funders":"","keywords":"Psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3114654,0.0009019609,0.003884936,0.03640974,0.00179981,0.008137873,0.00392967,0.002920598,0.002227385],"category_scores_gemma":[0.7468889,0.001224075,0.00595308,0.01555546,0.004794539,0.005115937,0.005613051,0.002017433,0.0008569505],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004061935,"about_ca_system_score_gemma":0.005291898,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002115603,"about_ca_topic_score_gemma":0.00291702,"domain_scores_codex":[0.5558174,0.1315102,0.1670214,0.01429768,0.128685,0.002668259],"domain_scores_gemma":[0.09840994,0.5553772,0.1655926,0.02000804,0.1562375,0.004374783],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003353354,0.000207353,0.8608756,0.0210191,0.008878134,0.0003774841,0.00669911,0.0008234889,0.001530771,0.00124623,0.004661398,0.09032802],"study_design_scores_gemma":[0.0008235067,0.002792656,0.9188753,0.02508085,0.005437991,0.002545951,0.00772467,0.006304295,0.002947781,0.00562886,0.02133117,0.0005069135],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7580951,0.1502244,0.03386139,0.008685347,0.005920361,0.005439911,0.007508911,0.0003492542,0.02991531],"genre_scores_gemma":[0.9686317,0.00890612,0.01680917,0.000700388,0.001096059,0.001400412,0.001910438,0.00008798957,0.0004577501],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6885346,"threshold_uncertainty_score":0.8490858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9723956434621527,"score_gpt":0.5521466201376565,"score_spread":0.4202490233244962,"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."}}