{"id":"W2588998449","doi":"10.4300/jgme-d-16-00694.1","title":"The Promise of Milestones: Are They Living Up to Our Expectations?","year":2017,"lang":"en","type":"article","venue":"Journal of Graduate Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; MEDLINE; Data science; Medical education; Psychology; Gerontology; Computer science; Biology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002176587,0.0001239367,0.0003193855,0.00025077,0.0004893313,0.00007086778,0.000540252,0.0001079994,0.00004235456],"category_scores_gemma":[0.05557628,0.00007880125,0.0001212189,0.0002022753,0.0001442055,0.0002071459,0.0000541921,0.00055099,0.00002274067],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001770842,"about_ca_system_score_gemma":0.002951419,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005025508,"about_ca_topic_score_gemma":0.00002272945,"domain_scores_codex":[0.9972117,0.00009845823,0.0009457066,0.0001355832,0.001409692,0.0001988645],"domain_scores_gemma":[0.9954864,0.0001642843,0.001480621,0.0005887913,0.002010132,0.0002697866],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003137861,0.001877034,0.03455833,0.000334634,0.0002191465,0.00001437754,0.02328653,0.000004013295,0.0006936459,0.001763821,0.3657413,0.5711933],"study_design_scores_gemma":[0.001938128,0.0009113182,0.8100641,0.008831653,0.0003888925,0.0009322737,0.1489965,0.0004789812,0.001389384,0.002978712,0.02274537,0.0003447291],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8333183,0.0002254014,0.0003825157,0.1603818,0.005030156,0.0003019052,6.06756e-7,0.000007977358,0.0003513976],"genre_scores_gemma":[0.993691,0.0001660166,0.00221913,0.001289923,0.001591218,0.00003579014,0.000003036352,0.00001885594,0.0009850113],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7755058,"threshold_uncertainty_score":0.952379,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05249932425210999,"score_gpt":0.4126720906755726,"score_spread":0.3601727664234626,"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."}}