{"id":"W2808722003","doi":"10.1373/jalm.2018.027037","title":"A Tautology Worth Repeating: Well-Characterized, Analytically Robust Assays Underpin Reliable Clinical Performance","year":2018,"lang":"en","type":"article","venue":"The Journal of Applied Laboratory Medicine","topic":"Prostate Cancer Diagnosis and Treatment","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Mount Sinai Hospital","funders":"","keywords":"Tautology (logic); Computer science; Reliability engineering; Artificial intelligence; 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":[],"consensus_categories":[],"category_scores_codex":[0.003497032,0.0002550764,0.0009793459,0.0001490553,0.0001702616,0.00001078899,0.0002457746,0.0001482315,0.0006041247],"category_scores_gemma":[0.0002457833,0.0001365669,0.00009486611,0.0005471444,0.0007437267,0.00008393902,0.00005690695,0.0007303056,0.00007649148],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000170349,"about_ca_system_score_gemma":0.0006134072,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006807589,"about_ca_topic_score_gemma":0.000006249837,"domain_scores_codex":[0.9973279,0.0001193502,0.001364377,0.0002431041,0.0005701904,0.0003750692],"domain_scores_gemma":[0.9969282,0.0003613968,0.0009620335,0.0005655692,0.000820093,0.0003627102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.06896949,0.005203579,0.351844,0.001053634,0.008212576,0.001960439,0.007232474,0.0003783703,0.06354737,0.00743135,0.4071641,0.07700267],"study_design_scores_gemma":[0.04826205,0.04337101,0.5137815,0.005003527,0.008598709,0.001739275,0.003614844,0.002478509,0.02408128,0.0009112948,0.3470802,0.001077831],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9743779,0.0009661283,0.0003379802,0.009804439,0.0008973841,0.0004353256,0.000004149785,0.00003578525,0.01314096],"genre_scores_gemma":[0.9858093,0.004307007,0.001278521,0.005882253,0.002478528,0.00001029881,0.000006885999,0.00003722149,0.0001900248],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1619375,"threshold_uncertainty_score":0.6614742,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03609169985436481,"score_gpt":0.3151984224260348,"score_spread":0.27910672257167,"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."}}