{"id":"W2080520080","doi":"10.1353/pbm.0.0079","title":"Making the Grade: Assuring Trustworthiness in Evidence","year":2009,"lang":"en","type":"article","venue":"Perspectives in biology and medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Credibility; Normative; Context (archaeology); Evidence-based medicine; Interpretation (philosophy); Scientific evidence; Soundness; Trustworthiness; Health care; Engineering ethics; Epistemology; Relevance (law); Articulation (sociology); Psychology; MEDLINE; Political science; Social psychology; Law; Computer science; Engineering; Philosophy","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":[0.005784546,0.0001305025,0.0003448563,0.0001807307,0.00034969,0.000002273073,0.0001838597,0.0002131339,0.0001094482],"category_scores_gemma":[0.00787033,0.00007862564,0.00001447034,0.0003847592,0.000424952,0.000163974,0.00006469881,0.00140918,0.000006396451],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001927844,"about_ca_system_score_gemma":0.0001277478,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001478858,"about_ca_topic_score_gemma":0.002201407,"domain_scores_codex":[0.995976,0.002647869,0.0004753797,0.0003667036,0.0001004092,0.0004336924],"domain_scores_gemma":[0.9950784,0.004341381,0.0001900971,0.0002487274,0.00008190161,0.00005951554],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001625088,0.00004974426,0.8158006,0.00007844392,0.000007574128,0.00002189048,0.0776863,0.0000012967,0.0001653621,0.09730978,0.0001856764,0.008530833],"study_design_scores_gemma":[0.0007503649,0.0001704273,0.909368,0.001166468,0.000008057063,0.000008408422,0.05921268,0.00004848992,0.000001403414,0.02757896,0.001603516,0.00008321134],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7657474,0.0278462,0.00007104751,0.2017935,0.0004614152,0.0005234986,8.460528e-7,0.00003457016,0.00352152],"genre_scores_gemma":[0.9862262,0.005235028,0.0000863786,0.007786405,0.0005268251,0.00004154925,0.000001225687,0.000005758671,0.00009061905],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2204788,"threshold_uncertainty_score":0.9422089,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7328428287410091,"score_gpt":0.6584308737913167,"score_spread":0.07441195494969244,"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."}}