{"id":"W4306181334","doi":"10.1136/bmjebm-2022-112065","title":"The Pandora’s Box of Evidence Synthesis and the case for a living Evidence Synthesis Taxonomy","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":24,"is_retracted":false,"has_abstract":true,"ca_institutions":"Public Health Ontario; St. Michael's Hospital; University of Toronto; McMaster University; Queen's University; Impact","funders":"National Health and Medical Research Council; University of Adelaide","keywords":"Context (archaeology); Pandemic; Nursing; Scope (computer science); Public relations; Psychology; Medical education; Business; Medicine; Coronavirus disease 2019 (COVID-19); Political science; Geography; Computer science","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":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.08895037,0.000451666,0.001244975,0.0002521058,0.005522552,0.00004450409,0.001224693,0.000182359,0.001028091],"category_scores_gemma":[0.5281563,0.0002359309,0.0002301779,0.0008683498,0.001749158,0.0008889062,0.0007103605,0.001494992,0.00001814769],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006320087,"about_ca_system_score_gemma":0.003719024,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01488642,"about_ca_topic_score_gemma":0.002445777,"domain_scores_codex":[0.9701009,0.02390796,0.002525732,0.0008852295,0.001479157,0.001101035],"domain_scores_gemma":[0.4435838,0.5511241,0.002320067,0.001669548,0.0009244581,0.0003780671],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.04064043,0.0006093096,0.1640456,0.0418591,0.00106093,0.001641681,0.0379085,0.0002609977,0.001410557,0.01555755,0.1904767,0.5045286],"study_design_scores_gemma":[0.007524949,0.005039957,0.02187499,0.3283021,0.007608755,0.001765634,0.3118049,0.01590508,0.001289133,0.00269304,0.2934649,0.00272655],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.09481427,0.104466,0.005204379,0.7740754,0.002053749,0.01891594,0.00009224699,0.0001951976,0.0001828316],"genre_scores_gemma":[0.9444512,0.009312398,0.001110325,0.0113905,0.0009802452,0.03221746,9.415832e-7,0.00006326784,0.0004736346],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.849637,"threshold_uncertainty_score":0.9998851,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7695973013225446,"score_gpt":0.5618726478439796,"score_spread":0.2077246534785649,"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."}}