{"id":"W4323542543","doi":"10.1136/bmjebm-2022-112100","title":"Proposed triggers for retiring a living systematic review","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Construct (python library); Computer science; Certainty; Actuarial science; Risk analysis (engineering); Business","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":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04519819,0.0003516391,0.002577207,0.0002344243,0.0001367039,0.00002366977,0.0005482389,0.0001609276,0.0003892443],"category_scores_gemma":[0.8755799,0.0002311395,0.0003409877,0.00120145,0.0002724007,0.0001076915,0.00007889716,0.0002872524,0.0001241308],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001427996,"about_ca_system_score_gemma":0.000292315,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000007094849,"about_ca_topic_score_gemma":0.000001583484,"domain_scores_codex":[0.9923989,0.002317822,0.002875955,0.0006157618,0.001194823,0.0005967488],"domain_scores_gemma":[0.676367,0.3209464,0.0009504853,0.00103998,0.0004232047,0.000272897],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000130078,0.00005597879,0.00002571996,0.9364212,0.0001105059,0.00006770914,0.0001032614,0.00000235164,0.0002719453,0.0101996,0.05135242,0.001259262],"study_design_scores_gemma":[0.0009989426,0.0006070661,0.0000210552,0.8616086,0.001750952,0.000009526775,0.0001098622,0.001887531,0.0002192663,0.1324111,0.0000876646,0.000288415],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0007034876,0.03192959,0.811725,0.09688094,0.003684792,0.05197398,0.0000602215,0.002237922,0.0008041331],"genre_scores_gemma":[0.006429647,0.009068269,0.9586734,0.01251451,0.00281206,0.007239824,0.000009766056,0.0002659906,0.002986588],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8303817,"threshold_uncertainty_score":0.9831694,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8168270424800806,"score_gpt":0.6492158980082676,"score_spread":0.167611144471813,"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."}}