{"id":"W3212167289","doi":"10.7554/elife.72185","title":"Consensus-based guidance for conducting and reporting multi-analyst studies","year":2021,"lang":"en","type":"article","venue":"eLife","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":54,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; University of Victoria","funders":"Fonds de recherche du Québec – Nature et technologies; Natural Sciences and Engineering Research Council of Canada; Institut Européen d'Administration des Affaires; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Weizmann Institute of Science; European Research Council; National Science Foundation","keywords":"Robustness (evolution); Lead (geology); MEDLINE; Work (physics)","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"theoretical_or_conceptual","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":[0.1211492,0.0001892807,0.003069406,0.0001137743,0.0002786044,0.0004219754,0.000294731,0.00004275105,0.0003777704],"category_scores_gemma":[0.5054755,0.0000989569,0.00100907,0.0006546908,0.00008103447,0.00005808548,0.00009565728,0.00007356898,0.0001021842],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001770712,"about_ca_system_score_gemma":0.0001211213,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004066258,"about_ca_topic_score_gemma":0.00006889569,"domain_scores_codex":[0.9822338,0.002713619,0.01181966,0.0009301948,0.002091481,0.0002112972],"domain_scores_gemma":[0.9647641,0.01136616,0.01566703,0.002360957,0.005704177,0.0001376138],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001821477,0.0001601567,0.6253378,0.0009407069,0.003417031,0.0002700945,0.003007465,0.001525671,0.01689747,0.001677905,0.3219826,0.02476491],"study_design_scores_gemma":[0.003238632,0.0001229769,0.03092325,0.0008331571,0.00338969,0.0003065012,0.06060098,0.2991026,0.04315358,0.002578693,0.5541148,0.001635103],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.916943,0.03112969,0.04530282,0.004354769,0.0006009133,0.0007486565,0.00002329609,0.0000159719,0.0008808776],"genre_scores_gemma":[0.8280459,0.00002347357,0.1584863,0.00121359,0.0001269935,0.00005474212,0.000004762405,0.00001232915,0.01203192],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5944145,"threshold_uncertainty_score":0.9049619,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9662418058042032,"score_gpt":0.6597044233950229,"score_spread":0.3065373824091803,"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."}}