{"id":"W2192978482","doi":"10.1136/ebmed-2015-110342","title":"Improving reporting and utility of evaluations of complex interventions","year":2015,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Psychological intervention; Computer science; Psychology; Psychiatry","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":"reporting","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7578909,0.005038891,0.009841763,0.02357375,0.003126807,0.02138597,0.0106005,0.007664463,0.01647322],"category_scores_gemma":[0.9214269,0.00291692,0.009717034,0.01884457,0.009188483,0.02418896,0.01785258,0.00914659,0.003672683],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01813356,"about_ca_system_score_gemma":0.02352851,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007410207,"about_ca_topic_score_gemma":0.006681892,"domain_scores_codex":[0.0818383,0.7062206,0.11961,0.01701086,0.07356816,0.001752079],"domain_scores_gemma":[0.03031524,0.7339755,0.07304443,0.0912085,0.06981925,0.001637127],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.007275926,0.001746124,0.09657317,0.05429226,0.01296122,0.0003301526,0.01029405,0.009321925,0.0009038256,0.06460989,0.07266612,0.6690254],"study_design_scores_gemma":[0.007687572,0.009070758,0.1704037,0.1720614,0.007934696,0.001236656,0.009117822,0.03655001,0.009164741,0.2958047,0.2791024,0.001865601],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"editorial","genre_scores_codex":[0.06652972,0.0886027,0.5064128,0.05046403,0.01423705,0.1112917,0.04597807,0.005013025,0.111471],"genre_scores_gemma":[0.4366729,0.01378692,0.3665646,0.01245436,0.003622006,0.1464059,0.01489051,0.001871222,0.003731548],"genre_candidate":"editorial","genre_consensus":null,"teacher_disagreement_score":0.2421091,"threshold_uncertainty_score":0.2985637,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8886211946752411,"score_gpt":0.7497801860907314,"score_spread":0.1388410085845098,"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."}}