{"id":"W4414083395","doi":"10.1111/jep.70272","title":"AppRaise: Software for Quantifying Evidence Uncertainty in Systematic Reviews Using a Posterior Mixture Model","year":2025,"lang":"en","type":"article","venue":"Journal of Evaluation in Clinical Practice","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Complement (music); Sensitivity (control systems); Mixture model; Systematic review; Quality (philosophy); Software; Systematic error","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1395785,0.005270961,0.005255313,0.01782764,0.001042275,0.01000532,0.004996186,0.004358604,0.06415857],"category_scores_gemma":[0.4014932,0.006464757,0.01456907,0.01265887,0.001729077,0.008419384,0.01107352,0.004410438,0.01387656],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004300082,"about_ca_system_score_gemma":0.01529723,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00350956,"about_ca_topic_score_gemma":0.006213334,"domain_scores_codex":[0.924678,0.04961535,0.01512986,0.003661431,0.006530855,0.0003844933],"domain_scores_gemma":[0.5406665,0.4058722,0.02441061,0.01184377,0.01611624,0.001090787],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002863943,0.0001737134,0.005605795,0.1409998,0.01206749,0.0005468763,0.003929537,0.05635039,0.002241574,0.06602137,0.1783585,0.5308411],"study_design_scores_gemma":[0.005558897,0.0007138374,0.005413791,0.0432689,0.01125982,0.001073556,0.000500047,0.2997711,0.004348374,0.2451037,0.3818935,0.001094474],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.001529561,0.004690018,0.8861833,0.003279405,0.0004596625,0.01018541,0.02813939,0.060417,0.005116231],"genre_scores_gemma":[0.01109378,0.002385186,0.9491975,0.0005099855,0.0001570599,0.0275758,0.004428165,0.003659419,0.000993181],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.8604215,"threshold_uncertainty_score":0.7381705,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9331737426296115,"score_gpt":0.7227301034102289,"score_spread":0.2104436392193826,"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."}}