{"id":"W4317738862","doi":"10.11124/jbies-22-00362","title":"Methodological quality assessment should move beyond design specificity","year":2023,"lang":"en","type":"article","venue":"JBI Evidence Synthesis","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Computer science; Scale (ratio); Quality (philosophy); Redundancy (engineering); Reliability engineering; Data mining; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.5305187,0.0004867867,0.005203585,0.0005648396,0.0003413689,0.001358147,0.003556004,0.0002428406,0.03225284],"category_scores_gemma":[0.5013151,0.0002335748,0.002744431,0.003813428,0.0001776423,0.0006365726,0.0004547723,0.0003466671,0.01765644],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001367766,"about_ca_system_score_gemma":0.0001556752,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004292517,"about_ca_topic_score_gemma":0.00001042963,"domain_scores_codex":[0.8537194,0.118646,0.01170323,0.002545071,0.01263109,0.0007552188],"domain_scores_gemma":[0.6121293,0.3708014,0.00628137,0.008720314,0.001550969,0.0005166005],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00008689714,0.0002498848,0.1012776,0.0002425174,0.0007792271,0.0001279238,0.0008401927,0.002435149,0.01017123,0.02206618,0.5073165,0.3544067],"study_design_scores_gemma":[0.0001031967,0.00009578282,0.7839165,0.0002461471,0.0006981288,0.00001515953,0.002908168,0.02578731,0.003247347,0.1410049,0.04103939,0.0009379116],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0761424,0.0007040773,0.8842441,0.01769693,0.000851209,0.002038317,0.00004089163,0.0001265714,0.01815546],"genre_scores_gemma":[0.7344651,0.0003416136,0.2462139,0.000947976,0.0003000362,0.0004794856,0.000002231785,0.00003185879,0.01721783],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6826389,"threshold_uncertainty_score":0.9996786,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.940347558989425,"score_gpt":0.6430948244968382,"score_spread":0.2972527344925868,"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."}}