{"id":"W2142171035","doi":"10.1186/1472-6963-5-25","title":"Systems for grading the quality of evidence and the strength of recommendations II: Pilot study of a new system","year":2005,"lang":"en","type":"article","venue":"BMC Health Services Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":327,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Medicine; Health informatics; Nursing research; Health administration; Grading (engineering); Public health; Nursing; Engineering","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3874785,0.0001453773,0.00265448,0.0003416119,0.0006329766,0.0002232218,0.002351671,0.00002885074,0.0001004634],"category_scores_gemma":[0.009119431,0.00004965682,0.0003075456,0.001977884,0.0001296895,0.0002115244,0.000366993,0.0001776826,0.00001340053],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006372695,"about_ca_system_score_gemma":0.0004526048,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01083413,"about_ca_topic_score_gemma":0.007649185,"domain_scores_codex":[0.9031665,0.0699087,0.01620102,0.0009072213,0.009277187,0.0005394234],"domain_scores_gemma":[0.9300452,0.05672522,0.006716674,0.003307452,0.0030208,0.0001845872],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.002455771,0.004021137,0.3426771,0.09864493,0.001756781,2.614437e-7,0.3079013,0.002333055,0.0001627012,0.1465658,0.02109291,0.07238834],"study_design_scores_gemma":[0.003652171,0.003916459,0.08510054,0.004506617,0.0002252625,0.000004977573,0.6720971,0.2142428,0.00002480536,0.001054666,0.01496497,0.0002096037],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9706154,0.01438975,0.00199278,0.003858687,0.0001241894,0.008741117,0.00004884055,0.000003297097,0.0002259329],"genre_scores_gemma":[0.9974241,0.0002982322,0.0009636635,0.00003671593,0.00008003185,0.0002203407,0.00000127533,0.000006650792,0.0009690226],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3783591,"threshold_uncertainty_score":0.9992272,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.9429551293519843,"score_gpt":0.694466003944852,"score_spread":0.2484891254071323,"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."}}