{"id":"W7111223050","doi":"10.1371/journal.pone.0325723.s002","title":"Methodological quality assessment of 14 included studies using the Newcastle-Ottawa quality assessment scale.","year":2025,"lang":"","type":"article","venue":"Figshare","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Quality assessment; Quality (philosophy); Risk assessment; Quality assurance; Conformity assessment; Quantitative assessment","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05865477,0.0008782174,0.002808022,0.0003396498,0.001663435,0.0006855537,0.003618316,0.0005333307,0.08035503],"category_scores_gemma":[0.0833898,0.0005323274,0.001294927,0.002716861,0.000522376,0.0005372709,0.004161048,0.00140707,0.0001215578],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001246877,"about_ca_system_score_gemma":0.002142347,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001897847,"about_ca_topic_score_gemma":0.0002600878,"domain_scores_codex":[0.9575323,0.02506821,0.006565836,0.002262959,0.007429114,0.001141539],"domain_scores_gemma":[0.9353095,0.04865303,0.004376615,0.004455375,0.006852486,0.0003530393],"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":[0.0005823129,0.007829572,0.1407565,0.01027059,0.00484163,0.0000328227,0.006597658,0.01716606,0.01373407,0.013415,0.6562522,0.1285215],"study_design_scores_gemma":[0.002709627,0.0006641317,0.8067921,0.009980191,0.0005339014,0.000004010527,0.02892827,0.01155123,0.003980526,0.04124898,0.09207927,0.001527692],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"empirical","genre_scores_codex":[0.3241753,0.03014054,0.04593802,0.05463883,0.01449496,0.02173614,0.4195766,0.0005123807,0.08878722],"genre_scores_gemma":[0.9618341,0.0001851285,0.03129048,0.002221474,0.0003191784,0.0005073909,0.001963409,0.00003061473,0.00164827],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6660357,"threshold_uncertainty_score":0.9997128,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8977663646866403,"score_gpt":0.6523932223448586,"score_spread":0.2453731423417816,"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."}}