{"id":"W4415443488","doi":"10.31222/osf.io/r9yh2_v1","title":"NOS-TLPlot: A Specialized Python Tool for Visualizing Newcastle–Ottawa Scale Risk-of-Bias Assessments in Systematic Reviews and Meta-Analysis.","year":2025,"lang":"","type":"preprint","venue":"","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Python (programming language); Visualization; Graphical user interface; Data visualization; User interface; Web application; Interface (matter); Flexibility (engineering)","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","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow"],"category_scores_codex":[0.07864458,0.001314924,0.01412335,0.002922273,0.0006798595,0.001012221,0.002746916,0.001294325,0.001702165],"category_scores_gemma":[0.01739386,0.001053026,0.006836793,0.006130464,0.0007547687,0.0005215112,0.002326107,0.001497916,0.00002802447],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008945769,"about_ca_system_score_gemma":0.001334764,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.02483201,"about_ca_topic_score_gemma":0.03574697,"domain_scores_codex":[0.9645945,0.01999368,0.008107641,0.00267635,0.003143206,0.001484613],"domain_scores_gemma":[0.9772954,0.01278214,0.005653759,0.002477423,0.00140536,0.0003858935],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.0005959234,0.00350379,0.01854336,0.4628266,0.4266833,0.00004157676,0.02836056,0.0006243691,0.0001936634,0.04111638,0.01028608,0.00722441],"study_design_scores_gemma":[0.00210166,0.0004475779,0.001028886,0.008148475,0.9180172,0.000002783268,0.01174922,0.01979222,0.0006147787,0.01751332,0.01754343,0.003040463],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01544825,0.05288799,0.5781181,0.004816587,0.002656967,0.2188051,0.002706436,0.000853618,0.123707],"genre_scores_gemma":[0.306444,0.07131114,0.4474263,0.000829235,0.0008330243,0.06235781,0.0002352199,0.0003950146,0.1101682],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4913339,"threshold_uncertainty_score":0.9999602,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5360303526392249,"score_gpt":0.5775934138020985,"score_spread":0.04156306116287356,"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."}}