{"id":"W7115678137","doi":"10.48448/ev62-sh58","title":"Factors Associated with the Reproducibility of Health Sciences Research: A Systematic Review and Evidence and Gap Map","year":2025,"lang":"","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"","keywords":"Reproducibility; Operationalization; Relevance (law); Reliability (semiconductor); MEDLINE; Scientific evidence; Quality (philosophy); Empirical research","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","metaepi_narrow","sts"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.2948176,0.00106687,0.004125169,0.002247639,0.004515858,0.0009539968,0.004431156,0.0002533993,0.0001848502],"category_scores_gemma":[0.1481731,0.0005615944,0.0001278307,0.01804868,0.05909704,0.001147378,0.002369641,0.001591532,0.00003781913],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001486005,"about_ca_system_score_gemma":0.01922899,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001755004,"about_ca_topic_score_gemma":0.00238887,"domain_scores_codex":[0.9674778,0.01221906,0.003046109,0.007508995,0.007411222,0.002336817],"domain_scores_gemma":[0.9656267,0.01280038,0.00676469,0.008898078,0.00513574,0.000774369],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00002298024,0.0006173635,0.009328457,0.9771193,0.0003793799,0.000003668094,0.002846004,0.00001449269,0.0001968177,0.00452415,0.004568289,0.0003791133],"study_design_scores_gemma":[0.0003348239,0.001780297,0.008709855,0.9794773,0.001287097,0.00001266919,0.005149391,0.001070638,0.00007532509,0.001201065,0.00007113934,0.0008304224],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.009967551,0.932539,0.0001412579,0.02897067,0.000290143,0.0241726,0.0003106543,0.0001571643,0.003450949],"genre_scores_gemma":[0.9472719,0.0338136,0.0005293479,0.0009824483,0.00004991259,0.0002548971,0.00001100813,0.0002423892,0.01684449],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9373044,"threshold_uncertainty_score":0.9996836,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4040927731384851,"score_gpt":0.4607792874953416,"score_spread":0.05668651435685657,"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."}}