{"id":"W3137174954","doi":"10.1038/s41597-022-01143-6","title":"A large-scale study on research code quality and execution","year":2022,"lang":"en","type":"preprint","venue":"Scientific Data","topic":"Scientific Computing and Data Management","field":"Decision Sciences","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Engineering and Physical Sciences Research Council; Alan Turing Institute; Alfred P. Sloan Foundation","keywords":"Computer science; Code reuse; Reuse; Replication (statistics); Coding (social sciences); Code (set theory); Transparency (behavior); Database; Set (abstract data type); Programming language; Software; Computer security","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06990214,0.0008477963,0.0007159939,0.009103291,0.002082751,0.00372628,0.002248487,0.001248934,0.001096009],"category_scores_gemma":[0.3982817,0.0009763747,0.001249484,0.0153818,0.003995713,0.005115981,0.002998148,0.002344195,0.001029853],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003238899,"about_ca_system_score_gemma":0.00657688,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01218605,"about_ca_topic_score_gemma":0.009795095,"domain_scores_codex":[0.9193344,0.03269953,0.008852768,0.01262036,0.02462724,0.001865784],"domain_scores_gemma":[0.3598469,0.3753698,0.08714671,0.0977001,0.07511025,0.004826216],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001098262,0.000487452,0.7800267,0.002220053,0.00115543,0.0006682705,0.0114534,0.01038941,0.01006427,0.007132849,0.01618883,0.1591151],"study_design_scores_gemma":[0.0001829937,0.0008642187,0.8768828,0.0009295586,0.0005272542,0.001211587,0.005136128,0.02658579,0.01876436,0.006757709,0.06180021,0.0003574535],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.938462,0.004957446,0.03383323,0.003652644,0.0002404683,0.0005629534,0.009512744,0.00283219,0.005946247],"genre_scores_gemma":[0.9344575,0.001630846,0.04012917,0.0008775594,0.0001987681,0.0005552792,0.01813219,0.00210147,0.001917346],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9300979,"threshold_uncertainty_score":0.3696823,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7493102534435432,"score_gpt":0.600662037413241,"score_spread":0.1486482160303022,"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."}}