{"id":"W4252955832","doi":"10.5194/gmdd-5-347-2012","title":"Assessing climate model software quality: a defect density analysis of three models","year":2012,"lang":"en","type":"preprint","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Executable; Context (archaeology); Software; Climate model; Computer science; Trustworthiness; Software quality; Quality (philosophy); Climate change; Software development; Geography; Geology; Programming language; Computer security","routes":{"ca_aff":true,"ca_fund":true,"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":[0.008777562,0.000661007,0.0006054104,0.002576587,0.0005241722,0.001631679,0.001191479,0.0008782675,0.000728225],"category_scores_gemma":[0.03427085,0.000405982,0.001704744,0.001791915,0.001166884,0.001589917,0.001102951,0.001010706,0.00009815545],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002495924,"about_ca_system_score_gemma":0.0007084646,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01548116,"about_ca_topic_score_gemma":0.008821555,"domain_scores_codex":[0.9977738,0.001147945,0.0001439161,0.0002795093,0.0005071517,0.0001475879],"domain_scores_gemma":[0.921496,0.06264917,0.005027983,0.00523723,0.004447084,0.001142454],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.0006671976,0.0006365416,0.4673708,0.00009317903,0.000528842,0.0001873445,0.0006706466,0.5094769,0.001850428,0.004161351,0.0008800733,0.01347668],"study_design_scores_gemma":[0.00002613699,0.0001900374,0.0814929,0.00001251979,0.00009033894,0.00005471851,0.000156652,0.9146847,0.0006525056,0.002430548,0.0001743192,0.00003465552],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9933569,0.00005239326,0.005818747,0.00008161915,0.000003185427,0.00002093628,0.0001295917,0.00008941476,0.0004472009],"genre_scores_gemma":[0.9975619,0.00002206337,0.002130765,0.000006915205,0.00000248481,0.00001462871,0.0001938974,0.00001565775,0.00005173698],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9912224,"threshold_uncertainty_score":0.04642069,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1484321936416739,"score_gpt":0.3733499920176819,"score_spread":0.224917798376008,"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."}}