{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002971038,0.0004339362,0.001189218,0.001114055,0.0001070067,0.0005270109,0.002340618,0.0004310814,0.00001417478],"category_scores_gemma":[0.0007174156,0.0004145025,0.0009797905,0.001485051,0.00006842442,0.001055423,0.006437903,0.000877634,0.00001031113],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002338302,"about_ca_system_score_gemma":0.0003761696,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006251517,"about_ca_topic_score_gemma":0.0001749208,"domain_scores_codex":[0.9958583,0.0001835562,0.0007508008,0.001053894,0.001298994,0.0008544466],"domain_scores_gemma":[0.9943176,0.001486426,0.0003477163,0.003038877,0.0005475607,0.0002618392],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000003477023,0.00007497295,0.06041829,0.0003369184,0.001150617,0.000003603342,0.0003098788,0.9235957,0.00004841977,0.008992692,0.00001458301,0.005050817],"study_design_scores_gemma":[0.00008134854,0.000005048897,0.05018562,0.00006457437,0.0004677473,0.000001207568,0.000005376309,0.9360627,0.0001581235,0.01256219,3.490308e-7,0.0004057143],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2458835,0.0002629817,0.7525595,0.00002816259,0.0001726705,0.0002034613,0.00002377801,0.0007120349,0.0001538638],"genre_scores_gemma":[0.620857,0.00001853129,0.3789956,0.00002163626,0.00002208239,0.00002468525,0.00002298558,0.00002344296,0.00001404216],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.3749735,"threshold_uncertainty_score":0.9998307,"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."}}