{"id":"W4388348706","doi":"10.2172/2004895","title":"DECOVALEX-2023 Task F Specification (Rev. 9)","year":2022,"lang":"en","type":"report","venue":"","topic":"Distributed and Parallel Computing Systems","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Nuclear Safety Commission","funders":"National Nuclear Security Administration; Sandia National Laboratories; U.S. Department of Energy","keywords":"Task (project management); Computer science; Programming language; Environmental science; Engineering; Systems engineering","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001630635,0.0003394068,0.0004889511,0.0002402189,0.0002523697,0.0003585651,0.002570226,0.0002077184,0.001148047],"category_scores_gemma":[0.0001030276,0.0003277063,0.0002680689,0.0007941861,0.00002212272,0.0001601854,0.0008531344,0.0005241181,0.0008625594],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005163424,"about_ca_system_score_gemma":0.001297125,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004229883,"about_ca_topic_score_gemma":0.00002029551,"domain_scores_codex":[0.9960715,0.0001740267,0.0007192221,0.0009836943,0.0016234,0.0004281909],"domain_scores_gemma":[0.9971232,0.0001411234,0.0005397998,0.001721188,0.0003425575,0.000132079],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000001108044,0.00005660443,0.0001193546,0.0001048111,0.00006096607,0.00005933009,0.00007183479,0.0005636836,0.000007959141,0.01071332,0.9697248,0.01851618],"study_design_scores_gemma":[0.00008710329,0.00004060168,0.0007451235,0.00004379113,0.00001178049,0.0001916143,0.00001287528,0.003337384,0.000004545052,0.0004304376,0.994692,0.0004026979],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.00002563749,0.001376754,0.40949,0.0002916451,0.00718957,0.0003047472,0.00007148496,0.0005919286,0.5806582],"genre_scores_gemma":[0.08650411,0.00252692,0.02773116,0.0005477503,0.005389588,0.000340458,0.003776063,0.0001688529,0.8730151],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.3817589,"threshold_uncertainty_score":0.9999175,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.062222863568236,"score_gpt":0.2991842548215344,"score_spread":0.2369613912532984,"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."}}