{"id":"W4405483738","doi":"10.2172/2481300","title":"DECOVALEX-2023: Task A Final Report","year":2024,"lang":"en","type":"report","venue":"","topic":"Distributed and Parallel Computing Systems","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Eidgenössisches Nuklearsicherheitsinspektorat; Korea Atomic Energy Research Institute; Japan Atomic Energy Agency; Nuclear Waste Management Organization; Canadian Nuclear Safety Commission; U.S. Department of Energy","keywords":"Task (project management); Computer science; Environmental science; Engineering; Systems engineering","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001955858,0.0005089699,0.0007213978,0.0003044085,0.0000961155,0.0009957339,0.001934991,0.0004530422,0.0001106912],"category_scores_gemma":[0.0002166099,0.0004262635,0.0004914283,0.00081612,0.00003733701,0.0001596796,0.001002655,0.0007263959,0.002235531],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003114133,"about_ca_system_score_gemma":0.00363449,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006202396,"about_ca_topic_score_gemma":0.00003791038,"domain_scores_codex":[0.9950948,0.00007407377,0.001080642,0.001443163,0.001712277,0.0005950206],"domain_scores_gemma":[0.9967822,0.0001156945,0.0004441875,0.001908044,0.0005438556,0.0002060387],"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":[8.512152e-7,0.00003218954,0.00006088349,0.0004342125,0.0001579001,0.01097641,0.00004049213,0.00009142753,0.000007813677,0.006363745,0.9740589,0.007775174],"study_design_scores_gemma":[0.00006966686,0.00004293961,0.0000804472,0.0005152787,0.00004142076,0.01237018,0.000003046034,0.003038825,0.00001109727,0.002231993,0.9810534,0.0005417133],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.00005686693,0.002568365,0.403242,0.0004883381,0.0116089,0.000221048,0.00004276609,0.001030393,0.5807413],"genre_scores_gemma":[0.05898925,0.0001866364,0.02127816,0.0002472521,0.002947856,0.000081165,0.0006157543,0.00009827384,0.9155557],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.3819638,"threshold_uncertainty_score":0.9998189,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06355249207204917,"score_gpt":0.3233694931286686,"score_spread":0.2598170010566194,"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."}}