{"id":"W4401786701","doi":"10.2172/2431152","title":"DECOVALEX-2023: Task F Specification (Revision 10)","year":2023,"lang":"en","type":"report","venue":"","topic":"Simulation Techniques and Applications","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Nuclear Safety Commission","funders":"","keywords":"Task (project management); Computer science; 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":[0.005996203,0.0003342855,0.000616918,0.0009292992,0.0002557214,0.0004522383,0.001432505,0.0005147987,0.02401135],"category_scores_gemma":[0.005439643,0.0002487514,0.0004353782,0.002556979,0.00007617069,0.0001967785,0.0003322986,0.0003218631,0.04096273],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002692856,"about_ca_system_score_gemma":0.0006180528,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001551978,"about_ca_topic_score_gemma":0.00005312224,"domain_scores_codex":[0.9918767,0.0001181334,0.001857789,0.001280653,0.004558751,0.0003079764],"domain_scores_gemma":[0.9917959,0.001897593,0.0009917377,0.002425024,0.002719147,0.0001705973],"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":[0.000002263768,0.00002132837,0.00009932383,0.000006342178,0.000007865507,0.000001621513,0.000008418971,0.00002879992,0.00002929149,0.00142788,0.8123593,0.1860076],"study_design_scores_gemma":[0.00006195586,0.00002076782,0.003674579,0.00004194401,0.00002096677,0.000007002513,0.00005841061,0.0008008088,0.0001123345,0.01899604,0.9759143,0.0002909081],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0002671046,0.0004644059,0.0689012,0.00196251,0.001378806,0.001254364,0.0002598655,0.001238831,0.9242729],"genre_scores_gemma":[0.01248754,0.002795557,0.004524302,0.0001243403,0.001095,0.0001442331,0.0006301271,0.00009175674,0.9781072],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.1857167,"threshold_uncertainty_score":0.9999965,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5601234120549604,"score_gpt":0.5507735863478135,"score_spread":0.00934982570714693,"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."}}