{"id":"W4362700320","doi":"10.31219/osf.io/d7pbh","title":"The Rational Agent Benchmark for Data Visualization","year":2023,"lang":"en","type":"preprint","venue":"","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Science North","funders":"","keywords":"Visualization; Computer science; Benchmark (surveying); Bounding overwatch; Information visualization; Task (project management); Data visualization; Rational agent; Creative visualization; Rational design; Human–computer interaction; Data mining; Machine learning; Artificial intelligence; 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":["scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0009671369,0.0001419445,0.0001209959,0.0000746832,0.0003327968,0.00120474,0.003646828,0.00009005471,0.00002151197],"category_scores_gemma":[0.0004226679,0.0001011284,0.00005613836,0.0002443702,0.00003096545,0.0002993362,0.005039914,0.00009140698,0.00007685993],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003171358,"about_ca_system_score_gemma":0.0002894617,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001865158,"about_ca_topic_score_gemma":0.00008550179,"domain_scores_codex":[0.9983776,0.00005900196,0.0003584574,0.0006274141,0.0003959423,0.0001815618],"domain_scores_gemma":[0.9971597,0.0003351305,0.0001831801,0.002045362,0.000220926,0.00005576782],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[8.574288e-7,0.00001598617,0.00002199613,0.00002360006,0.00002828676,3.285394e-7,0.00003614078,0.0005810207,0.000001497512,0.621733,0.3742591,0.003298268],"study_design_scores_gemma":[0.00006356623,0.000006293137,0.00007304649,0.0000136971,0.000009079155,2.689762e-7,0.00001058164,0.7407815,0.00001502583,0.02332725,0.2355928,0.0001068405],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.000003165973,0.00004711927,0.9926543,0.003974495,0.001774499,0.0003927338,0.0002923241,0.0002624503,0.0005989152],"genre_scores_gemma":[0.01503743,0.005408312,0.5121246,0.01330855,0.004656603,0.001060982,0.2539741,0.0003023029,0.1941271],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7402005,"threshold_uncertainty_score":0.9998321,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.17094996800839,"score_gpt":0.408702341540989,"score_spread":0.237752373532599,"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."}}