{"id":"W4319596722","doi":"10.1007/s10694-023-01371-1","title":"Roxborough Park Community Wildfire Evacuation Drill: Data Collection and Model Benchmarking","year":2023,"lang":"en","type":"article","venue":"Fire Technology","topic":"Evacuation and Crowd Dynamics","field":"Engineering","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada","funders":"National Research Council Canada; University of Colorado Boulder; National Institute of Standards and Technology; Horizon 2020 Framework Programme; Lunds Universitet; RMIT University; Universidad de Cantabria; European Commission; U.S. Department of Commerce; Fire Protection Research Foundation; Imperial College London","keywords":"Benchmarking; Drill; Data collection; Engineering; Transport engineering; Environmental science; Forensic engineering; Computer science; Business; Statistics","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":[],"consensus_categories":[],"category_scores_codex":[0.0002656312,0.000112593,0.0001268072,0.000203845,0.0002670268,0.0000234877,0.0002910264,0.0002341385,0.000007851047],"category_scores_gemma":[0.00009709171,0.0001287949,0.00001159929,0.0008447871,0.00005942517,0.000178487,0.0002331475,0.0003620288,0.00002391871],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006172554,"about_ca_system_score_gemma":0.00001802274,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003598672,"about_ca_topic_score_gemma":0.0002678316,"domain_scores_codex":[0.9993726,0.00002709764,0.0001656782,0.0001560738,0.00009122981,0.0001873259],"domain_scores_gemma":[0.9992452,0.00005493736,0.00002887768,0.0006236369,0.00002139692,0.0000259962],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00002873908,0.0001051476,0.02281057,0.0002950596,0.0002311969,0.00001511624,0.002764555,0.2840897,0.005535065,0.01106475,0.09788795,0.5751721],"study_design_scores_gemma":[0.0001760298,0.00002144876,0.002870305,0.00001859946,0.00001303723,0.000008192815,0.0001990362,0.9882501,0.00008264719,0.006661714,0.001573531,0.0001253112],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9747138,0.0001767574,0.01940346,0.0009232655,0.0002037518,0.0001821546,0.00003557751,0.003111962,0.001249275],"genre_scores_gemma":[0.9967264,0.0004432264,0.002100742,0.00003350122,0.00002153453,0.00002768662,0.0004952326,0.00002644148,0.0001252677],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7041605,"threshold_uncertainty_score":0.5252103,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03873732058073268,"score_gpt":0.2724278314773267,"score_spread":0.233690510896594,"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."}}