{"id":"W4306844731","doi":"10.31223/x52938","title":"In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance","year":2022,"lang":"en","type":"preprint","venue":"","topic":"Hydrology and Watershed Management Studies","field":"Environmental Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Horizon 2020 Framework Programme; Austrian Science Fund; Nvidia","keywords":"Benchmarking; Computer science; ENCODE; Suite; Set (abstract data type); Variety (cybernetics); Metric (unit); Data science; Preference; Machine learning; Artificial intelligence; Management science; Mathematics; Statistics; Political science; Economics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05172278,0.00242358,0.001629128,0.004166181,0.001110307,0.006836826,0.001909987,0.003387165,0.003085955],"category_scores_gemma":[0.3143665,0.0005113805,0.001000275,0.004719348,0.003354618,0.01168877,0.004041241,0.002950291,0.001419197],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00223232,"about_ca_system_score_gemma":0.002047941,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00288195,"about_ca_topic_score_gemma":0.002459961,"domain_scores_codex":[0.9265425,0.04326617,0.007435462,0.008085446,0.01325267,0.001417862],"domain_scores_gemma":[0.7320001,0.1597506,0.02807273,0.043279,0.03312964,0.003767884],"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":[0.001691489,0.000868869,0.190452,0.003147281,0.001656664,0.0004575885,0.01167168,0.05748558,0.01830249,0.1232369,0.04343575,0.5475937],"study_design_scores_gemma":[0.000211592,0.001756492,0.1170337,0.001686577,0.0003847529,0.00091215,0.00513537,0.3972044,0.01952872,0.3821045,0.07332703,0.0007147514],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2403632,0.003403295,0.7067101,0.005755452,0.0008085233,0.0009591746,0.003197491,0.001928266,0.03687448],"genre_scores_gemma":[0.8755247,0.0003362033,0.1190179,0.0008815211,0.0001648304,0.000725799,0.001580144,0.0004222393,0.001346678],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.05172278,"threshold_uncertainty_score":0.2735395,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04928111287632808,"score_gpt":0.2742894665626755,"score_spread":0.2250083536863474,"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."}}