{"id":"W4378214151","doi":"10.1029/2022wr033918","title":"In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance","year":2023,"lang":"en","type":"article","venue":"Water Resources Research","topic":"Hydrology and Watershed Management Studies","field":"Environmental Science","cited_by":57,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Horizon 2020 Framework Programme; Österreichische Forschungsförderungsgesellschaft; Bundesministerium für Bildung, Wissenschaft und Forschung; Austrian Science Fund; European Commission; Nvidia","keywords":"Benchmarking; Computer science; ENCODE; Variety (cybernetics); Suite; Set (abstract data type); Data science; Metric (unit); Machine learning; Artificial intelligence; Geography","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":[0.05605761,0.002411104,0.001638464,0.004547571,0.001108113,0.006643297,0.001895265,0.003272102,0.002854479],"category_scores_gemma":[0.3223341,0.0005024092,0.001016854,0.005084375,0.003182562,0.011755,0.003919257,0.002828164,0.001284566],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002335587,"about_ca_system_score_gemma":0.002041199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003152635,"about_ca_topic_score_gemma":0.002849254,"domain_scores_codex":[0.919576,0.04810984,0.008964169,0.00817361,0.01368926,0.001487134],"domain_scores_gemma":[0.7125325,0.1763583,0.03070614,0.04194729,0.03492447,0.003531283],"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":[0.001771267,0.0009400543,0.2175972,0.003115487,0.001780072,0.0004586058,0.01190921,0.05671025,0.01717434,0.1045689,0.039735,0.5442395],"study_design_scores_gemma":[0.0002321514,0.002103823,0.1391391,0.001978594,0.000436525,0.0009752392,0.006244833,0.4260499,0.01849109,0.3323352,0.07117869,0.0008348555],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2638462,0.003325739,0.6814882,0.005877512,0.000783882,0.001170101,0.003325069,0.001907177,0.03827624],"genre_scores_gemma":[0.8840308,0.0002974833,0.1110539,0.0008221606,0.0001384008,0.0007946797,0.001375812,0.0003326419,0.001154089],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.05605761,"threshold_uncertainty_score":0.2964646,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1030342086215556,"score_gpt":0.3333135966286256,"score_spread":0.23027938800707,"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."}}