{"id":"W4408389320","doi":"10.1029/2024ms004379","title":"Benchmark Framework for Global River Models","year":2025,"lang":"en","type":"article","venue":"Journal of Advances in Modeling Earth Systems","topic":"Hydrology and Watershed Management Studies","field":"Environmental Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"New Energy and Industrial Technology Development Organization; Google","keywords":"Benchmark (surveying); Computer science; Climatology; Environmental science; Geology; Geodesy","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":[0.01262243,0.001602254,0.001388619,0.00362682,0.0009760365,0.004129256,0.0037138,0.001289735,0.008678274],"category_scores_gemma":[0.02829464,0.0005786265,0.001485571,0.003329392,0.0009543639,0.003455513,0.00357121,0.002259832,0.001472441],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002672261,"about_ca_system_score_gemma":0.003594506,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01281277,"about_ca_topic_score_gemma":0.008469055,"domain_scores_codex":[0.994063,0.003168589,0.0004392189,0.0004755311,0.001560218,0.0002935552],"domain_scores_gemma":[0.9900514,0.003616667,0.0006581629,0.001515687,0.003794257,0.0003637996],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00008006152,0.00008176026,0.001735437,0.0001740782,0.00009303405,0.00009147288,0.00006160556,0.7257167,0.0004969704,0.2230475,0.006919037,0.0415024],"study_design_scores_gemma":[0.00001474864,0.00003216408,0.0001481876,0.00004466964,0.00001104186,0.0000123896,0.00002223496,0.9417343,0.000329832,0.05139787,0.006243275,0.000009306945],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.006758009,0.0004381061,0.974691,0.0003318147,0.0001335236,0.000277287,0.0019728,0.002864578,0.01253291],"genre_scores_gemma":[0.2915667,0.0008472967,0.6906225,0.0002578738,0.0002022467,0.001770771,0.009047369,0.001384292,0.004300954],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01281277,"threshold_uncertainty_score":0.06675458,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01378801733937394,"score_gpt":0.2828322716202503,"score_spread":0.2690442542808764,"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."}}