{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005537433,0.0001005135,0.0002579015,0.00005517314,0.00009100533,0.00001651688,0.0002140561,0.00006796331,0.000006865797],"category_scores_gemma":[0.0000558784,0.00008082764,0.00007470162,0.0001691131,0.00007305063,0.000526308,0.00008145045,0.0001341587,0.000003907775],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008017659,"about_ca_system_score_gemma":0.000008568031,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002501999,"about_ca_topic_score_gemma":0.00002997424,"domain_scores_codex":[0.9989769,0.00004427001,0.000442256,0.000147396,0.000178482,0.0002107265],"domain_scores_gemma":[0.9996014,0.00007454187,0.0001633387,0.0001079208,0.00002278688,0.00003006966],"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":[0.00006743302,0.00002524194,0.0132814,0.00004049222,0.00002354764,0.000004763739,0.0001420959,0.9731377,0.000002561422,0.0116832,0.0001497843,0.001441771],"study_design_scores_gemma":[0.0003956268,0.00006214144,0.0002500298,0.0002546125,0.00002202021,0.000004017117,0.0001930806,0.7147693,0.00000306223,0.2787926,0.005169095,0.0000844427],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05412244,0.004203347,0.9339123,0.0002431945,0.00117143,0.000204046,0.000002370267,0.000007581394,0.006133272],"genre_scores_gemma":[0.9717203,0.0008265102,0.02705459,0.0001533573,0.00006696997,0.0000103695,3.639073e-7,0.000003505052,0.0001640521],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9175978,"threshold_uncertainty_score":0.3296055,"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."}}