{"id":"W6931843350","doi":"10.5281/zenodo.7640347","title":"Data and Code: In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance","year":2023,"lang":"en","type":"other","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"","field":"","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"ENCODE; Code (set theory); Data collection; Encoding (memory); Source code; Human genome; Key (lock)","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006511132,0.002705948,0.001605479,0.00297123,0.0009499409,0.004037805,0.003225613,0.002550688,0.2013498],"category_scores_gemma":[0.06169642,0.001477789,0.001708357,0.004117548,0.001206602,0.003882971,0.004342838,0.003816934,0.2399039],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001662062,"about_ca_system_score_gemma":0.00353331,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008485015,"about_ca_topic_score_gemma":0.007616621,"domain_scores_codex":[0.9913867,0.001347167,0.001331886,0.001175997,0.004278899,0.0004792247],"domain_scores_gemma":[0.9591418,0.01419847,0.001830024,0.01051682,0.01299353,0.001319409],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001640485,0.00005321543,0.000849666,0.0007414019,0.00002761294,0.00003392022,0.0001015272,0.0007489868,0.000603551,0.002485516,0.9815927,0.01259773],"study_design_scores_gemma":[0.0004803125,0.0001024791,0.005804321,0.0007513817,0.00004070165,0.0002015886,0.0001205654,0.00895041,0.00579609,0.01572589,0.9618535,0.0001729336],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"methods","genre_scores_codex":[0.001447859,0.0001443817,0.02369192,0.001225549,0.0006547825,0.000492532,0.7847623,0.1710883,0.01649237],"genre_scores_gemma":[0.00936503,0.00020103,0.02391645,0.0007622117,0.0001575878,0.001232545,0.8248288,0.1282954,0.01124098],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9934888,"threshold_uncertainty_score":0.6735821,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1490168144278838,"score_gpt":0.3087946782547413,"score_spread":0.1597778638268576,"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."}}