{"id":"W4408473307","doi":"10.22541/essoar.174196646.65056548/v1","title":"Systematic Benchmarking of Climate Models: Methodologies, Applications, and New Directions","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Climate Change Policy and Economics","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Environment and Climate Change Canada","funders":"","keywords":"Benchmarking; Computer science; Climate model; Environmental science; Management science; Climate change; Engineering; Business; Geology; Oceanography","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":[0.1029559,0.002216282,0.003278463,0.009473186,0.001439579,0.01068303,0.005383991,0.002894145,0.002560364],"category_scores_gemma":[0.171145,0.000988659,0.00177243,0.01734748,0.004856899,0.01246076,0.006061103,0.004547813,0.001001032],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005052382,"about_ca_system_score_gemma":0.01121955,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007392692,"about_ca_topic_score_gemma":0.006763159,"domain_scores_codex":[0.9134746,0.05323534,0.006882023,0.00445978,0.02078423,0.001164045],"domain_scores_gemma":[0.8666711,0.08088116,0.009099818,0.01357938,0.02846841,0.001300238],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00008901497,0.0001551676,0.005860278,0.01218529,0.0005360491,0.00009268017,0.001272579,0.03074491,0.0009558044,0.2254142,0.02578413,0.6969099],"study_design_scores_gemma":[0.00008969365,0.0004321171,0.009684684,0.03863202,0.0006321536,0.0002615029,0.002109798,0.06034981,0.005185608,0.4501963,0.4319877,0.000438509],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.01035043,0.4817947,0.4498048,0.02724059,0.002614311,0.0006978786,0.001351108,0.001195862,0.02495045],"genre_scores_gemma":[0.1393758,0.4705043,0.3758319,0.004272482,0.002152259,0.001694626,0.002789706,0.001335914,0.002043104],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1029559,"threshold_uncertainty_score":0.5444895,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3005333969278088,"score_gpt":0.3358679345200598,"score_spread":0.03533453759225097,"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."}}