{"id":"W2003658455","doi":"10.1007/s10652-009-9163-2","title":"A framework for evaluating regional-scale numerical photochemical modeling systems","year":2010,"lang":"en","type":"article","venue":"Environmental Fluid Mechanics","topic":"Atmospheric chemistry and aerosols","field":"Earth and Planetary Sciences","cited_by":229,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"National Institute of Environmental Health Sciences; U.S. Environmental Protection Agency","keywords":"Credibility; Probabilistic logic; Computer science; Scale (ratio); Process (computing); Statistical model; Quality (philosophy); Bayesian probability; Data mining; Air quality index; Operations research; Machine learning; Industrial engineering; Artificial intelligence; Meteorology; Engineering","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002594373,0.0001778672,0.000183331,0.000003448407,0.0002027617,0.00004167486,0.0002230777,0.00020769,0.001544249],"category_scores_gemma":[0.00005619326,0.0001659794,0.0001048492,0.00005110663,0.00002882268,0.00009095891,0.00002259372,0.0003359964,0.00008282261],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001175789,"about_ca_system_score_gemma":0.00002027981,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004044002,"about_ca_topic_score_gemma":0.000002559246,"domain_scores_codex":[0.9986821,0.00001833518,0.000251304,0.0003779069,0.0003320494,0.000338313],"domain_scores_gemma":[0.9993324,0.000161625,0.00006253459,0.0002512886,0.000005377675,0.0001867422],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001619678,0.0001079942,0.001902848,0.0000434632,0.00003717626,0.000005047699,0.0002241339,0.1418525,0.8511881,0.0008003776,0.0002382717,0.003438078],"study_design_scores_gemma":[0.0002136654,0.00007927338,0.00003405277,0.0000187008,0.000019653,0.00003357954,0.0002704098,0.9861373,0.009099353,0.003155714,0.0007206532,0.0002176023],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7677473,0.000184015,0.2310215,0.00005652277,0.0004362483,0.0003439428,0.00007194404,0.00004179993,0.0000966994],"genre_scores_gemma":[0.9417063,0.0000183928,0.0574679,0.0001496877,0.0003347991,0.00002275697,0.0001856001,0.00001049282,0.0001040451],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8442848,"threshold_uncertainty_score":0.9993685,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02571951257004748,"score_gpt":0.2487872620602434,"score_spread":0.2230677494901959,"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."}}