{"id":"W2124476392","doi":"10.1002/met.52","title":"Forecast verification: current status and future directions","year":2008,"lang":"en","type":"article","venue":"Meteorological Applications","topic":"Meteorological Phenomena and Simulations","field":"Earth and Planetary Sciences","cited_by":301,"is_retracted":false,"has_abstract":true,"ca_institutions":"Environment and Climate Change Canada","funders":"","keywords":"Computer science; Forecast verification; Event (particle physics); Process (computing); Functional verification; Data verification; Verification and validation of computer simulation models; Software verification; Data science; Operations research; Formal verification; Data mining; Software; Forecast skill; Software development; Geography; Meteorology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.000142728,0.000158266,0.000195741,0.00007402864,0.0008531893,0.00003045827,0.0001581263,0.00009941267,0.002945809],"category_scores_gemma":[0.00003406095,0.0001138524,0.00006403891,0.0004746272,0.0003005661,0.0001276764,0.00001841434,0.0002240466,0.0003253349],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000006365004,"about_ca_system_score_gemma":0.0000253391,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000478481,"about_ca_topic_score_gemma":0.00005182541,"domain_scores_codex":[0.9986864,0.00009476082,0.0002737421,0.0004197307,0.0001799605,0.0003454547],"domain_scores_gemma":[0.9990283,0.0002552787,0.00007588612,0.0002779513,0.00006324702,0.0002993962],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00003525688,0.0001503451,0.2984517,0.000009566973,0.00002246989,0.000002082965,0.0001878524,0.0003156266,0.00003356619,0.02202075,0.0003594431,0.6784114],"study_design_scores_gemma":[0.0001032337,0.00005947039,0.5033866,3.657284e-7,0.00001058094,0.000009963881,0.00001848543,0.0007824739,0.000002113964,0.006877921,0.4886545,0.00009430147],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9298806,0.0197849,0.00920349,0.00277525,0.000418046,0.001339934,0.0003317338,0.0004873032,0.03577871],"genre_scores_gemma":[0.9883618,0.006162012,0.004222546,0.000242096,0.0005120541,0.0001004395,0.0002886866,0.000003602793,0.0001067754],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6783171,"threshold_uncertainty_score":0.9979656,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03968620783540099,"score_gpt":0.2469762135315269,"score_spread":0.2072900056961259,"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."}}