{"id":"W4408440538","doi":"10.5194/egusphere-egu25-4596","title":"ContrailBench: evaluating the performance of contrail models","year":2025,"lang":"en","type":"preprint","venue":"","topic":"Data Quality and Management","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Environmental science","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":[0.01642013,0.0002118213,0.0005703537,0.0001840155,0.0001534088,0.0003117668,0.003160629,0.0001391179,0.0008234482],"category_scores_gemma":[0.002083486,0.0001147241,0.0002237822,0.0003275766,0.0002127777,0.0002606992,0.003135521,0.0003978456,0.00005733472],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002773188,"about_ca_system_score_gemma":0.0003476887,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003262052,"about_ca_topic_score_gemma":0.0001283572,"domain_scores_codex":[0.9949383,0.0005751363,0.001358216,0.0006600146,0.002219399,0.0002489636],"domain_scores_gemma":[0.993834,0.002683155,0.0006655997,0.002127535,0.0006447067,0.00004496572],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001051175,0.0001055722,0.0001420451,0.0002374813,0.0002070073,7.535398e-7,0.001913591,0.1722259,0.0000463908,0.4393019,0.04493507,0.3407791],"study_design_scores_gemma":[0.0003142451,0.00007917736,0.0004679356,0.0001488541,0.0000732039,2.951633e-7,0.001121974,0.796823,0.0002006103,0.1931821,0.007412392,0.0001761882],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.2205769,0.0008807807,0.1506733,0.008584431,0.002839778,0.002545349,0.0008074922,0.0001108941,0.6129811],"genre_scores_gemma":[0.9530395,0.0001406205,0.002854314,0.001658829,0.00005911552,0.00007882763,0.00002807136,0.000004907962,0.04213584],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7324626,"threshold_uncertainty_score":0.9016181,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5004050430154415,"score_gpt":0.5058007409941256,"score_spread":0.005395697978684111,"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."}}