{"id":"W4385977944","doi":"10.1007/s11242-023-01977-7","title":"The FluidFlower Validation Benchmark Study for the Storage of CO$$_2$$","year":2023,"lang":"en","type":"article","venue":"Transport in Porous Media","topic":"CO2 Sequestration and Geologic Interactions","field":"Environmental Science","cited_by":55,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Deutsche Forschungsgemeinschaft; Energi Simulation; National Science Foundation","keywords":"Benchmark (surveying); Benchmarking; Computer science; Petrophysics; Software deployment; Ranking (information retrieval); Scale (ratio); Process (computing); Data mining; Machine learning; Engineering; Geology","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002692927,0.000821748,0.0006064051,0.0005883527,0.0007242913,0.0008085546,0.001030945,0.0009879534,0.0006447601],"category_scores_gemma":[0.003215341,0.0001829209,0.0004653657,0.0005487232,0.0007743834,0.001052288,0.0006297985,0.0008517595,0.0001339613],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001184678,"about_ca_system_score_gemma":0.001175698,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01251298,"about_ca_topic_score_gemma":0.0103259,"domain_scores_codex":[0.9991756,0.0002757247,0.00005774517,0.0001460062,0.0002157062,0.0001291306],"domain_scores_gemma":[0.9976051,0.0009874123,0.0001709895,0.0003902957,0.0006891237,0.000156989],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001188564,0.002108425,0.02802632,0.000180611,0.0001115942,0.0004521617,0.0002949546,0.9098052,0.02379536,0.003026946,0.00390224,0.02710767],"study_design_scores_gemma":[0.0003049142,0.003186199,0.02270621,0.00003177418,0.00003971193,0.0001111086,0.0003679923,0.873833,0.09425984,0.001187371,0.003866313,0.0001055579],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9936265,0.00005626311,0.003145125,0.0001162222,0.00002548244,0.0001064839,0.0009630202,0.000184502,0.001776394],"genre_scores_gemma":[0.9932788,0.00003346529,0.004520437,0.00002024288,0.000005934722,0.00007378056,0.001588896,0.00002168782,0.0004567014],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01251298,"threshold_uncertainty_score":0.02488029,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02389097692745137,"score_gpt":0.2883201582530722,"score_spread":0.2644291813256208,"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."}}