{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007753693,0.00006935397,0.0000813168,0.00002195228,0.0001530047,0.000007342891,0.0001915503,0.00003048667,0.001030529],"category_scores_gemma":[0.0000603677,0.00004078749,0.00004203124,0.0002484244,0.0001134372,0.00007592696,0.000008847313,0.0000892137,0.00006932644],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003659658,"about_ca_system_score_gemma":0.00001381143,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004394743,"about_ca_topic_score_gemma":0.005353325,"domain_scores_codex":[0.9991868,0.00003561315,0.0002581296,0.0001349495,0.0002320445,0.0001524562],"domain_scores_gemma":[0.9991856,0.0005195035,0.00005686194,0.0002042949,0.00001076545,0.00002294412],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000399272,0.001094359,0.8479478,0.00001991105,0.00009428043,0.00006716956,0.05594628,0.02726858,0.005758598,0.001545751,0.0345152,0.02534277],"study_design_scores_gemma":[0.0003469345,0.0001066743,0.9774031,0.000003553582,0.00002067417,0.000001406612,0.003462105,0.0006599471,0.0004410363,0.0005309562,0.01695174,0.00007187149],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9940139,0.00001420188,0.0002465427,0.0009508205,0.0005082103,0.0007113655,0.0000215036,0.0000310497,0.003502382],"genre_scores_gemma":[0.9985446,0.00004357657,0.00002241841,0.00005104412,0.00002849982,0.0001320923,0.00005582685,0.000005406795,0.001116503],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1294553,"threshold_uncertainty_score":0.9998826,"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."}}