{"id":"W7081946901","doi":"10.25675/3.03303","title":"Controlled testing of next generation leak detection and quantification solutions to evaluate performance and develop consensus assessment metrics","year":2025,"lang":"en","type":"other","venue":"Colorado State University","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Leak; Methane; Leak detection; Acceptance testing; Continuous monitoring; Greenhouse gas","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004372734,0.0001197455,0.0002282546,0.0003133771,0.0002312183,0.00004157651,0.0001497078,0.00009325164,0.000006657973],"category_scores_gemma":[0.0003126202,0.0001320498,0.00001553596,0.0007406544,0.00006087853,0.00006656751,0.0002139966,0.00009837835,0.000001349155],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008451439,"about_ca_system_score_gemma":0.0002594318,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001556817,"about_ca_topic_score_gemma":0.0002547286,"domain_scores_codex":[0.9991802,0.0001069056,0.000143239,0.0003178315,0.0001186833,0.0001331319],"domain_scores_gemma":[0.9989663,0.000159951,0.0002360459,0.0001805773,0.0004075591,0.00004952162],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0007654104,0.0005543781,0.01135413,0.003186262,0.001352753,0.00008179283,0.002506798,0.03121294,0.1815387,0.02642042,0.07294741,0.668079],"study_design_scores_gemma":[0.001926467,0.0001655404,0.007207511,0.0001442162,0.0001107974,0.00000705397,0.0001701942,0.9434529,0.001186419,0.0001166914,0.04518072,0.000331485],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05238973,0.0002323102,0.8446057,0.0009249852,0.0003612565,0.001132231,0.00003779919,0.0001966228,0.1001194],"genre_scores_gemma":[0.8656262,0.0002425104,0.04072613,0.00002625889,0.00001817008,0.000003632719,0.00001335356,0.000007849137,0.09333593],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.91224,"threshold_uncertainty_score":0.5384833,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07005774383698825,"score_gpt":0.254007166457778,"score_spread":0.1839494226207897,"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."}}