{"id":"W2051694665","doi":"10.1190/1.1843294","title":"Reid Mahaffy Test Site, Ontario Canada: an example of benchmarking in airborne geophysics","year":2004,"lang":"en","type":"article","venue":"","topic":"Geophysical and Geoelectrical Methods","field":"Earth and Planetary Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"Condor Petroleum (Canada)","funders":"","keywords":"Benchmarking; Mineral exploration; Test site; Test (biology); Geophysical survey; Remote sensing; Computer science; Geology; Engineering; Mining engineering; Geophysics; Management","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":[0.0001629635,0.0001259468,0.0002340023,0.00004439779,0.00005210104,0.00001397856,0.0001681225,0.00004708,0.00130607],"category_scores_gemma":[0.00003353966,0.0001004308,0.00003496587,0.0004854114,0.00003403449,0.0001495798,0.00001084775,0.000196813,0.00002132223],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003787161,"about_ca_system_score_gemma":0.0006091864,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9996645,"about_ca_topic_score_gemma":0.9998415,"domain_scores_codex":[0.9988337,0.00004214368,0.0002382816,0.0002763231,0.0002570881,0.0003524213],"domain_scores_gemma":[0.9992839,0.0002776087,0.0000564918,0.0001931721,0.00003277431,0.0001561135],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004438076,0.000260947,0.7389643,0.00002586511,0.0000116974,0.00006174274,0.0007010917,0.0134009,0.0004772388,0.001076051,0.0001565003,0.2448194],"study_design_scores_gemma":[0.0002440646,0.000364349,0.98909,0.00000976077,0.000005020425,0.000002052618,0.00002442688,0.0006449192,0.0007527726,0.007715075,0.0009989716,0.0001485725],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916342,0.00002420775,0.0001179656,0.0001095569,0.00009114415,0.0000873563,0.00001440324,0.00001320811,0.007907958],"genre_scores_gemma":[0.9923238,0.000002276203,0.006963211,0.0002546233,0.00007163382,5.380525e-7,0.00006625617,0.000002196575,0.0003155163],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2501258,"threshold_uncertainty_score":0.9996068,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01999440769715646,"score_gpt":0.2051776537762561,"score_spread":0.1851832460790997,"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."}}