{"id":"W4412779116","doi":"10.1111/jmg.70004","title":"Calibration, Validation and Evaluation of Machine Learning Thermobarometers in Metamorphic Petrology: An Application to Biotite and Outlook for Future Strategy","year":2025,"lang":"en","type":"article","venue":"Journal of Metamorphic Geology","topic":"Geological and Geochemical Analysis","field":"Earth and Planetary Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"European Research Council","keywords":"Biotite; Geology; Metamorphic rock; Geochemistry; Calibration; Petrology; Paleontology; Statistics","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":[0.002212003,0.0001036881,0.0003687912,0.0003905075,0.00005320572,0.00001544166,0.0001120962,0.0001138185,0.0003055438],"category_scores_gemma":[0.0002746674,0.00007343881,0.0000560719,0.0003555048,0.00008112175,0.0001580276,0.00001290007,0.0001734523,7.929243e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00000429162,"about_ca_system_score_gemma":0.00004550147,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004401252,"about_ca_topic_score_gemma":0.001745182,"domain_scores_codex":[0.9985781,0.0004338454,0.00047477,0.0001962887,0.0001784912,0.0001385117],"domain_scores_gemma":[0.9990713,0.0002090926,0.0003011414,0.0000805346,0.0002615299,0.00007638401],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002258086,0.00005384663,0.8618234,0.00003334864,0.00013108,0.000001058674,0.000127579,0.09014621,0.00699969,0.000495258,0.000009605363,0.03995306],"study_design_scores_gemma":[0.0008010833,0.0006490329,0.7603175,0.000006856192,0.000303607,0.00002029582,0.0001927628,0.2106944,0.000684849,0.026088,0.000159639,0.00008198272],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9910645,0.002126807,0.003251361,0.003211467,0.00006778427,0.0002099149,0.000007102258,0.000003628357,0.00005736265],"genre_scores_gemma":[0.9980209,0.0001271426,0.001550483,0.0001635703,0.00005111023,0.000003727613,0.00005967014,0.000001378117,0.00002198302],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1205482,"threshold_uncertainty_score":0.334549,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01834397933654942,"score_gpt":0.25653122207409,"score_spread":0.2381872427375405,"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."}}