{"id":"W4416225816","doi":"10.21203/rs.3.rs-7763957/v1","title":"Flexible Uncertainty Calibration for Machine-Learned Interatomic Potentials","year":2025,"lang":"","type":"preprint","venue":"Research Square","topic":"Machine Learning in Materials Science","field":"Materials Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Parameterized complexity; Calibration; Uncertainty quantification; Quantile; Baseline (sea); Adaptability; Interatomic potential","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch","metaepi_narrow","sts","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.02750053,0.001304331,0.001929831,0.001961871,0.003021118,0.005630601,0.005552061,0.001334592,0.009660857],"category_scores_gemma":[0.01247511,0.001255827,0.0007756355,0.002003058,0.001599589,0.0008934952,0.007313346,0.003050725,0.0008718148],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001538564,"about_ca_system_score_gemma":0.004780029,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005162279,"about_ca_topic_score_gemma":0.0003457283,"domain_scores_codex":[0.9775064,0.008178273,0.00257206,0.004395671,0.00384904,0.003498548],"domain_scores_gemma":[0.9874653,0.004054063,0.001033037,0.003441806,0.003218197,0.0007876597],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003851285,0.0008342816,0.0006205062,0.01790989,0.000172836,0.00004245391,0.002997573,0.3266307,0.5982128,0.0257139,0.0144531,0.008560731],"study_design_scores_gemma":[0.002098028,0.001164443,0.0003796525,0.006577965,0.0001023202,0.00001636042,0.0007854338,0.7221437,0.2027896,0.03640629,0.0258533,0.001682967],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2644374,0.003180764,0.612084,0.02441189,0.02434658,0.03720011,0.01732961,0.002341284,0.01466838],"genre_scores_gemma":[0.930186,0.0008798019,0.02274934,0.0004272309,0.002049343,0.003784233,0.001726522,0.0002217377,0.03797577],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6657487,"threshold_uncertainty_score":0.9999709,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08326280331758985,"score_gpt":0.4455504691928044,"score_spread":0.3622876658752145,"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."}}