{"id":"W6930692377","doi":"10.5281/zenodo.13645316","title":"Improving gyrochronology: An expanded benchmark data set and enhanced age inference framework","year":2024,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"T-cell and B-cell Immunology","field":"Immunology and Microbiology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Benchmark (surveying); Inference; Set (abstract data type); Data set; Cluster (spacecraft); Bayesian inference; Field (mathematics); Bayesian probability","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01376466,0.001408441,0.001457335,0.003176189,0.000973266,0.002756036,0.003575437,0.001895215,0.002064147],"category_scores_gemma":[0.03186386,0.0005428536,0.001643899,0.002230981,0.0008868128,0.003013114,0.002847896,0.002461907,0.001574369],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001539502,"about_ca_system_score_gemma":0.001940429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01663821,"about_ca_topic_score_gemma":0.01503468,"domain_scores_codex":[0.995663,0.001607971,0.0002569106,0.001164763,0.001061742,0.0002456103],"domain_scores_gemma":[0.986134,0.004325529,0.0007279809,0.004524815,0.003824299,0.0004633066],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.000984753,0.0006002322,0.1221411,0.000727454,0.0008946454,0.0003222346,0.000300344,0.5210775,0.01011855,0.02144275,0.03832152,0.283069],"study_design_scores_gemma":[0.0001317445,0.0001768186,0.02452149,0.00008799675,0.0001195765,0.0002734468,0.00007146748,0.9063912,0.01157097,0.02833784,0.02821731,0.0001001479],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2093387,0.002686372,0.730207,0.001361806,0.0004248798,0.0004124865,0.0362993,0.01311222,0.006157204],"genre_scores_gemma":[0.4133198,0.000582898,0.5018038,0.0004920651,0.0003048767,0.0003219425,0.07954165,0.001227137,0.00240583],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01663821,"threshold_uncertainty_score":0.07279527,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04408839984589833,"score_gpt":0.2827046024787479,"score_spread":0.2386162026328496,"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."}}