{"id":"W4405425641","doi":"10.48550/arxiv.2408.11307","title":"Improved thresholds for e-values","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Statistics; Mathematics; Psychology","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001022313,0.000377767,0.0005449106,0.000392777,0.0002157087,0.0003666221,0.001783916,0.000352556,0.00009538385],"category_scores_gemma":[0.002400782,0.0003500298,0.0003721564,0.0007233549,0.00024288,0.0002232608,0.002546778,0.0007864419,0.000298943],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002324108,"about_ca_system_score_gemma":0.000239812,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002694227,"about_ca_topic_score_gemma":0.00002280666,"domain_scores_codex":[0.9967561,0.00008012632,0.0004810791,0.001915338,0.0002942017,0.0004731214],"domain_scores_gemma":[0.9957828,0.00188995,0.0002978418,0.001182622,0.0006083322,0.000238459],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005288763,0.0001637247,0.002300371,0.0006890098,0.0003288885,0.0004790004,0.0006127703,0.25957,0.0004974431,0.6977029,0.01014426,0.02698279],"study_design_scores_gemma":[0.0001931705,0.0000432463,0.00006478943,0.00006070845,0.00008066792,0.000001014996,0.0002357982,0.2815134,0.0001779574,0.71579,0.001540864,0.0002984684],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.07376781,0.0003443559,0.9158674,0.0001569281,0.003262967,0.0006480704,0.0003570004,0.0002953281,0.005300082],"genre_scores_gemma":[0.9763115,0.00003966004,0.005360195,0.00003951795,0.0004004125,0.000005513371,0.00001316233,0.00004776022,0.01778229],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9105073,"threshold_uncertainty_score":0.9998952,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2671037112618906,"score_gpt":0.3336684743511958,"score_spread":0.0665647630893052,"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."}}