{"id":"W4411689481","doi":"10.3847/1538-4357/adda3a","title":"Blinded Mock Data Challenge for Gravitational-wave Cosmology. I. Assessing the Robustness of Methods Using Binary Black Hole Mass Spectrum","year":2025,"lang":"en","type":"article","venue":"The Astrophysical Journal","topic":"Pulsars and Gravitational Waves Research","field":"Physics and Astronomy","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Guelph; Perimeter Institute","funders":"Agence Nationale de la Recherche","keywords":"Physics; Cosmology; Gravitational wave; Binary number; Astrophysics; Robustness (evolution); Binary black hole; Gravitation; Astronomy","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":[0.02846469,0.001094391,0.0008233696,0.001312999,0.001498155,0.002295129,0.002074976,0.002789679,0.001645611],"category_scores_gemma":[0.1030769,0.0004938226,0.001339616,0.0006757601,0.002566845,0.002489498,0.003361642,0.002181581,0.0006049284],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001314922,"about_ca_system_score_gemma":0.001341782,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004428027,"about_ca_topic_score_gemma":0.003082738,"domain_scores_codex":[0.9827858,0.009650654,0.0007373831,0.003514595,0.002829986,0.0004816309],"domain_scores_gemma":[0.9016765,0.05674678,0.008891859,0.02444744,0.006005841,0.002231621],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01065824,0.001670929,0.3743196,0.001777649,0.004331318,0.001098026,0.001902582,0.3575088,0.03335165,0.06649707,0.02777866,0.1191055],"study_design_scores_gemma":[0.000664904,0.001699863,0.1152883,0.0004392621,0.0003705005,0.001061025,0.0009067379,0.756358,0.04206592,0.05948583,0.0212802,0.0003795407],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8091253,0.001630226,0.1735215,0.001455919,0.000670506,0.0005780495,0.005211406,0.002140554,0.005666525],"genre_scores_gemma":[0.9131186,0.0001145753,0.0787776,0.0006136468,0.0001233264,0.0002678981,0.005648552,0.0003231031,0.001012801],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02846469,"threshold_uncertainty_score":0.1505375,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1206561136389411,"score_gpt":0.4578499823589734,"score_spread":0.3371938687200323,"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."}}