{"id":"W4416284809","doi":"10.1177/20592043251384138","title":"Evaluating Musical Predictions with Multiple Versions of a Work","year":2025,"lang":"en","type":"article","venue":"Music & Science","topic":"Music and Audio Processing","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Variation (astronomy); Benchmarking; Feature (linguistics); Mode (computer interface); Feature extraction; Musical","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":[0.0007080571,0.00009423055,0.0001279527,0.0002252566,0.0006364381,0.0001421671,0.001085871,0.0000200238,0.00002092371],"category_scores_gemma":[0.0003458515,0.00007282421,0.00003330068,0.004118361,0.000648899,0.0006143084,0.0004443766,0.0001244934,0.000007155429],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005361156,"about_ca_system_score_gemma":0.0008243938,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002747259,"about_ca_topic_score_gemma":0.00001096945,"domain_scores_codex":[0.9984109,0.00002743385,0.0001886618,0.0004746049,0.0005925457,0.000305894],"domain_scores_gemma":[0.9988891,0.0001707164,0.00009451894,0.0005282082,0.0002326828,0.0000847414],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00008368909,0.0007774192,0.04512955,0.0001816227,0.00005629009,0.00002332651,0.01676631,0.01951081,0.1071777,0.07408897,0.01317876,0.7230256],"study_design_scores_gemma":[0.001661895,0.000310961,0.1991047,0.00120704,0.00005473156,0.00001687206,0.000552061,0.7755842,0.01387845,0.002704975,0.004425195,0.0004988884],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3742682,0.00003238402,0.6166295,0.001006887,0.0004248864,0.0001274466,7.953337e-7,0.00009806688,0.007411759],"genre_scores_gemma":[0.8976063,5.733514e-7,0.1015613,0.000576833,0.0000256318,0.00001149304,2.199895e-7,0.000002264055,0.000215417],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7560734,"threshold_uncertainty_score":0.4895031,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04534455672637839,"score_gpt":0.3094654762339179,"score_spread":0.2641209195075395,"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."}}