{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008443012,0.001006475,0.0006275076,0.002573168,0.0005678018,0.002816817,0.001079882,0.001022691,0.00219609],"category_scores_gemma":[0.05456737,0.0003165772,0.0007372568,0.001605939,0.0007541375,0.003308957,0.001865353,0.0008628449,0.001419684],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007394239,"about_ca_system_score_gemma":0.000478484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001976056,"about_ca_topic_score_gemma":0.002301421,"domain_scores_codex":[0.992416,0.001429975,0.00048681,0.00189829,0.003503996,0.0002648982],"domain_scores_gemma":[0.9575829,0.02708465,0.004191903,0.004630704,0.005571316,0.0009384938],"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":[0.003829515,0.0008849649,0.4131123,0.0007067627,0.0008580071,0.0006166628,0.002124402,0.110648,0.02787504,0.002991804,0.006257425,0.4300953],"study_design_scores_gemma":[0.0001100145,0.002148934,0.3486457,0.0001380576,0.000274574,0.0004335057,0.001562642,0.6060623,0.0273692,0.006705004,0.00637347,0.0001766341],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.8763981,0.0004001572,0.1130254,0.0002264518,0.0001573224,0.0002420532,0.00191022,0.001880022,0.005760072],"genre_scores_gemma":[0.9662095,0.00008016913,0.02923844,0.00003550593,0.00004506444,0.0001384974,0.003220213,0.0001493747,0.0008832554],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.008443012,"threshold_uncertainty_score":0.04465145,"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."}}