{"id":"W4308866009","doi":"10.3390/vision6040065","title":"Musical Novices Are Unable to Judge Musical Quality from Brief Video Clips: A Failed Replication of Tsay (2014)","year":2022,"lang":"en","type":"article","venue":"Vision","topic":"Neuroscience and Music Perception","field":"Neuroscience","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of New Brunswick","funders":"Office of Defense Programs; Natural Sciences and Engineering Research Council of Canada; U.S. Department of Defense","keywords":"Musical; Replication (statistics); Quality (philosophy); Psychology; Cognitive psychology; Computer science; Visual arts","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reproducibility","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reproducibility","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008082613,0.0008892869,0.0009334125,0.0008065562,0.000617868,0.001903166,0.001559793,0.0009739241,0.00650754],"category_scores_gemma":[0.05665879,0.000391668,0.0007472311,0.0003495146,0.001791248,0.002528409,0.002822691,0.001850163,0.001903304],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004074917,"about_ca_system_score_gemma":0.0005883242,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003656308,"about_ca_topic_score_gemma":0.00361887,"domain_scores_codex":[0.9936339,0.000956505,0.0009753639,0.00135198,0.002818707,0.0002635971],"domain_scores_gemma":[0.9568649,0.01618452,0.005803147,0.01145372,0.007679007,0.002014648],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.007134498,0.00605275,0.470072,0.003102239,0.001501131,0.004875428,0.05849892,0.002058891,0.1763642,0.002428986,0.01318824,0.2547227],"study_design_scores_gemma":[0.0003651951,0.006186393,0.936601,0.0004928834,0.0003244976,0.003032945,0.008498612,0.004368161,0.02355204,0.004082365,0.01223206,0.0002637989],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9927211,0.000226159,0.002425763,0.0001666487,0.0001049478,0.0002705329,0.0003052994,0.00005614637,0.003723291],"genre_scores_gemma":[0.9932745,0.0001351939,0.002600363,0.0004280104,0.00003424966,0.0004038688,0.0004039663,0.00005929652,0.002660573],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9919174,"threshold_uncertainty_score":0.04274547,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05938127848475169,"score_gpt":0.3531853783222646,"score_spread":0.2938040998375129,"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."}}