SALIENCE listener vs EEG: replication and attention test on MAD-EEG (results)

Pre-registered replication (plan: doi:10.5281/zenodo.23964890) of the SALIENCE listener vs EEG result on polyphonic music, plus a test of attention, using MAD-EEG (8 listeners attending to one instrument in duos and trios). The result did not replicate: the listener’s surprise did not improve EEG prediction over the envelope (3 of 8 listeners, p = 0.47) or over envelope and spectral flux (4 of 8, p = 0.68). Surprise in the attended instrument was not tracked more than in ignored instruments (p = 0.96). The analysis pipeline was intact (correct alignment; flux adds beyond the envelope in all 8 listeners), but stimulus-locked prediction was about a third as strong as in the Bach data. Includes code (MIT), features, per-listener scores and a dated deviation log.