Thank u this is such a great question! In the paper, what we tried to do as a control is the amplitude predictive modeling (similar to IC), so it is technically a multi variate model of amplitude and did not work well. Although we are open that maybe certain multivariate decoding could work well!