Multiple p-mods in SPM

Does anyone know how orthogonalization and mean-centering of multiple p-mods work? In my study, in the final design matrix, p-mods are orthogonal but the mean isn’t zero. However, according to this post SPM mean-centers the p-mods during orthogonalization.

SPM mean-centres and orthogonolises the p-mods, then it performs convolution with the HRF and orthogonises again. After convolution the mean is small but several orders of magnitude larger than before, so there is a small correlation between p-mods.