One thousand reionization histories sampled from their prior distribution given the different models we consider. In the PCA case, the prior over PCA modes is taken to be uniform within the physicality region; in the two knot cases, the prior is uniform on the position and/or amplitude of each knot (thus none of these priors correspond to a flat τ priors). The left panel shows the extent to which the PCA procedure allows unphysical models, even when using the optimal physicality priors. The middle panel shows that the HS17 prior creates odd features and tighter peaked priors in the regions between the knot locations (which are indicated with arrows). The right panel is the prior for the FlexKnot model and represents our best solution for creating a reasonable and “homogeneous” prior.