I've tested that all three types of PTC work with the BFK code on this branch.
As for making it more consistent: it might be worth adding this kind of feature to the PTC dataset itself; when it unpacks the amplifier entries, it might be best to remove the padded entries. I think this would make len(finalMeans) = len(finalVars) = len(covariances) for any amplifier. This would require repadding if it's rewritten, but I'm not sure that's something that happens.
I've added a ticket (
DM-31938) if you think that's worth doing.