Good question, @profricecake. What I did to avoid that sort of problem is only compare std’s from the same feature groups, just in different eras. (Except at the end, where I take the inner products on the Training era groups)
The reason for doing it that way was because I noticed that the std in two particular feature groups (data columns 13 to 136j varied much more than that of the rest, particularly from era to era. There’s a bit more detail in this post.
There are probably better ways to go about this, but this was easy to set up.