I think your setup is great!
Note that I'm really punching above my weight when it comes to statistical analysis. I wouldn't be able to do anything without the help of people like Nablator, Marco and Rene.
That said, there should be some way to calculate whether your detected pairs are statistically relevant. You would probably have to know how likely it is for a VM word type to have more than one token? Example, If, say, 50% of words are hapax legomena, then you can expect 50% of your gallow words to be unique as well. And the other 50% will be expected to have multiple tokens in the MS. If 50% of them have the gallow replaced by the same glyph (or have a second identical token), your result would certainly be relevant.
Note that I'm really punching above my weight when it comes to statistical analysis. I wouldn't be able to do anything without the help of people like Nablator, Marco and Rene.
That said, there should be some way to calculate whether your detected pairs are statistically relevant. You would probably have to know how likely it is for a VM word type to have more than one token? Example, If, say, 50% of words are hapax legomena, then you can expect 50% of your gallow words to be unique as well. And the other 50% will be expected to have multiple tokens in the MS. If 50% of them have the gallow replaced by the same glyph (or have a second identical token), your result would certainly be relevant.
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)