I find the approach very interesting, but the main problem I see is that it is based on whether the Eva representation of the character is a vowel or a consonant. Given that the Eva representation is essentially arbitrary, this is likely to end up with a result that does not differ much from 'chance'. I believe that this is what @obelus has demonstrated.
It might be worth doing the same thing, but classifying the initial characters by one of the attributes of its shape.
It might be worth doing the same thing, but classifying the initial characters by one of the attributes of its shape.
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)