The Cardan grille approach to the Voynich MS taken to the next level

[ReneZ]

My point was that we do not observe independent choices. 

Rare combinations of the parts can correspond to rare words in the plain text.

If the wheels worked independently, every combination would have the same probability.

The paper addresses two possible uses of the wheels:
1. to generate words "on the fly" and write them down as they appear. This is consistent with the original proposal of Gordon Rugg.
2. to build up a vocabulary of words, to be used to translate some plain text document

What you write is true in the first case, but not in the second case.
Effectively, the second case is much closer to what we see in the MS text.

To be more precise, your proposed 'flat' probability distribution would be true if the text were generated by a computer, but we know, of course, that it was generated by a human. So, it would not be entirely flat. This last point is an important aspect of the arguments of both Gordon Rugg and Torsten Timm.

As described in the paper, Tiltman's table generates only 240 different word types, out of ca. 9500 needed.

This just means that your hypothesis is inconsistent with Stolfi's and Tiltmann's observations.

I am not presenting a hypothesis. I am presenting a different way to look at, or analyse the text.

I also don't understand this comment.
Tiltman's observation is fully in line with the method presented. He just said that this covers "many words", not "all words".

Stolfi's observations are extremely important. They were input to Rugg's proposal, but it remains to be seen whether the n-wheels approach could generate his word grammar.
I never said that it could.