The Naibbe cipher is interesting, but he says himself it doesn't achieve quite a lot of the statistical properties of Voynich, so as a demonstration of form and method it's useful, but it's far from conclusive. I've played around with a dozen ciphers, including Cardan grilles and Volvelles together to try to find something that works, but have only hit 87% of Bowern-Gaskill statistics. It seems high until you realise that the key ones such as autocorrelation are extremely difficult. Then you need to decide if it's word level you're looking at or character level.
You also have quite different measurements depending which section of the manuscript you're comparing against. Looking at the recipes section, the nearest I've come has been something templated. Both Arabic and Chinese formulary texts match well, but then have the wrong word length. So the tricky thing is which kind of text in which language against which part of the Voynich, with which kind of cipher on top. There are almost too many variables to control for. On top of that you then need to decide whether you are using the script itself or EVA, and then which EVA, and then how you treat tokenisation.
I call Voynich an example of Chop Suey pareidolia.. Everything fits depending on the individual choices you make.
You also have quite different measurements depending which section of the manuscript you're comparing against. Looking at the recipes section, the nearest I've come has been something templated. Both Arabic and Chinese formulary texts match well, but then have the wrong word length. So the tricky thing is which kind of text in which language against which part of the Voynich, with which kind of cipher on top. There are almost too many variables to control for. On top of that you then need to decide whether you are using the script itself or EVA, and then which EVA, and then how you treat tokenisation.
I call Voynich an example of Chop Suey pareidolia.. Everything fits depending on the individual choices you make.