The VMS as a possible chain encryption ( Mod 23 ).

[oshfdk]

After converting all letters to capital letters in the original text, the entropy value is 3.9962.

Also the increase from 3.9962 to 4.4779 is not minor, it's almost as huge as it can get. Maximum possible character entropy for an alphabet of 23 characters is log2(23) ~ 4.5236. The value of 4.5236 would mean the same probability of any character coming after any other character (edit: it's single character probabilities, the wording suggested conditional probabilities), it cannot get more random than that. 4.4779 is just ~0.05 bits from the absolute limit. To see the difference it would be better to talk about the redundancy (if I remember correctly what it's called), which for a 23 character alphabet is 4.5236 - entropy. The redundancy of plaintext Latin is 4.5236 - 3.9962 = 0.5274 bits per character. After the encryption it's 4.5236 - 4.4779 = 0.0457, a decrease of more than an order of magnitude.

I meant that a word can only be identical if the last letter at the end of the previous word is also identical.

I don't think it's only if the last letter is identical, two different words can also produce the same ciphertext starting from two different ciphertext letters if their letter values are the same up to some constant shift. But this situation probably will be much rarer than just a repeated plaintext word.

This property should therefore theoretically also be found in the VMS.

I'm not sure how to evaluate this encoding in the context of the Voynich Manuscript. For starters, the character frequency statistics in the manuscript are very wrong for this method. This method produces a text where all characters appear with roughly the same probability, there are no very frequent or very rare characters. So, for this encoding to work (edit: for the Voynichese), the basic element should be some combination of glyphs and not a single glyph.