Matching planet labels

[MarcoP]

If this method works on character level, has few or no nulls and is additive in nature, then in general the longer the word or phrase you encode the longer the result. Under these assumptions it makes sense to compare the lengths of various labels even if we have no idea about the specifics of the algorithm used.

my favorite hypothesis on VMS writing is that it is a many-to-many character based scheme, where depending on some unknown factor the same plaintext can be encoded in different ways and at the same time the same transliteration of the encoded text can produce different plaintexts.

I am not sure I understand correctly, but I don't know how these ideas explain the available evidence.

1. Though average word length is comparable for Voynichese and ordinary European languages, the  shape of the distribution is significantly different. In particular, Voynichese has fewer short and long word tokens than ordinary languages (see You are not allowed to view links. Register or Login to view.).

2. A simple example of a many-to-many cipher would be a Caesar cipher where words are written in "Boustrophedon" style, i.e. alternating left-to-right and right-to-left for each line, so that 'wolf' can be ciphered as 'XPMG' or 'GMPX', and both 'XPMG' and 'GMPX' can be decoded as both 'wolf' and 'flow'. This would increase character conditional entropy (while we are looking for methods that decrease entropy) since each bigram would occur in both its original and reversed form with the same frequency (e.g. TH vs HT, CK vs KC). Also, this cipher does not explain why consecutive words tend to be similar in Voynichese (e.g. <f108r.42> okeey.qokeey.qokeedy.qokeey). Maybe this method is not what you are thinking of, but it seems compatible with the description you gave.