The below is a speculation, but if the Voynich manuscript is a ciphered codex, it seems to make sense.
The marginalia are mnemonics for the key that would allow a person who knows the principles upon which the cipher is built to read the text.
Let's assume Voynichese is a cipher. Given then the Voynichese is not fully made of characters of any known alphabet, the cipher almost certainly has a substitution step, where after some preprocessing (whatever it be: chunking, permutations, transpositions, numeric coding, etc) resulting Voynichese sequences are substituted with plaintext (let's suppose Latin) characters. This mapping of the preprocessed result to Latin is likely arbitrary, in the sense that you still need to remember the actual mapping tables, even if you knew exactly how the preprocessing step worked. This creates a problem for the author of the encoding system, if the substitution tables are lost or forgotten, recovering them may be a very hard process, and even knowing how the cipher works won't easily yield the plaintext. Basically, we have just had You are not allowed to view links.
Register or
Login to view. (good example, bad puzzle), where the basic principle was identified very quickly, but the actual mapping took years to uncover with complex modern computational techniques.
The solution to this problem is too keep the substitution table in the manuscript as a quick mnemonic that would be self evident for a person who knows how the cipher works, but won't make any sense for anybody else. Maybe there is some interplay between the list of Voynichese characters from 1r or 57v with the Latin phrase from 116v, maybe they are unrelated and the whole mnemonic is contained in 116v, or maybe parts of the mnemonic are hidden on different pages (116v and 17r).
Including Voynichese in the mnemonic may let the reader quickly verify that they understood the mnemonic correctly, in which case Voynichese probably just encodes the plaintext sequence that comes immediately before or after it. The table is structured in a way that makes it similar to some phrase, to hide its purpose.
A quick practical example of how a mnemonic may work, that certainly has nothing to do with the actual Voynichese. Imagine a cipher that is made of glyphs @#%^*+$ (this exact sequence of seven glyphs is written at the beginning of the book), and you know that 2 glyphs should encode a single Latin character.
On the last page of the book there is the following:
MICHITON ODALABAS MULTAS PORTAS ZYWEG FOX VHEK JUSQ MARIA (49 Latin characters with some spaces thrown in to make it look like a phrase, which it is not)
Then there is the following line
#@@##@++#^ NINJA SEARV WHIL IT'S STIL COLDE
Can you use this mnemonic to recreate the cipher table and decode the following:
#%@$ *%*^ +*#@^%*% @#*+ @+@^+^ $%#^#@%^%#@%$*@*^#%+ $+^@ #+ +%*$%@+@
How this works exactly:
Just write the phrase as 7x7 table and use the glyph sequence for rows and columns, then you can map glyphs to the plaintext by selecting the first glyph for the row and the second for the column. #@@##@++#^ decodes to NINJA, which lets you quickly verify that the mapping actually works. SEARV WHIL IT'S STIL COLDE is a meaningless padding to make the whole text look like a recipe in an obscure language.
Code:
@#%^*+$
@ MICHITO
# NODALAB
% ASMULTA
^ SPORTAS
* ZYWEGFO
+ XVHEKJU
$ SQMARIA
This way you only have to remember the basic idea that you map two glyphs to a single character and that by combining the sequence of glyphs @#%^*+$ from the first page with the strange phrase from the end of the book you can always recreate the mapping.
There is also a not very encouraging alternative that the manuscript is meaningless and the inscription on the last page is just made to look like a mnemonic for the key to give prospective buyers some hope of decoding the manuscript, but actually means nothing. In this case including the pieces of the ciphertext is just a way of showing that this inscription was made by the original author and not by a later clueless owner.