Why not positional variation?

[Koen G]

RadioFM: 

When considering exceptions like [i] at the end of words, I personally feel like these can safely be ignored. There is almost always something weird going on (like obstruction by a plant drawing) and their occurrence is three orders of magnitude smaller than that of [in]. 

For me, the biggest problem with [n] being the final form of [i] is the existence of [ir]. One option is just that the agreement was "you flourish your final minim". Something like this was done in accounting with Roman numerals, where the final I was written like J. In that case, the scribe might sometimes use the other flourish, without a change in interpretation. 

If it is indeed the case that [ir] is equivalent to [in] since it doesn't matter which flourish they use, then we should expect a difference between scribes. In Herbal A, [in]/[ir] = 6%, while in Herbal B with a different scribe, [in]/[ir] = 22%. 

There is also the phenomenon that some glyphs could easily be merged without much loss of information. Like you say, [y] and [a] is an option, but for example also [y] and [q]. If the rule is "[y] before [o] becomes the ligature [qo]", then they are following this quite well. Voynichese.com gives 5289 for [qo] and only 40 for [yo] (which includes weird examples like the one below).