Relation between word types and vocabulary density.

[npcompl33t]

Very interesting @KoenG. I have actually been digging into MATTR related stats for some tests I am doing with Timm's self citation model, and I agree something very odd is happening, that I specifically was hoping to discuss with you. The overall self-citation tests are not finished yet, so I don't want to focus too much on that now, but here is what the MATTR looks like when plotted across my proposed MS ordering.

   

It matches up with what you are describing: the start of the MS and the end of the MS are the lowest, but when arranged like this, you can see that MATTR is roughly stable,. but starts increasing about half way into 1b / Herbal A, until it hits a new sustained level the 1.5a section. 

1a, 1b, and 1.5a are all Herbal A pages. 1.5a in particular comes from a few different quires, that are shuffled throughout herbal in the current MS ordering.  (These are the scattered Herbal A dots at the top side of your chart. )

The MS stays roughly constant at this higher level for over half its total length, aside from a slight dip at the beginning of 3a (Herbal B), and pharma, both of which show up on your plots. 

This higher rate continues until near the end of the stars 3c section, where the decline that eventually will become bio low starts. You can see this trend continues through the 4c stars section, and drops continuously as bio progresses.

So yes, I 100% agree: the chart you show above with the MATTR arch, and its resemblance to the chronology,  means the MATTR shifts must be related to the overall gradient / evolution of voynich. Whatever changes are occurring are extreme enough to significantly alter the MATTR comparisons against references. 

Timm's self citation model normally falls below the MS in terms of MATTR, It has trouble maintaining enough lexical diversity within a single page to hit the MS targets, with no additional text reserve (like a written external sheet). 

However the MATTR decline in stars 4c and bio 4a stretch actually closely matches the expected MATTR from self citation, limited to the current page. some sort of serious degradation is occurring.

Section or topic vocabulary don't seem like suitable explanations, because the drops and rises clearly happen within sections, not in between them. 

I think this is very difficult to explain with any theory that assumes Voynich text at all parts of the MS are equivalent, or were created in the same way. I'm still struggling to interpret this myself, so would definitely love to hear other thoughts!