I tried Koen's vocabulary richness measure (MATTR: how many different words appear in each stretch of 100 words) on my skribe. The skribe is a program that writes a fake Voynich with no meaning (see You are not allowed to view links. Register or Login to view.). It follows the same procedure on every page. Each hand and section has its own glyph habits, learned from the real text, but it copies words at the same rate everywhere and knows nothing about the order of the pages.
It still makes Q13 the poorest quire, exactly like the real manuscript (0.74 in both), and it gets the other sections roughly right too. The reason is not that the scribe copied more in Q13. Even if I switch copying off completely, Q13 stays the poorest. If instead I give the whole manuscript one shared set of glyph habits, Q13 becomes average:
So Q13 is poor because of the way its words are built: the habits of that section push most words into a few forms like shedy, chedy and qokedy. The 20 most common words make up 31% of Q13, against 17% elsewhere, and the skribe gets 30% just by writing glyph by glyph with those habits.
What the skribe doesn't reproduce is that some folios are richer or poorer than others within the same section. That goes by sheet, not by position in the book. In Q20 the two halves of each bifolio match: f103+f116 and f108+f111 are poor, f104+f115, f105+f114 and f106+f113 are rich. Across the whole manuscript, two folios of the same sheet are more alike than chance, while neighbouring folios in the binding are not. It looks like each sheet, written in one sitting, came out more or less repetitive.
This doesn't mean the proposed order is wrong. It means that MATTR and the seriation agreeing is not evidence of a chronology: a text written the same way everywhere, with sessions that are more or less repetitive, gives the same picture.
It still makes Q13 the poorest quire, exactly like the real manuscript (0.74 in both), and it gets the other sections roughly right too. The reason is not that the scribe copied more in Q13. Even if I switch copying off completely, Q13 stays the poorest. If instead I give the whole manuscript one shared set of glyph habits, Q13 becomes average:
| Text | Q13 | Other quires | Q13's place (of 11) |
| Real manuscript | 0.736 | 0.828 | poorest |
| Skribe as it is | 0.741 | 0.817 | poorest |
| Skribe with no copying at all | 0.771 | 0.846 | poorest |
| Skribe with one set of glyph habits for the whole book | 0.831 | 0.833 | 5th |
So Q13 is poor because of the way its words are built: the habits of that section push most words into a few forms like shedy, chedy and qokedy. The 20 most common words make up 31% of Q13, against 17% elsewhere, and the skribe gets 30% just by writing glyph by glyph with those habits.
What the skribe doesn't reproduce is that some folios are richer or poorer than others within the same section. That goes by sheet, not by position in the book. In Q20 the two halves of each bifolio match: f103+f116 and f108+f111 are poor, f104+f115, f105+f114 and f106+f113 are rich. Across the whole manuscript, two folios of the same sheet are more alike than chance, while neighbouring folios in the binding are not. It looks like each sheet, written in one sitting, came out more or less repetitive.
This doesn't mean the proposed order is wrong. It means that MATTR and the seriation agreeing is not evidence of a chronology: a text written the same way everywhere, with sessions that are more or less repetitive, gives the same picture.