Bifolia created using previous as reference

[Mauro]

If I got it right, this is not related to sections and hands. Inside each section the position and slant of the text seems to change gradually, and this change seems correlated with the changes in the word structure already identified.

Mauro, I see that it’s within sections, not across them. My problem is with the time scale. A book like this took weeks or months, with several hands and pauses. A layout bias (margin, slant) that carries over from one bifolium to the next over that span needs something that lasts that long. A scribe’s habit doesn’t, and it certainly doesn’t pass from one hand to another.

I agree, it would mean that the layout bias had to be incorporated in the production process. For instance if they used the previous as a guide for the margins or layout. It would have to be embedded into the production process. And perhaps the reason some don't show clear pairs means that the production process was different for those.

@quimqu's objection makes a lot of sense and identifies a real problem. Your proposal looks reasonable, but not so self-evident: it may be, but it's not so clear how. But you found a trend which deserves an explanation (I will confess I did not thoroughly verify your graphs, but I see no reasons for doubting the trend is real at the moment), so all this is quite interesting, however it'll end!

A question: a scribe's habit

certainly doesn’t pass from one hand to another.

As I said I did not thoroughly check the graphs, do margin distance and/or slant 'reset' at the start of a new hand, or do they continue a previous trend?