(28-10-2026, 02:11 PM)npcompl33t wrote: You are not allowed to view links. Register or Login to view.

(28-10-2026, 08:40 AM)quimqu wrote: You are not allowed to view links. Register or Login to view.@npcompl33t, I agree that the hands sit at different points of your gradient. I measured it: the hand and section of each page explain 94% of where the page falls on the gradient. (My skribe text it is 97%). But that is exactly the problem. If the hands explain almost all of the gradient, the gradient cannot tell hands apart from time. That the hands follow one another in time is what needs to be shown; it can't be the starting point.

So I tried a test that can tell them apart. If the gradient is time, it should also run inside a single hand, and follow the order of that hand's folios. My text has no order inside a hand by construction, so it gives the baseline.

- Q13 (all hand 2): the main direction in which its pages differ has nothing to do with your gradient (r 0.04).
- Herbal, hand 1: a weak drift with the folio number, but mostly because of its herbal pages in the late quires. In quires 1–7 alone it is weak (p 0.10).
- Q20 (hand 3, stars): here there is something. Its pages make their own small horseshoe, on the same axis as your gradient, and my text doesn't do that. It doesn't follow the folio order, though (p 0.29).

So I would say: the big arc is the hands and sections, and inside the hands there is little, except in Q20, which looks worth a closer look.

I also read reproduce.py on your page. As far as I can see, it averages fits that were computed beforehand, and it only keeps the fits where 1a comes before 1b (and, in the other paths, the order of 3a/3b and 4a/4c is chosen). So the direction and part of the order are inputs, not results. Is there a step that goes from the text alone to the order of Q13 and Q20? With that I could run it on my text and see if it also finds an order there.

I strongly recommend you try Herbal A, because there is significant variation within Hand 1 in that section, and is where the effect would be the most noticeable. You should also note that Timm's theory is that You are not allowed to view links. Register or Login to view., at different points in time, which would perfectly explain your results. 

FOr the current ordering (not sure if you are looking at an older version) -- i used ElPiGraph to fit about 1500 curves in the space to see which orderings are subject to change and which are more secured. That is discussed in You are not allowed to view links. Register or Login to view..

I independently reproduced Layfield and Davis' LSA work, that should be reproducible. As for why our methods are different, theirs uses LSA, which operates on full tokens, while mine uses glyph markov representations.