28-10-2026, 10:45 PM
(28-10-2026, 09:37 PM)Mauro wrote: You are not allowed to view links. Register or Login to view.You guys help me sleep better tonight, I hate contradictions
Mauro, I think the contradiction goes away if we separate "there is a gradient" from "the gradient is time".
My skribe does have a gradient, just not a chronological one. Each sheet picks its own habits at random around the section, so the sheets end up spread along a few directions, and one of them is much stronger than the rest: how much a sheet leans to words with e. When things are spread along one main direction, MDS draws a horseshoe. That is a known effect of the method, not a sign of time. Stolfi’s bookshelf would give a horseshoe too, and the order would be real, but it would be the order of height, not of when the books were bought. So the methods are right to find something. The question is only what it means.
The agreement with Lisa in Q13 and Q20 is the same point. One in a few thousand is the chance of two random orders agreeing, but these two methods aren’t independent. Both look at which pages resemble each other, so if one property dominates, both will follow it. Two rulers measuring the same books will agree on the order.
And with npcompl33t’s new code, Q13 in the real text isn’t secure: about 0.54, with no pair of bifolios firmly placed, while my text with no order comes out firmer (0.65 to 0.76). So I wouldn’t put much weight on the Q13 match. Q20 is different. There the real text is very stable (0.95) and beats all five of my runs, so Q20 deserves a closer look.
On my reruns: they are my own reimplementations, not exact copies, so I wouldn’t build anything on where they differ from the published sequences. That is why I used npcompl33t’s own code this time.
Hope you sleep better now
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)
