A One-Page Ledger Method for Generating Voynich-Like Text

[Dunsel]

If you compute the correlation between S2 Herbal B and S2 Quire 13, I expect it will be lower than your S2 vs S3 correlation of 0.948 — despite being the same "scribe." That would mean the within-scribe variation exceeds the between-scribe variation, which is hard to reconcile with a multiple-scribe model but follows naturally from a continuous evolutionary gradient.

I ran the comparison you suggested, and yes, your prediction is technically correct — but the separation is not large.

Using Zandbergen/Landini:

comparison	Pearson	cosine
S2 Herbal B vs S2 Quire 13	0.918	0.928
S2 Herbal B vs S3 Quire 20	0.942	0.949
S2 Quire 13 vs S3 Quire 20	0.926	0.934

Using Takahashi, where my other computations are from:

comparison	Pearson	cosine
S2 Herbal B vs S2 Quire 13	0.887	0.898
S2 Herbal B vs S3 Quire 20	0.945	0.951
S2 Quire 13 vs S3 Quire 20	0.925	0.931

So, yes: S2 Herbal B is closer to S3 Quire 20 than to S2 Quire 13 in both transcriptions.

But the size of the effect matters. In Zandbergen/Landini, the Pearson gap is only 0.024 and S2 Quire 13 vs S3 Quire 20 is 0.926, only 0.001 higher than the same comparison in Takahashi at 0.925. So this is not a large collapse of scribal structure. It is a small but real sign that section/regime can cut across scribal hand.

That is actually compatible with what I am arguing: scribal hand, section, and production weighting are separable variables. All of my modeling to this point has focused on Scribe 1.  To clarify that method before looking into the <ed> heavy later scribes.  The chart below shows the same point visually. That <ed> does not divide perfectly by Davis scribe, but it does separate Scribe 1 very cleanly from the later <ed>-heavy sections.  And, aside from the Herbal section which has scribe 2 sheets mixed into quires with scribe 1 sheets, you can almost pick out each section by scribe and <ed> density.