Different parchment color through Herbal section?

[Koen G]

Many years ago, I undertook the cumbersome (and mostly fruitless) task of sampling all painted surfaces in the Herbal section in Photoshop, hoping to find differences between Herbal A and Herbal B. Apart from a shade of red that's famously absent from B, nothing was found. (See here: You are not allowed to view links. Register or Login to view. )

Now I thought, what if I dump the info from my spreadsheet in Claude, and let it find any patterns? Not much either. I also got it to resample some things - there may be something going on with differences between the greens, but that's not relevant now. Somewhere along the way, it spontaneously complained that Herbal B bifolios are lighter than Herbal A bifolios, so the comparison may be unfair.

What?

So it had been sampling parchment for some reason and decided that B parchment was lighter. This would be interesting if true, since A and B folios are mixed, so we could rule out the effect of photography.

I asked it to take new measurements. The new method was to take "400 random patches of clean parchment, about 2 mm square, from anywhere on the page. A patch only counted if it was uniform and at least about 4 mm from any ink or paint. That’s about 48,000 patches in total." The results matched, and became even a bit sharper.

It first tried to sell me on a scatter plot, since I generally prefer those. It looks impressive: B is mostly sorted, apart from Q8's usual shenanigans. The fact that precisely the tormented 55/66 bifolio sticks out like a sore thumb is remarkable.
   

However, you can never trust robots to see the full picture. I asked it to make a bar chart according to the current binding order:
   

This confirmed my suspicion: the first handful of folios are just kinda gross. Look at the difference between f2 - f7. Part of A being darker can be explained by increased exposure, which has little to do with the bifolio's surface.

Quire 4 is the first one where bifolios mix, and this contains some of the darkest A folios. But check the bar chart with individual leafs: bifolium 28-29 seems out of place. It is normal for A, but lives in a quire with A's two biggest outliers. 

In all the other quires though, B is much lighter than A.

Results for Quires 4–7, where A and B bifolios are bound together:

Herbal B is lighter in every quire: by 0.9 in Quire 4, 1.0 in Quire 5, 1.8 in Quire 6 and 1.5 in Quire 7. If A and B labels were shuffled at random within each quire, a difference this large would come up about 0.5% of the time (p ≈ 0.005).
Adding Quires 8 and 17: the difference is +1.0 (p ≈ 0.03). Quire 8 is still the exception, because of f57/f66.
Bootstrap: I resampled both the patches on each page and the bifolios in each group, 1,000 times. Herbal B came out lighter every time, with a 95% range of +0.8 to +1.8.
Each part of the page separately: top half, bottom half, gutter side, outer side, rectos only and versos only. Herbal B is lighter in every quire in every case, by +1.0 to +1.4.
Hair and flesh sides: the difference is smaller on hair sides (+0.7) than on flesh sides (+1.9), but positive in every quire both times.
Yellowness: Herbal B is also less yellow, by about 1.7, consistently in all four quires.

So with a different sampling method and position taken out of the equation, Herbal B still sits on lighter, less yellow skins.

I don't know if this is convincing or not, or if any of this makes sense at all, but it sounds impressive. What's also remarkable is that in the "lost" Herbal folios among the pharma section, the A/B "lightness" split remains.