Yesterday, 05:28 PM
This started with something small. A while ago I said in another thread that I had a result about the letter t that wasn't published yet. This is it, and it took me further than I expected. It ends up being about the parchment more than about the text. Long post, sorry.
In the herbal pages of hand 1, the outer pages of each folded bifolio (the first recto and the last verso) have more t, compared to k, than the two inner pages. I took the 21 bifolios of hand 1 that have all four pages and compared t/(t+k) outside and inside. The outer pages have more t in 18 of the 21, by about 0.13 on average (p 0.0002 if you shuffle the four pages inside each bifolio).
My first idea was the skin. The outside of a folded bifolio is one side of the skin and the inside is the other, so maybe t is written differently on hair and on flesh. To check that I needed to know which side each page is, and everybody says that in this manuscript you can't tell. Well, I tried...
I tried colour first (hair is usually yellower) and it didn't work. What does seem to work is the texture of the blank parchment. One side of each leaf is clearly grainier than the other, and often has small dark dots that look like follicle holes. In the pictures the grainy side is on the left and the smooth side on the right, always two sides of the same leaf. The bottom row is the same patch with more contrast.
f14 is the biggest difference in the book. In f94 the grainy side is the verso, and you can see the dots. f112 is there because arca_libraria wrote that You are not allowed to view links. Register or Login to view. is a hair side ("you can easily see the follicle dots"), and in my measurement You are not allowed to view links. Register or Login to view. is indeed the grainier side. That is the only outside reference I have, so when I say grainy = hair it rests on that one page.
How I measured it, because I know someone will ask. I used the full-size Beinecke images, around 300 ppi. On each page I cut 10% off every edge, so the border and the neighbouring page are not counted. I converted to Lab colour and only kept pixels close to the normal colour of that page (within 2 MAD of the median in L, a and b), which leaves out the ink, the paint and the paint that shows through from the other side. On what's left I took the lightness minus a blurred copy of it (Gaussian, 3 pixels), which keeps only the very fine variation, less than about half a millimetre. The number for the page is the spread of that (MAD × 1.4826). Grainy pages give around 4 to 5, smooth ones 3 or less. There is no fixed threshold: for every leaf I compare the recto with its own verso and call the higher one grainy. That way differences in lighting between photos don't matter much. For a bifolio I add the differences of its two leaves, and if the total is under 0.25 I call it doubtful and measure again in many small windows over each side. I wrote these settings down before measuring and didn't change them after.
Before believing it I needed a way to check it, and the manuscript gives one. The two leaves of a bifolio are the same piece of skin, so both must say the same thing about which side is outside, and they are photographed separately. They agree in 46 of 50 bifolios. The four that don't have one leaf where the two sides are almost the same, or come from foldout photos that I had to cut in two, which is the roughest part of all this. The second check is the middle of each quire (f4v/f5r for example), which is always the same side of the same skin. The grain gets it right in 12 of 14 quires, and the two misses are again cut foldout photos. Colour gets only 3 of 7, which is why I dropped it.
So, back to t. In hand 1's herbal the side that ends up outside changes from bifolio to bifolio, 8 grainy-out and 13 smooth-out. The outer pages have more t in both groups (7 of 8 and 11 of 13), and there is no relation with the skin side (correlation 0.10, p 0.68). It's not the skin. The scribe wrote more t on the outer pages of the folded sheet, whichever side they were. I can't tell from the text if that is the start and end of a writing session, or writing those pages with the sheet closed.
Once I could read the sides I couldn't resist looking at Gregory's rule, the usual practice of making facing pages always hair against hair and flesh against flesh. In a quire of nested bifolios that means the outer side alternates from one bifolio to the next. I downloaded the pages and foldouts I didn't have and did all of them:
Quire D follows it exactly. C, E and F have two of each and would follow it if two bifolios were swapped. A, B, G and the balneological quire can't follow it whatever the order, and in A all four bifolios have the same side outside. The two pharma quires have both bifolios the same way round, which doesn't work for a nested quire but works fine if the pharma bifolios were loose single bifolios. Stolfi and Lisa have already suggested that, and the two lost pharma bifolios were singletons. In total I get 24 bifolios grainy-out, 27 smooth-out and 1 I can't decide.
Then I looked at the missing leaves. The herbal block (quires A to H) had 33 bifolios before it was mixed up: seven quires of four and quire H with five. If all of them followed Gregory you would need 17 of one side and 16 of the other. What survives is 14 grainy and 16 smooth, and the three missing ones are 59/64, 60/63 and 61/62 in the middle of H, which Lisa thought might have been herbal. The numbers only come out right if those three were all grainy, or two grainy and one smooth.
Just as a puzzle I tried to put the 33 back into eight quires that all follow Gregory, keeping bifolios with similar text together (glyph transitions, nothing else). What comes out is that npcompl33t's groups make complete quires on their own: 1.5a one quire, 1b two quires, and the Herbal B pages another. The three missing bifolios always end up with group 1a, which is the group with too many smooth-out bifolios. One of the possible orders, outside to inside, * for the missing ones:
1/8, 43/46, 42/47, 58/65
2/7, 20/21, 4/5, 28/29, 49/56 (the quire of five, all 1a)
25/32, 59/64*, 35/38, 60/63*
9/16, 19/22, 10/15, 36/37 (1b)
11/14, 12/13, 18/23, 44/45 (1b)
17/24, 3/6, 51/54, 52/53 (1.5a)
33/40, 41/48, 34/39, 50/55 (Herbal B)
26/31, 27/30, 57/66, 61/62*
Please don't take this as the original order. The skin only says the sides must alternate, and in a quire of four with two of each that leaves 8 of the 24 possible orders. There are many other arrangements about as good as this one. All it shows is that an herbal made of Gregory quires with similar text in each is possible with what we have.
Now the weak points, and there are a few. This isn't a standard method and nobody has checked it against the real parchment. Which side is hair depends only on f112r, and what I call grain could be the follicle holes of the hair side or the velvety surface of the flesh side, I don't know which. That doesn't matter for Gregory, where only "same or different" counts, but it does matter for "24 hair-out". It depends on the photos, focus and JPEG compression, which is why I only ever compare the two sides of the same leaf and use the two leaves of a bifolio as the check. At 300 ppi a follicle is one to four pixels. From the 4 bad bifolios out of 50, a single leaf is wrong maybe once in 25. Some leaves don't show any difference (f73, f76, f85, f90, f95), and f73 is a single leaf, so it stays undecided. The cut foldout photos are the least reliable part.
So, the conclusions: the t effect is only in hand 1's herbal, and in the other hands it's weak or not there. The missing bifolios in the puzzle are only placeholders for their skin side. The two-and-two balance inside groups 1b and 1.5a could happen by chance about a third of the time. And npcompl33t's groups and my text similarity come from the same kind of measure, so it's no surprise that they agree.
If someone who can see the manuscript (Lisa? René?) wants to check, these are the leaves where my measurement is clearest. The recto should be the grainier side in f14, f32 and f8, and the verso in f54, f99, f66, f107 and f65. If that holds on the parchment, the rest probably holds too. If not, I've been measuring something else...
In the herbal pages of hand 1, the outer pages of each folded bifolio (the first recto and the last verso) have more t, compared to k, than the two inner pages. I took the 21 bifolios of hand 1 that have all four pages and compared t/(t+k) outside and inside. The outer pages have more t in 18 of the 21, by about 0.13 on average (p 0.0002 if you shuffle the four pages inside each bifolio).
My first idea was the skin. The outside of a folded bifolio is one side of the skin and the inside is the other, so maybe t is written differently on hair and on flesh. To check that I needed to know which side each page is, and everybody says that in this manuscript you can't tell. Well, I tried...
I tried colour first (hair is usually yellower) and it didn't work. What does seem to work is the texture of the blank parchment. One side of each leaf is clearly grainier than the other, and often has small dark dots that look like follicle holes. In the pictures the grainy side is on the left and the smooth side on the right, always two sides of the same leaf. The bottom row is the same patch with more contrast.
f14 is the biggest difference in the book. In f94 the grainy side is the verso, and you can see the dots. f112 is there because arca_libraria wrote that You are not allowed to view links. Register or Login to view. is a hair side ("you can easily see the follicle dots"), and in my measurement You are not allowed to view links. Register or Login to view. is indeed the grainier side. That is the only outside reference I have, so when I say grainy = hair it rests on that one page.
How I measured it, because I know someone will ask. I used the full-size Beinecke images, around 300 ppi. On each page I cut 10% off every edge, so the border and the neighbouring page are not counted. I converted to Lab colour and only kept pixels close to the normal colour of that page (within 2 MAD of the median in L, a and b), which leaves out the ink, the paint and the paint that shows through from the other side. On what's left I took the lightness minus a blurred copy of it (Gaussian, 3 pixels), which keeps only the very fine variation, less than about half a millimetre. The number for the page is the spread of that (MAD × 1.4826). Grainy pages give around 4 to 5, smooth ones 3 or less. There is no fixed threshold: for every leaf I compare the recto with its own verso and call the higher one grainy. That way differences in lighting between photos don't matter much. For a bifolio I add the differences of its two leaves, and if the total is under 0.25 I call it doubtful and measure again in many small windows over each side. I wrote these settings down before measuring and didn't change them after.
Before believing it I needed a way to check it, and the manuscript gives one. The two leaves of a bifolio are the same piece of skin, so both must say the same thing about which side is outside, and they are photographed separately. They agree in 46 of 50 bifolios. The four that don't have one leaf where the two sides are almost the same, or come from foldout photos that I had to cut in two, which is the roughest part of all this. The second check is the middle of each quire (f4v/f5r for example), which is always the same side of the same skin. The grain gets it right in 12 of 14 quires, and the two misses are again cut foldout photos. Colour gets only 3 of 7, which is why I dropped it.
So, back to t. In hand 1's herbal the side that ends up outside changes from bifolio to bifolio, 8 grainy-out and 13 smooth-out. The outer pages have more t in both groups (7 of 8 and 11 of 13), and there is no relation with the skin side (correlation 0.10, p 0.68). It's not the skin. The scribe wrote more t on the outer pages of the folded sheet, whichever side they were. I can't tell from the text if that is the start and end of a writing session, or writing those pages with the sheet closed.
Once I could read the sides I couldn't resist looking at Gregory's rule, the usual practice of making facing pages always hair against hair and flesh against flesh. In a quire of nested bifolios that means the outer side alternates from one bifolio to the next. I downloaded the pages and foldouts I didn't have and did all of them:
Quire D follows it exactly. C, E and F have two of each and would follow it if two bifolios were swapped. A, B, G and the balneological quire can't follow it whatever the order, and in A all four bifolios have the same side outside. The two pharma quires have both bifolios the same way round, which doesn't work for a nested quire but works fine if the pharma bifolios were loose single bifolios. Stolfi and Lisa have already suggested that, and the two lost pharma bifolios were singletons. In total I get 24 bifolios grainy-out, 27 smooth-out and 1 I can't decide.
Then I looked at the missing leaves. The herbal block (quires A to H) had 33 bifolios before it was mixed up: seven quires of four and quire H with five. If all of them followed Gregory you would need 17 of one side and 16 of the other. What survives is 14 grainy and 16 smooth, and the three missing ones are 59/64, 60/63 and 61/62 in the middle of H, which Lisa thought might have been herbal. The numbers only come out right if those three were all grainy, or two grainy and one smooth.
Just as a puzzle I tried to put the 33 back into eight quires that all follow Gregory, keeping bifolios with similar text together (glyph transitions, nothing else). What comes out is that npcompl33t's groups make complete quires on their own: 1.5a one quire, 1b two quires, and the Herbal B pages another. The three missing bifolios always end up with group 1a, which is the group with too many smooth-out bifolios. One of the possible orders, outside to inside, * for the missing ones:
1/8, 43/46, 42/47, 58/65
2/7, 20/21, 4/5, 28/29, 49/56 (the quire of five, all 1a)
25/32, 59/64*, 35/38, 60/63*
9/16, 19/22, 10/15, 36/37 (1b)
11/14, 12/13, 18/23, 44/45 (1b)
17/24, 3/6, 51/54, 52/53 (1.5a)
33/40, 41/48, 34/39, 50/55 (Herbal B)
26/31, 27/30, 57/66, 61/62*
Please don't take this as the original order. The skin only says the sides must alternate, and in a quire of four with two of each that leaves 8 of the 24 possible orders. There are many other arrangements about as good as this one. All it shows is that an herbal made of Gregory quires with similar text in each is possible with what we have.
Now the weak points, and there are a few. This isn't a standard method and nobody has checked it against the real parchment. Which side is hair depends only on f112r, and what I call grain could be the follicle holes of the hair side or the velvety surface of the flesh side, I don't know which. That doesn't matter for Gregory, where only "same or different" counts, but it does matter for "24 hair-out". It depends on the photos, focus and JPEG compression, which is why I only ever compare the two sides of the same leaf and use the two leaves of a bifolio as the check. At 300 ppi a follicle is one to four pixels. From the 4 bad bifolios out of 50, a single leaf is wrong maybe once in 25. Some leaves don't show any difference (f73, f76, f85, f90, f95), and f73 is a single leaf, so it stays undecided. The cut foldout photos are the least reliable part.
So, the conclusions: the t effect is only in hand 1's herbal, and in the other hands it's weak or not there. The missing bifolios in the puzzle are only placeholders for their skin side. The two-and-two balance inside groups 1b and 1.5a could happen by chance about a third of the time. And npcompl33t's groups and my text similarity come from the same kind of measure, so it's no surprise that they agree.
If someone who can see the manuscript (Lisa? René?) wants to check, these are the leaves where my measurement is clearest. The recto should be the grainier side in f14, f32 and f8, and the verso in f54, f99, f66, f107 and f65. If that holds on the parchment, the rest probably holds too. If not, I've been measuring something else...
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)