The Voynich Ninja

(29-08-2026, 09:26 AM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.And while you have found samples from the early 1400s that match many of the letters, you have not found a single sample whose handwriting matches the whole You are not allowed to view links. Register or Login to view. text -- have you?

It seems to me that this may be due to the fact that the quill was unsuitable. The author apparently tried to copy a common type of handwriting, lacking the skills of skilled writing and the right quill.
For example, the letter d in the word “oladabas” is somewhat similar to the letter d in the Rotunda font.
[attachment=17468]
[attachment=17469]

(28-08-2026, 11:14 AM)eggyk Wrote: You are not allowed to view links. Register or Login to view.

(28-08-2026, 11:06 AM)MarcoP Wrote: You are not allowed to view links. Register or Login to view.This is pure speculation. Since we have hard data, I think sticking to them is the best possible approach. More carbon testing could potentially find that there are pages that are more than 60,000 years old, therefore time-travel or alien technology is involved.

Yes, that point is pure speculation.
But why are you characterising what i've said in that way? I haven't claimed once that the results for the samples are incorrect. How is saying "we might find more samples like one of the samples we used" in any way similar to potential time travel and alien influence?

They are similar in all being pure speculation, so you get my point. It’s probably just me, but I cannot make sense of arguments that compare evidence with speculation.
Of course, there is speculation and speculation, and aliens are in different category, I apologize.

---

Maybe this has been mentioned already, I don’t know much about statistics, it’s possible I misunderstood something or made errors, but this is how I understand it:


The sample from f26 results in a carbon measure (14C fraction) of 0.9380 ± 0.0041 (You are not allowed to view links. Register or Login to view.).

I used a python script to repeatedly extract 4 samples from a gaussian with mean 0.9380 and sigma 0.0041. This simulates four samples taken from a single page, different values are due to instrument error alone. The script computes the mean standard deviation for these groups of 4 samples, and I get 0.0038; I understand that this is the expected standard deviation for 4 samples taken from a single page. The 4 VMS measures are closer to each other, with a standard deviation of 0.0030.
In 66% of cases, four measures taken from the same page would result in a standard deviation >0.0030.

The 4 VMS measures are less spread-out than what is expected for 4 samples from the same page. As Hodgins says, this suggests that “a single object was measured 4 times”.

(29-08-2026, 01:25 PM)MarcoP Wrote: You are not allowed to view links. Register or Login to view.They are similar in all being pure speculation, so you get my point. It’s probably just me, but I cannot make sense of arguments that compare evidence with speculation.
Of course, there is speculation and speculation, and aliens are in different category, I apologize.

I appreciate that. There are different levels of speculation, with some speculation being more reasonable, and other speculation being unreasonable. In my opinion, speculating that if we were to take more samples that we may find more of them like one of the samples already taken is basically the exact same assertion that the conclusion of the report has. What is being asserted here is -as far as I can see- that if we were to take samples from the rest of the folios they would likely all be similar to the 4 already taken. All I said is that many of them could be similar to 1 of those samples (f68). 

Now, maybe there is no strong evidence that the other pages would/should be any different, but is a sample size of 4 pages from only 2 sections truly enough evidence to say that all of the rest of the pages are almost certainly the same? The sample size is too low to be representative for the entirety of the VMS in my view. 

On the other hand, if my assertion is that the manuscript could possibly be a composite split by topical section, I also have to admit that 3 samples from 1 section does constitute a pretty reasonable sample size, so I should, would and do accept the dating range from the averaging of those 3 samples for that section.

(29-08-2026, 12:01 PM)oshfdk Wrote: You are not allowed to view links. Register or Login to view.Given that the folios nowadays belong and for a long time have belonged to the same object and share the script, the image styles, and are of compatible sizes, one cannot ignore the probability of 4 random samples from this object producing a coherent largely overlapping result in the carbon isotope measurement.

But I must insist: even if the whole VMS was scribed in the space of a few years (which I am willing to concede that it is still the most likely scenario), it does not follow at all that the vellum of all its bifolios was also made in a similarly short time span.   There is no common-sense argument or physical evidence indicating that the sheets came all from the same vellum maker and from the same lot, or that they were made shortly before the Author acquired them.  

The C14 results only say that the three Herbal bifolios that were tested were, with high probability, made within a 67 year interval centered around ~1420; more specifically, in the interval 1391-1458.  They don't say that they were made within a couple of years of each other And, again, it is about equally likely that the vellum of bifolio f67-f68 was made around 1335 or around 1410.

Quote:Your logic would totally apply if we were identifying when 4 totally distinct objects were produced.

Actually, we are dealing with 50 different objects, which must have come from at least 15-20 different animals, individually slaughtered, skinned, and processed.  They only became the same object when the the Author used them to write his book (which definitely looks like a compilation of at least four completely unrelated books, except that written in (mostly) the same language and/or code and illustrated by the same artist).

All the best, --stolfi

PS. The best violin bows are made from a wood that violin makers call "pernambuco", but is otherwise known as "brazilwood".  It has a purplish-red core from which one once extracted a prized red cloth dye.  The brazilwood tree was once extremely common in this country, so much so that it is claimed that the country was named after it.  But it was so thoroughly exploited starting in the 1500s that today it is an endangered species, and the export of brazilwood logs or timber was strictly prohibited in 1996.  Therefore, many violin bows manufactured today outside Brazil are supposedly made from stocks of "pernambuco" that were amassed by violin makers more than 30 years ago...

(29-08-2026, 08:59 PM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.The C14 results only say that the three Herbal bifolios that were tested were, with high probability, made within a 67 year interval centered around ~1420; more specifically, in the interval 1391-1458.  They don't say that they were made within a couple of years of each other And, again, it is about equally likely that the vellum of bifolio f67-f68 was made around 1335 or around 1410.

If we found bifolio 67-68 just as a separate artifact with no other folios of the manuscript, yes, absolutely, C14 results would about equally likely date it to ~1335 or ~1410. However, I totally agree with the report that in the context of the results for the other folios it's overwhelmingly likely that this bifolio was produced around ~1420 and certainly no earlier than ~1390. The probability that it was produced circa 1335 is there, but it's certainly not of the same order as the probability that it was produced circa 1420.

Even if we ignore the other folios, there is one more issue, which maybe you can explain, because I have no real education in statistics and I don't understand what's happening here. I've marked on the image below the 2 sigma range of the measurement for f67, with the yellow horizontal band. The probabilities as computed in the report - 42.5% for 1308-1362 and 52.9% for 1385-1436, how are they computed exactly? They add up to 2 sigmas, so I assume they somehow distributed 95.4% between two buckets, but I'm not sure this was done correctly. The way I see it, no more than 45% of the yellow band passes through the ~1335 peak, while 100% of this band passes through ~1410 slope. So, if the question is "given this measurement distribution, what is the probability it corresponds to the 1335 peak or 1410 slope", I would say the numbers are 45 / (100 + 45) ~ 31% for 1335 and 100 / 145 ~ 69% for 1410 (both can be multiplied by 0.954 to make them sum up to 2 sigmas, if needed for some reason). And this is even generous, it's hard to measure this exactly from the image, maybe we are talking even 35/135=26% for 1335 vs 100/135=74% for 1410.

[attachment=17479]

Lisa told us that there is no difference in parchment quality between sections. It's all equally bad. How realistic is it that the author(s) acquired bad parchment, and similarly bad parchment decades later? That looks like a bit of a stretch to me. Not impossible, but unlikely. At least I see no evidence for it.

About the bimodal distribution: The calibration curve is from 2005. I wonder if anything substantial would change when recalibrating the values with a more modern curve. I believe the most recent one is from 2020, maybe there's an even newer one.

If we accept Taccola as a source, this would place the lowest boundary for at least the Balneo and Cosmo sections to the 1420s. More realistically the 1430s. That's within the upper range of the younger 1385-1438 peak, but the older 1308-1362 peak implies that the parchment would have been at least 60 years old when used, even up to 125 years.

(30-08-2026, 02:16 AM)Bernd Wrote: You are not allowed to view links. Register or Login to view.Lisa told us that there is no difference in parchment quality between sections. It's all equally bad. How realistic is it that the author(s) acquired bad parchment, and similarly bad parchment decades later? That looks like a bit of a stretch to me. Not impossible, but unlikely. At least I see no evidence for it.

Why is it less likely that someone would source the same quality parchment years later? Is it somehow more likely that after decades that they would suddenly acquire better quality parchment? For all we know, the VMS author may have even made it themselves, explaining the consistently bad quality. 

Also, two samples were deliberately taken from a currier A page and a currier B page, due to the possibility that a change in textual properties may have indicated a change in time. As far as I know, those pieces of parchment are also similar in size and quality, so why was it thought possible that they could date differently? It was clearly entirely plausible to the team that two similar pieces of parchment could reasonably date differently, otherwise they would not have thought it worthwhile to check. 

(30-08-2026, 02:16 AM)Bernd Wrote: You are not allowed to view links. Register or Login to view.About the bimodal distribution: The calibration curve is from 2005. I wonder if anything substantial would change when recalibrating the values with a more modern curve. I believe the most recent one is from 2020, maybe there's an even newer one.

A long while ago I downloaded a similar software to the one mentioned in the report (or a newer version, maybe), and using the most modern curve available through the software at the time (maybe 2020?) I believe the results looked much the same. But i'm not professionally trained to use the software, and I may have made a mistake somewhere. It may be valuable to check, though!

(29-08-2026, 11:16 PM)oshfdk Wrote: You are not allowed to view links. Register or Login to view.However, I totally agree with the report that in the context of the results for the other folios it's overwhelmingly likely that this bifolio was produced around ~1420 and certainly no earlier than ~1390.

Again, why?  Even if the Author acquired that bifolio in the same buying/begging/stealing/rummaging transaction as the other folios, it does not follow that they were all manufactured at about the same time.   Vellum is a non-perishable commodity, like timber, leather, copper, or vitriol, so a merchant's stock could well have new vellum mixed with decades-old vellum.  Especially considering that the Author was happy to take off-cuts and grossly defective sheets.  Are you perchance assuming that the Author made the vellum himself?

Quote:there is one more issue... how are [the probabilities computed exactly?

Refer to the following sketch:
[attachment=17486]
Based on what I know of statistics (not specifically C14 dating), what they seem to have done was

1. took several measurements of the C14/C12 ratio from each sample.
   
2. computed the average and standard deviation of those values.
   
3. assumed that the measurement errors had a Gaussian distribution with those parameters (red curve).
   
4. cut that distribution into thin slices.
   
5. mapped those slices through the calibration curve (blue line).
   
6. assembled the mapped slices into a distribution for the times (black curve).
   
7. computed ranges R,S that contained 68.4% or 95.2% of the total probability (of the gray area).
   

For a point on the vertical axis like A, that meets the blue curve at only one spot, step 5 is easy. They only had to pick the height of the mapped slice A' so that the its area (black) was the same as the area of the original slice (black).  That would make the probability of the time being in the small range A' the same as the probability of the true C14/C12 ratio being in the small range A.  The height of the black area at A' is the height at A times the slope of the blue curve at that point.

For a point like B, that meets the blue curve at three spots, they had to divide the small black area at B into three parts; that is, assume that the time could be in any of the three small intervals B1', B2', B3'.  If I am not mistaken, the proper (Bayesian) way to split is to make each part inversely proportional to the slope at each of the three points.  Anyway, the total small black area at B1', B2' and B3' should be equal to the small black area at B.

The result of of steps 4-6 is the probability distribution for the date of the object (gray area). The pink area being 1, the total gray area will be 1 too.  If the blue curve were a straight line, the gray distribution would be a Gaussian too.  But in general it is deformed, as in this case.

If the gray curve curve had a single hump, it could be summarized by finding an interval that includes 68.2% or 95.4% of the probability, and is either the smallest possible, or leaves out equal amounts of area on ether side, or is centered on the mean of the distribution.  I don't know which of these is the "standard" criterion.

(Those specific percentages are conventionally used because, in a Gaussian distribution, they are the areas contained within one and two standard deviations away from the mean.  For a non-Gaussian distribution, they are somewhat arbitrary.)

But in this case that single-interval summary would be rather uninformative, even misleading, because the distribution has two well-separated humps with non-trivial areas.  Thus a better summary is to split the gray distribution into two curves and find the 68.2% or 95.4% intervals R,S for each hump.  The areas of the humps for the f68 sample seems to be about 0.445 and 0.555. Then we can say that the date is within the R interval with 0.445 x 0.954 probability and within the S interval with 0.555 x 0.954 probability.

All the best, --stolfi

(30-08-2026, 02:16 AM)Bernd Wrote: You are not allowed to view links. Register or Login to view.Lisa told us that there is no difference in parchment quality between sections. It's all equally bad.

To the extent one can see from the scanned images, that seems to be true.  But it does not mean that all sheet are equally bad.  Wasn't there a testimonial that the folios varied in thickness and finish?

From the images, the Herbal folios seem to be generally uniform, but there are many exceptions:

• f5, f13, f26, f37, f38, and others have a slanted free edge.
  
• f9 and others have a tear that was sewn closed during manufacture.
  
• f10, f13, f14, f129, and many others seems to be 1 cm narrower than most.
  
• f15, f36, and others seem to be 2 cm narrower than most.
  
• f14, f30, and others have big holes or skin edges at the free edge.
  
• f28, f34 and others have holes from the original skin.
  
• f29, f50, f55, and others have conspicuous creases.
  
• f66 has half of a big hole along the bottom edge
  

Similar defects are seen in the Bio and Starred Parags sections.  In the latter,

• f105+f114 and f104+f115 have a slanted bottom edge, 1-2 cm too short at one end.
  
• f112 has a surface finish defect at the top right that makes it unwritable.
  

But at least all the bifolios in those sections have very roughly the same dimensions.  Whereas the fold-out bifolios (that include f67+f68) have "non-standard" width, and often end with a flap with irregular free edge that is only a fraction of the width of a normal folio.  And of course the Rosettes bifolio has a very dfferent size and shape too.

Those defective folios look like off-cuts or rejects, that a merchant could have sold at a discounted price, or that some professional scribe or scriptorium could have donated to the Author. Why should we assume that they all came from the same manufacturer and were all made within a range of a few years?

Quote:How realistic is it that the author(s) acquired bad parchment, and similarly bad parchment decades later?

If his means were scarce when he started creating the VMS, why would his financial condition improve over the course of the next 30 years?

But, again, even if he acquired all that third-grade vellum in the span of a few months, it does not follow that it was all made in the span of a few years.

Quote:If we accept Taccola as a source, this would place the lowest boundary for at least the Balneo and Cosmo sections to the 1420s. More realistically the 1430s. The older 1308-1362 peak implies that the parchment [of that section] would have been at least 60 years old when used, even up to 125 years.

  

No objections to that for the Balneo section.  But is there any plausible Taccola influence on Astro/Cosmo, which contains f68? 

Anyway, given that the vellum of f68 looks like an off-cut, it would not be unlikely that it was left in a "rejects" box in the corner of a scriptorium or someone's attic for a century or more.  Vellum is not perishable and does not change appreciably in that time scale.

All the best, --stolfi

(30-08-2026, 10:30 AM)Jorge_Stolfi Wrote: You are not allowed to view links. Register or Login to view.

Quote:If we accept Taccola as a source, this would place the lowest boundary for at least the Balneo and Cosmo sections to the 1420s. More realistically the 1430s. The older 1308-1362 peak implies that the parchment [of that section] would have been at least 60 years old when used, even up to 125 years.

  

No objections to that for the Balneo section.  But is there any plausible Taccola influence on Astro/Cosmo, which contains f68? 

Not necessarily for the Cosmo diagrams, I think.  But the fact that some of the celestial bodies are personifications (full heads) rather than just having a smiley face is either very early (inherited from Classical sources) OR broadly renaissance (Taccola does this as well). If we move into the Zodiac section, the fashion of the figures is not plausible before some time in the 1390's.