The evolution of P/F vs T/K raises important questions

[Jorge_Stolfi]

Here too, I do not see why [f76r] should be included in Bio.

I think layfield and davis convincingly show that it is connected, simply by being a part of the bifolia.

It is probably connected, and was probably written right before the other three pages of that bifolio.  But its different structure suggests that it may well have a different topic, formula, style, etc. -- which could result in noticeably different statistics. Like the preface to a chemistry book may not have a single word ending in -gen, -on, -um, -ide, -ate, -ol, -ic, etc. 

The negative intercept does not indicate a “puff-free” beginning of the line.

Not with certainty, true.  But a fit Y=AX+B implies that, for sufficiently long head lines, the ratio of puffs to glyphs should approach A.    With a zero intercept, that fit would predict the same ratio even for very short lines.  But with a negative intercept predicts that short lines (shorter than the line's root -B/A) would have a much lower puff/glyph ratio than A.  

Hence my guess that the line-initial puffs were not being counted and/or that there was a "low-puff" zone at the start of the head lines. 

However your first set of plots (thanks for them!) show that the situation is more complicated.  As expected from visual inspection of the text, there is an extraordinary number of head lines that begin with a puff.  But then, in all major sections except Bio there are no puffs among the first M glyphs, where M varies from 5 to 9 depending on the section -- except that the very first glyph may be a puff, but only in a fraction of the cases.

In Stars, for instance, 37% of the head lines have practically no puffs in the first 9 glyphs, while 63% start with a puff but have practically no puff among the next 8 glyphs.

In Herbal-A there seems to be two separate "puff-free" zones, at glyph indices 1-5 and 11-13.  But the second one may be a sampling noise artifact.

This pattern seems too consistent to be the result of the Scribe randomly deciding to "puff up" some gallows on the head lines.   Will those gaps remain if all four gallows ptkf are plotted together in the same way?

Herbal B’s slope is steeper: 1.14 additional p/f occurrences/10 additional glyphs, versus 0.80 for Herbal A

So that is another A/B difference?  Or just Herbal-B?

removing every first line with unreadable words leaves the negative Herbal intercepts almost unchanged

Well, so that part of my guess was wrong...

An executive summary of your recent plots could be:  

• In all sections, a large fraction of the parags starts with a puff: Pharma ~30%, Herbal-B ~50%, Herbal-A and Stars ~60%, Bio ~70%.
  

Excluding those initials:

• There are no puffs in the first M glyphs of the head lines, where M can be 5 to 10 or more depending on the section.
  
• In Stars, puffs are almost entirely confined to the head lines, where they are ~4% of the glyphs.
  
• In Herbal-A, ~6% of the head line glyphs are puffs, and also ~2% of the body words have puffs.
  
• In Herbal-B head lines, between 4% to 7% of the head line glyphs are puffs, depending on the line length; and also ~2% of the body words have puffs.
  
• In Bio, puffs occur at the same rate in head and body, in about 2-3% of the words.
  
• In Pharma, the head line has about 2 puffs, and the body about one more, independently of the parag's width or length.
  

 
Makes sense?

All the best, --stolfi