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

[npcompl33t]

About the second set of plots (puff counts as function of head line length in glyphs),

There are two different calculations here, I should have distinguished more clearly.
The lines on the first-line plots were constructed from the aggregate rate: rate = total puff count / total glyph count
They are proportional-rate reference lines, not correlation. You are right that a positive line of that kind can appear even when line length and puff count are uncorrelated. The ratios are valid descriptions of observed frequency, but they do not demonstrate that longer lines contain more puffs. 

The paragraph-size analysis was different. It already included an intercept for each Golf group:
puff count = group-specific intercept + B * paragraph word count

Consequently, the reported within-group association was not produced by forcing the regression through the origin. It also was not a claim that every group has a positive association. 1b, for example, is essentially flat. 

I checked the identities of the points you mentioned:

• The three short first lines in 3c are f86v3.15, f86v3.18 and f86v3.29, not the titles on f104v, You are not allowed to view links. Register or Login to view. and f108v. Their transcription-defined blocks contain 3, 5 and 8 lines respectively.
  
• Of the 13 first lines shorter than 20 glyphs in 2a, 12 come from f67r2, and the remaining one starts at f99r.47.
  
• You correctly identified the nine-puff first line in 4c as f111r.36. The nine-puff first line in the Bio group is f76r.1.
  
• The current data do indeed count f84v.1–2 as one paragraph.
  

The three very long 4c blocks you identify contain 261, 364 and 293 words, but only 1, 2 and 4 puffs. Their segmentation certainly deserves examination, but they are not all high-puff outliers that inflate a positive correlation.

the Golf groups are textual clusters, not exact equivalents of the illustrated manuscript sections, and i'm going to actually take back what i said in the previous post, i can see your point on this one because it is combining different paragraph types. Here is MS section, and by layout. 

The charts now show fits with free intercepts and actual correlations.