Bifolia created using previous as reference

[npcompl33t]

I completed testing of Layfield and Davis' proposed orderings. I compared their two orderings against the original Golf 2.1, and against the version identified above, using only these new page statistics. Not using any word or glyph statistics, so a completely independent test. 

The numbers below are mean mismatch, lower means it is a better fit.

• Layfield and Davis' first proposal: 77–82 |78–81 |75–84 | 76–83 | 79–80 : 0.472 
  
• Layfield and Davis' 2nd proposal: 76–83 | 77–82 | 79–80 | 75–84 | 78–81 :0.374 
  
• Default Golf 2.1:  79–80 | 76–83 | 78–81 | 75–84 | 77–82: 0.466 
  
• Current candidate : 79–80 | 77–82 | 76–83 | 78–81 | 75–84 0.338
  

Layfield and Davis' second proposal substantially beats their 1st proposal and the golf ordering. The main issue is the 75|84 bifolio. There isn't a clear mapping from it to the other bifolio in the set. This bifolio has arguably the most complex typography in the entire MS, so it makes sense that the previous pages may not have been used to help with its layout. 

Below is L&D's 2nd proposal.

   


Skipping the 75|84 bifolio is a much better fit, as seen below. It could be possible that in this case, due to the nature of 75|84 s odd layout, 76–83 was used as the reference for 78–81?