Correlation of PINDA result and reality  

Correlation of PINDA result and reality

So yes, another perfect first layer question, but this one has me stumped.

Background:
- Prusa mk3s
- in general printer prints fine (over 1 yr owned), didn't really worry about leveling until I actually wanted a perfect bottom layer for some parts
- nylock bed level mod
- Octoprint with Prusa Mesh Leveling plugin ( https://github.com/avianna/OctoPrint-PrusaMeshMap , my fork, minor improvements from original)
- all leveling tests done with bed stable at 60c
- 7x7x3 bed level mode with magnetic compensation enabled

With some effort, using a combination of adjusting with the nylock screws, bed level corrections, and even super thin steel inserts, I got my bed to a Total Bed Variance value of 0.064 mm (from the Prusa Mesh Level plugin).

However, when I print a 200mm x 0.2mm square test sheet centered on the bed, the left side the nozzle is clearly far too close, and on the right side of the bed the nozzle is so far that the diagonal infill isn't even touching each other.

At this point I got annoyed and reset the bed level correction settings, removed all the steel inserts, and did bed leveling the old fashioned way:
- print 30mm square at each corner of bed and at centre (no perimeter, just infill)
- measure their thickness (should be 0.2mm) and use that as a guide as to how much to adjust the nylock screws by
- I didn't try to get all of them to 0.2mm thickness, but I did try to get all of them to the same thickness +/- 0.01mm)
- and then repeat this with progressively larger squares (50mm, 75mm) until all the squares are smooth and are equal in thickness

Finally, I printed the 200mm square again, and it looks like this:

It's nearly perfect! Great right? Sure, let's check the values from the PINDA and the Prusa Mesh Level heatmap:

What the heck?!?!?

How does the <0.07mm mesh level result in a terrible actual first layer print, and yet, this awful mesh level result in a nearly perfect first layer print?