The first samples are ready.
Firstly, I print a control sample with K=0 and without extra prime amount
It was good as expected
IMG_20200825_122059.jpg
Next, I decreased K factor by half and got almost expected result with overcompensation
IMG_20200825_122108.jpg
Then I printed the K=0.3 sample. 0.3 is the recommended starting point for my Bowden length. It still looks overcompensated on the text. The bottom one was printed with my default settings. For the upper one, I use @Phaedrux advice and raised extruder speed/accel/jerk settings. It's maybe slightly better, but I can't notice a difference.
IMG_20200827_005704.jpg
Then I thought, maybe my old K calibrations are not applicable after microstepping change and I print another calibration pattern.
IMG_20200824_155347.jpg
And it points back to K=1. Why?
Now I'm trying to print a single-walled cube test. Now my cubes look like PA isn't exist at all and the width of the layers is the same on all of the length. Even while acceleration to 80mm/s. And the cura sometimes wants to fill tiny gaps on the corners, but these are other topics for googling.
On the next step, I will try to recalibrate flow, maybe PA also affects that.
PS. All my PA samples printed with 5mm retraction, maybe I should lower it more?