I used the recommended method of taking a dozen readings, discarding the top and bottom values and averaging the remainder. For what it's worth, when I finally got it working, the consistency of adjustment in the 105 Micro was extremely stable, varying by only 1 point over the sequence and with an average value of -12. Though I don't know how well that consistency correlates to accuracy of adjustment in practice.
The 60mm Micro that I tried subsequently was initially far less consistent, bouncing all over the place from +3 to +20. I refocused that one and tried again with better results, the bulk of the run being stable around +12 but with several outliers at +19.
For my art repro work where focus is absolutely critical I use Live View anyway, so any slight inaccuracy in the PD autofocus system isn't an issue; unfortunately the sculpture photography project for which the D850 was purchased won't permit that luxury, so accurate push-button focus is a necessity.
On the back of your comment I'm minded also to try the manual method later when I have more time, if only for confirmation, but hopefully the auto adjustment will be 'close enough' for the purpose.