This is horshack (dpreview) / snapsy (FM), the guy who developed this AF tune technique. Don't get hung up on Live View - the fact that LV is used to establish critical focus (step 1) is only incidental to the process. You can use whatever focus method you want, including viewfinder AF or MF, LV contrast AF or MF...doesn't matter, as long as you establish and verify critical focus prior to starting the other steps.
I wouldn't call it incidental. As you point out, establishing 'critical focus' is, well...critical. If LV is inconsistent, as the data indicate, there's no simple way to establish that baseline, certainly not with a single shot. Have you tested repeatability - several rounds of a single LV focus, resulting in the same selected AFMA value every time?
Have you, for example, used your method and selected an AFMA value 'on the edge' (one that gives confirmation where the next value out does not) then re-checked that value multiple times and seen confirmation every time, then the next value out and observed a lack of confirmation every time?
If by LV being inconsistent you mean its CDAF then sure it's by no means infallible, although it is rather precise, both in my experience and in Roger's LensRental tests. But again you don't have to rely on its consistency because you can verify critical focus visually at 10x and adjust manually if necessary before proceeding with my method. Or focus entirely manually if you prefer. If by inconsistent you mean that LV @ 10x isn't sufficient for visually confirming critical focus, then that I would disagree with that. But you can always take a photo to confirm focus if you don't feel LV can be trusted.
As for the repeatability of my technique, so far it's been very consistent in my tests, to within one or two AF tune units across iterations, when using the same target, focus distance, and lighting conditions between iterations. If you change any of these factors then you may certainly get different results, but those differences will arise from the sensitivity of the PD phase sensing mechanism itself rather than from this specific technique (ie, you'd get the same AF tune deltas using other MFA techniques in those situations).
As for the accuracy of the technique, I developed and tested it first on a D800 with 5 different lenses, all of which produced AF tune values equal or better than what I achieved via LensAlign. I have also tested it on a 5DM3 with two lenses so far, with the same level of accuracy. On my original dpreview for the Nikon version of the technique (link: http://forums.dpreview.com/forums/post/50774257
), everyone who has tried it so far has achieved similar levels of accuracy. I think a few have compared it to Focal as well.