Thanks for your help. I just went through the process of testing a number of 4.x versions with my system...in sort of a binary search, backing my way from 4.16 which worked to the present versions.
As you recall, the problem started when I went from 3.16 Release to 4.21d Master.
I was able to find a boundary between exhibiting the problem high speed tracking behavior and not having the high speed behavior. Note: this is with 128 microsteps per and 4 microsteps for the goto (not the OFF setting that I found stopped the problem with 4.21d). Specifically:
*Tracking way too fast happens on 4.20i , specifically change described as "TMC driver mode switch design change" with number 4c28f05
*Tracking appears to be correct on 4.20i, specifically the immediately preceding change described as "Slight PEC design change" with number 794aa05
I diff'd the two versions but the changes were too complex for me to easily understand.
I totally understand your comment that what I am reporting is "very very unusual". All I can think is there something specific to my HW that is highlighting something in the SW. At this point, I will stay at the 4.20i pre-driver mode change and play with the system for a while.
One related question: When the STMcube downloader erases the memory, does it erase *all* the flash on the STM32 or do some programmable settings persist? I have been wondering if some of this weirdness is due to some setting that is counted on but is not actively set ???