AgentX20 It's the nature of this dilemma that we go looking for any such patterns right!?
Me, I have a theory that there's more configuration held in the video card that can change - outside of the VBIOS itself. This is how good cards can go bad, perhaps as a result of a new video driver changing the underlying card configuration.
If that actually is the case, it's going to be much harder to solve. Video cards are enough of a black box on their own, and Nvidia don't really work with outsiders all that well (but what major tech company does?)
Tell you what. If you have an alleged 9xx card that was "turned bad" from a software update, take an old computer that can run Windows 7, install known good Nvidia drivers from circa 2014 or so and see if it's still "bad". If it is I have some ideas on how a VBIOS dump could potentially isolate any differences.
Another idea I've been curious about for quite some time is if a reliable correlation of VBIOS versions between good cards and bad cards can be found, it may be possible to attempt to convert a bad card into a good card. It's not for the faint of heart and is just an idea, but I have some understanding on how this could be done, how to avoid bricking (well, I would be more concerned about doing the reverse and losing a "good" card), etc.
As @degen said above: two of the reported "good" cards have a VBIOS dating from 2014 and two of the reported "bad" cards have VBIOS's from 2015 and 2016, so maybe there is credence to my theory. Need more data.
This all being said, the 9xx series are rather old, and given the mixed reports on newer cards I'm not holding much hope there at all. The silver lining is the 970 makes an interesting case study because we have an initial version that was "good" and through some mechanism later units were "bad", hence my proposal to investigate this.