So here we've identified a very specific problem, and we're making progress in generating the information that will be required to fix it. It's very hard to tie this to older problems that maybe haven't been adequately root-caused.
Let's not assume those 2016 reports refer to the same issue unless there is hard information available that we can use to regression test the current models (such as the slow-motion videos and/or reproducer MP4s from the earlier problems).
Certainly, a refund will be required if a fix is not forthcoming, and it's pretty disappointing that a problem of this magnitude has slipped by a QA department that hasn't got its head up its backside.
But let's try to work through for a fix first rather than assuming the worst.