I don’t think that’s correct, but even if it is it omits the obvious possibility that the first mutation was AT to GC, making the back mutation more likely than the first one.
Fixations aren’t relevant.
Are you familiar with the notion that intron length can function as a hardwired delay register — a non-coding sequence whose physical length determines a precise time interval, like a countdown timer encoded in the genome itself? Different genes have different intron lengths, producing different delays, which allows a single developmental signal to trigger sequential waves of gene expression — genes responding at different times to the same initial trigger, purely as a function of their intronic length. Where we see that absence of conservation is not absence of function.