The fundamental reason for SF success is it is strong and works out of box for everyone. Nobody really cares about how complicated the code is as long as they are not garbage.
There you have this NN black box that nobody understands how it produced better evals, then to argue that 100 lines of vectorized code which nobody should touch are complications of maintenance. There are plenty of such code everywhere, removing 100 lines here alone does not make it any less complicated, let alone other obvious complications it can cause.
There are also plenty of occurrences historically that optimizations are made in favor of strength at the cost of making someone else's life harder, RefreshTriggers for one, canSaturate for another.
Clear principles are to be made and followed to the text, I can like it or not like it, but now with this double standard at all levels I don't see there is any future down this path, therefore I will no longer contribute to the testing framework.