A non-trivial component of the replication failure is the extreme incentives to cheat, p-hacking and friends, because if you call your shot, and reality says "you're wrong", the real current system, the one that actually determines who gets promotions and more grant money, says that scientist fails.

Sure, the pretend system says this scientist has done noble and laudable work by showing some hypothesis was wrong, but who cares what the pretend system says? Certainly not anyone who wants promotions (or even just job security) and grant money.