But in nature most mutations are neutral, period. It is likely to be more don't cares although actual improvements are unlikely. Good outcomes from random chances are rare but that doesn't mean that bad outcomes must be the dominant component.

Just look at Atomic gardening [https://en.wikipedia.org/wiki/Atomic_gardening] - a practice which should be /way/ worse if mutations really did produce bad effects most of the time. Instead the effects were either dud seeds or a few useful byproducts as opposed to 'this grapefruit is now rich in cyanide'.