I think their point is that Andrew Wiles' flawed proof was shown to a community and improved without any automated testing because something can be verified bef

I think their point is that Andrew Wiles' flawed proof was shown to a community and improved without any automated testing because something can be verified before it's integrated and that verification never changes. You don't need testing for verified code interacting with verified code in an environment verified to run verified code, but testing takes half the time for "good enough" reliability so we don't spend the time.