Maybe we should run the code twice on different cpu architecures and only accept the results if they are identical. I’ve heard of high reliability environments doing this, but maybe there are cases for doing it in web/IT as well.

This would help catch a large variety of possible errors, including but not limited to cpu bit flips, cpu bugs, memory errors.