> This seems to assume that the tests will be higher quality than the problematic code in the first place.

If the problem lies with probematic code then tests are not the problem. At most they're just yet another way that problematic code affects the problem.

Let's put it this way: would the problems go away if the tests were ripped out?