It works really, really well, in part because it's also capable of tracking where an error is turned into another. And the error return trace ends where the stack trace begins (when an error is turned into a panic) so you don't even have to know what an error trace is, because it just makes sense intuitively.