The problem with the opposite is, since everything gets invisibly bubbled up to the top, you are not able to tell what errors do need to be handled. You only find those out from runtime failures and that's no good if you care about reliability.