An Empirical Method for Harmless Error