Although some errors may be detected by checking work, those checks are themselves vulnerable to the same kinds of adverse influences that caused the original errors. Indeed where capable processes result in few errors to find the probability of detection tends to be especially low. Second and subsequent checks yield an even smaller proportion of detectable errors and may even have counter-productive effects on vigilance. Relying on checking is a flawed strategy for improving reliability of human output because it does little to help address root causes. Even where checks detect errors, these are the consequences of failed processes, not causes. Further investigation will still be needed to identify all contributory causes and this work could have been carried out proactively, without incurring failure costs.