I suggested code testing and code coverage. They did not like this. The answer they wanted was; a statistical analysis of the results to show that the differences were 'good enough' and therefore justified updating the code.