On the effectiveness of run-time checks

Run-time checks are often assumed to be a cost-effective way of improving the dependability of software components, by checking required properties of their outputs and flagging an output as incorrect if it fails the check. However, evaluating how effective they are going to be in a future application is difficult, since the effectiveness of a check depends on the unknown faults of the program to which it is applied. A programming contest, providing thousands of programs written to the same specifications, gives us the opportunity to systematically test runtime checks to observe statistics of their effects on actual programs. In these examples, run-time checks turn out to be most effective for unreliable programs. For more reliable programs, the benefit is relatively low as compared to the gain that can be achieved by other (more expensive) measures, most notably multiple-version diversity.