Both manual and automated tests tend to be limited by the imagination of the person(s) designing the tests. It is extremely difficult to write tests for things that we cannot foresee, and as a result, we're likely to suffer crashes due to unexpected combinations of events when our systems go live. In this talk, I will briefly talk about how we have used randomised testing to find real bugs, by poking the system in ways that no sane tester would. Beyond that, I will talk about what happens when you instruct the tool to be evil in a controlled random way, stimulating the system using input that is almost, but not quite, legal. At the end of this talk, the audience should have a new perspective on the potential of robustness testing, and some insight into how it can change the way we think about software design.