One of the things that we learned was that replicating the paper ballot made it a very difficult test. Each ballot was to look different, so that the people who counted wouldn't be able to tell that this had been done by the machine versus something that people had marked themselves.
I think we need to have reasonable expectations of what we're looking for. Is it really reasonable to replicate squiggles in people's writing style? That's one thing we need to be looking for.
Also, if you're working to make it for people with disabilities, then involve people with disabilities right from the very beginning stages and throughout the process, so that it's not that they come in after you've selected a device to test. You start right at the very beginning stage and involve that community so that you have an idea of all the concerns from the get-go, when the test is designed.