Ars has a good write-up of a recent report on problems with the 2006 elections:
The report puts to rest the commonly held belief that screen calibration problems could account for all of the reported instances of vote flipping. Some of the offending machines were not touchscreen models–voters used a selection wheel to make their choices. In other cases, the voters would make a touchscreen selection for one slate of candidates, only to have the summary screen (and in some cases the paper tape) report that half or more of the selections had been flipped.
Notably, there were reports of vote-flipping in the hotly contested FL-13 race in Sarasota County, FL. Most recently in the legal contest over that disputed race, a federal judge has declined the Jennings camp’s demands to see the source code to the voting machines used. (In effect, the judge has declared that America’s citizens are not allowed to see how the votes were counted in this very close race, because to reveal that information would violate the voting machine company’s “trade secrets.”)
Other problems described in the report included difficulties with printing the voter verified paper trails (VVPATs) that are required by law in a few states, and that may soon be required by federal law in all states. In some cases, the VVPATs didn’t match voter choices, but the most common problem was that they were simply unavailable, typically due to printing problems.
This is why it’s critical that the paper be the official record for the election result. If I were going to steal a DRE election, the first thing I would do is cause a “malfunction” with the machines’ printer–something that’s rather easy to simulate in software. Therefore, the paper trail is absolutely useless if voters are allowed to continue using machines that are unable to produce paper records. Only if pollworkers know that the paper ballot is what will actually get counted will they ensure that every voter’s vote is properly recorded on paper.