The November 2006 race for Florida's 13th Congressional District resulted in a 369 vote margin of victory for the winning candidate with more than 18,000 undervotes recorded on the ES&S iVotronic touch-screen voting machines used in Sarasota County. Since then, the losing candidate and a coalition of local voters have filed suit against the state and local election officials (among other defendants), seeking a judicial order to rerun the election. A key question is whether a system malfunction may have induced the undervote rate. We evaluate the two major efforts previously undertaken by the State: a mock election, conducted by the State, and an analysis of the iVotronic source code, conducted by academic computer scientists under contract to the State. Press reports and summaries of the State's findings have created a public perception that the investigation was thorough and that the voting machines have been exonerated of contributing to the undervote. Based on our evaluation of the investigation, this perception is not justified.
There are many significant gaps in the tests conducted by Florida and its experts. The defined scope of the mock election specifically excluded examination of the vote selection process, which, based on voter complaints, should have been a major focus of the investigation. The tests were conducted in an artificial setting with the iVotronics mounted vertically, unlike their horizontal orientation in real elections. Furthermore, the State's report claims that there were no anomalies observed during the vote, yet video recordings of the test show occasional vote selections not registering on the machines.
The State's inspection of the iVotronic's software was also incomplete. The State's academic team read the source code but performed limited hands-on experimentation with demonstration machines. They made no attempt to examine whether the hardware functioned properly, nor did they examine iVotronic machines that were used in the actual election. The team performed no analysis based on compiling and executing the software, either on iVotronic hardware or in a "test harness." Such testing is commonly used to identify bugs that may manifest themselves only under obscure conditions. Likewise, the team did not review internal ES&S documents, such as their bug tracking systems or software repositories, which might contain important clues about the problem. For key issues, including how the iVotronic screen is calibrated and how its smoothing filter operates, the final report contained insufficient detail to determine how these issues may have impacted the undervote.
In total, the State's investigations have provided no persuasive explanation for Sarasota's undervotes. We recommend additional testing and analysis of both the software and hardware used in Sarasota. We also recommend analysis of ES&S's internal documents, including their bug tracking system and other versions (earlier and later) of their software. We estimate that this additional investigation could be conducted by an appropriate team of experts with about a month of work.