<
http://www.ss.ca.gov/elections/touchscreen.htm>
<
http://www.ss.ca.gov/elections/november2004_pmp_report.pdf> (203 pp with appendices)
The California Secretary of State has completed their report of testing in the ten California counties that used touch screen voting - six were Sequoia, two were Diebold, one was Hart/eSlate, and one was ES&S. The accuracy was near 100%.
But the state was clear that this level of testing was not enough. On the first page they wrote: "Notwithstanding this additional level of testing, there are forms of malicious code that could affect the accuracy of a voting system that would not be detected by federal, state, local or parallel testing. Other detection methods, such as the Accessible Voter Verification Paper Audit Trail (AVVPAT), are necesary to expose these types of election tampering."
The machines were selected, at random, approximately two weeks before the election. They were then sealed and segregated from the remainder of the touch screen machines. The testing occurred on November 2 with teams of local testers.
I would think that this type of random testing would catch any problems emanating from the source code, i.e. the manufacturer or distributor, but not any tampering that may have occurred between the date of selection for the test and the election. Nor would it catch any tampering with the central tabulator/memory card reader at each county.
In 9 counties the printout of the test results was made at the end of the day and all testing materials, including the printout, were taken by the test leader personally to Sacramento, delivered to the state the next day, Nov. 3. The exception was Riverside. A printout of the test results could not be made in Riverside because their DREs did not have printer capability. Instead, Riverside overnighted the memory card of the DRE to arrive at the state on Nov. 4 and the test results were printed out there.
The report summary says there was 100 % accuracy (after resolving - by videotape - some cases of voter/tester error), but the report detail says that in one case (Merced) the tester attempted to vote for Bush (confirmed by videotape) but instead was recorded as for Peroutka. It was not a case of voter (tester) error.
I'm glad the state took this step and I'm glad that all California DREs will require an AVVPAT by the 2006 election. But this testing still had too many loopholes to be completely convincing.