This is really quite disturbing.
The internal logs of at least 40 Sequoia touch-screen voting machines reveal that votes were time and date-stamped as cast two weeks before the election, sometimes in the middle of the night.
Black Box Voting successfully sued former Palm Beach County (FL) Supervisor of Elections Theresa LePore to get the audit records for the 2004 presidential election.
After investing over $7,000 and waiting nine months for the records, Black Box Voting discovered that the voting machine logs contained approximately 100,000 errors. According to voting machine assignment logs, Palm Beach County used 4,313 machines in the Nov. 2004 election. During election day, 1,475 voting system calibrations were performed while the polls were open, providing documentation to substantiate reports from citizens indicating the wrong candidate was selected when they tried to vote.
Another disturbing find was several dozen voting machines with votes for the Nov. 2, 2004 election cast on dates like Oct. 16, 15, 19, 13, 25, 28 2004 and one tape dated in 2010. These machines did not contain any votes date-stamped on Nov. 2, 2004.
You can find the complete set of raw voting machine event logs for Palm Beach County here: http://www.bbvforums.org/forums/messages/2197/6628.html
Note that some items were not provided to us and are ommitted from the logs.
The logs rule out the possibility that these were Logic & Accuracy (L&A) test results, and verified that these results did appear in the final totals. In addition to the date discrepancies, most had incorrect polling times, with votes appearing throughout the wee hours of the night. These machines were L&A tested, and the L&A test activities appeared in the logs with the correct date and time.
According to the voting machine assignment log, these machines were not assigned to early voting locations. The number of votes on each machine also corresponds with the numbers typical of polling place machines rather than early voting.
Many of these machines showed unexplained log activity after the L&A test but before Election Day. In addition, many more machines without date anomalies showed this log activity, which revealed someone powering up the machine, opening the program, then powering it down again. In one instance, the date discrepancy appeared when someone accessed the machine two minutes after the L&A test was completed
(Nod to NP, who is always outside one kind of box or another)
UPDATE: Props to ST at Poliblogger, who has been saying for a while that cards are better.