Hard Evidence of Voting Machine Addition Errors

goombah99 writes "Princeton Professor, Ed Felton, has posted a series of blog entries in which he shows the printed tapes he obtained from the NJ voting machines don't report the ballots correctly. In response to the first one, Sequoia admitted that the machines had a known software design error that did not correctly record which kind of ballots were cast (republican or democratic primary ballots) but insisted the vote totals were correct. Then, further tapes showed this explanation to be insufficient. In response, State officials insisted that the (poorly printed) tapes were misread by Felton. Again further tapes showed this not to be a sufficient explanation. However all those did not foreclose the optimistic assessment that the errors were benign — that is, the possibility that vote totals might really be correct even though the ballot totals were wrong and the origin of the errors had not been explained. Now he has found (well-printed) tapes that show what appears to be hard proof that it's the vote totals that are wrong, since two different readout methods don't agree. Sequoia has made trade-secret legal threats against those wishing to mount an independent examination of the equipment. One small hat-tip to Sequoia: at least they are reporting enough raw data in different formats that these kinds of errors can come to light — that lesson should be kept in mind when writing future requirements for voting machines."

Comments

lots of stuff going on

[goombah99]

In this case there are almost certainly multiple errors, one of which is the design error sequoia explained that causes the wrong ballot to be recorded.

Another plausible error mode here is the one the ES&S ivotronics had (and ones with old firmware still have). Certified voting machines are required to redundantly store the votes, usually 3 times, and there may be some effort to have these in different memory modules.

A while back ES&S had a bug that was triggered by a low battery voltage. The low battery condition would cause the logger to report this in the log. However the log entry was too long and cause a buffer over flow that over wrote the header of one of the redudant vote files. When the votes were read out at the precinct the machine did not notice the corrupt header and a second programming bug caused the malformed headers to cause other problems including mis-reported various things (like the maching ID) which then caused all sorts of downstream problems.

When the votes were read out by another method the corruption of the primary vote file was detected and it silently failed over to the secondary record. This produced a vote report that did not match up with the first one.

A reveiew of multiple systems was done by the Florida election supervisor who estimated about 1 in 7 machines reported wrong. He was fired.