From: "Douglas W. Jones"
Subject: Re: Washington state emergency
Date: Tue, 10 Aug 2004 09:20:11 -0500
To: "Ellen Theisen at VotersUnite!"
Cc: "John Gideon"
On Aug 9, 2004, at 5:49 PM, Ellen Theisen at VotersUnite! wrote:
(Ellen Theisen) John and I just found out that six counties in WA have just now installed new software for the September primary and the November general election. Five are optical scan, one is DRE+optical scan. It was done because the old s/w wouldn't handle the new ballot design, but it was a choice of the counties to use the new ballot design. Sam Reed (SoS) has used his powers to call this an emergency change and the only certification that is being done on this uninspected s/w is running ballots through the opscans. I don't know what they did for the DREs.
(Doug Jones) This is not a good procedure. No software upgrade should be allowed without going through the ITA process.
In this case, if Washington used California-style mandatory recounts for the sake of checking the integrity of the systems, I'd be willing to accept this kind of "late in the game" patching. Without that insurance, I'd say it's a very bad idea.
(ET) Alta Price (Iowa) told me today that you said software shouldn't be changed after the candidate list is known.
(DJ) Indeed. If you allow patches after the candidates are known, the patch can be rigged to favor a specific candidate unless the vote tabulating machine has no knowledge of candidate names or party affiliations, but only counts votes by
Things are even worse if the patch is installed after assignment of candidates to specific ballot positions, because then, the patch can be rigged to favor one specific ballot position.
Such patches can easily be rigged to pass most logic and accuracy tests with flying colors while being biased with real votes. The reason is that most logic and accuracy testing is done with "perfectly marked" ballots. The bubbles on a blacken the bubble ballot are perfectly blackened. The perverse patch could favor one candidate in some ballot position by having a different counting threshold for that candidate than for their opponent, so it would count the mark if 15 pixels were dark for the favored candidate, but require 35 dark pixels in their opponent's ballot position, where a perfectly blackened oval had 50 dark marks.
There are ways to improve logic and accuracy testing to touch on this issue. The best tests include not only perfectly marked ballots, but ballots that are carefully marked with a graded series of marks, from blank ballots to perfectly marked. I always include such tests in my state certification testing of mark-sense scanners, but I gather that many states and many vendors do not routinely perform such tests. I consider this bad.