Press reports on the recent New Jersey voting discrepancies have been a bit vague about the exact nature of the evidence that showed up on election day. What has the county clerks, and many citizens, so concerned? Today I want to show you some of the evidence.
The evidence is a “summary tape” printed by a Sequoia AVC Advantage voting machine in Hillside, New Jersey when the polls closed at the end of the presidential primary election. The tape is timestamped 8:02 PM, February 5, 2008.
The summary tape is printed by poll workers as part of the ordinary procedure for closing the polls. It is signed by several poll workers and sent to the county clerk along with other records of the election.
Let me show you closeups of two sections of the tape. (Here’s the full tape, in TIF format.)
Above you can see the vote totals on this machine for each candidate. On the Democratic side, the tally is Obama 182, Clinton 179. On the Republican side it’s Giuliani 1, Romney 13, McCain 40, Paul 3, Huckabee 4.
Above is the “Option Switch Totals” section, which shows the number of times each party’s ballot was activated: 362 Democratic and 60 Republican.
This doesn’t add up. The machine says the Republican ballot was activated 60 times; but it shows a total of 61 votes cast for Republican candidates. It says the Democratic ballot was activated 362 times; but it shows a total of 361 votes for Democratic candidates. (New Jersey has a closed primary, so voters can cast ballots only in their own registered party.)
What’s alarming here is not the size of the discrepancy but its nature. This is a single voting machine, disagreeing with itself about how many Republicans voted on it. Imagine your pocket calculator couldn’t make up its mind whether 1+13+40+3+4 was 60 or 61. You’d be pretty alarmed, and you wouldn’t trust your calculator until you were very sure it was fixed. Or you’d get a new calculator.
This wasn’t an isolated instance, either. In Union County alone, at least eight other AVC Advantage machines exhibited similar problems, as did dozens more machines in other counties.
Sequoia, the vendor, is trying to prevent any independent investigation of what happened.
Tomorrow: Sequoia’s story about how this happened, and why it’s inadequate.
UPDATE (March 20): We now have copies of nine anomalous tapes, including the one shown above. They’re on our New Jersey voting documents page.
I saw that the Census beurea scrapped PDAs for because pollers did not know how to work computers. Felton probably advised on that one too
I never heard of such system with two processors from different manufacturers but it sounds like a great idea.
I heard Sequoia is changing their company slogan:
“2 + 2 = 5 (for extremely large values of 2)”
There’s even a T-Shirt over at ThinkGeek for them:
http://www.thinkgeek.com/tshirts/generic/60f5/
On a more serious note, could it be that the two sections (per candidate and per option) are stored on two separate pieces of electronic hardware? That would be a REALLY good idea for a manufacturer to do so that the system could perform a checksum (i.e. 20 != 21) and verify that the data was correct.
Hardware fails all the time, shouldn’t there be some sort of built in system to verify the calculations?
At my company we have some system boards with two processors on them from two different manufacturers. The commands are run through both processors and the results are compared to verify they are the same. That way you avoid any silly Pentium floating point problems or other possible hardware flaws.
You would think Sequoia would come out and say THANK YOU for pointing out our failsafe checksum feature! A failed checksum should prompt the county to go through and hand-count each ballot to verify the results recorded on the machine.
I get the sense that TOm just doesn’t get it…
I get the sense that Felton would only be happy with chads and ballots, pencils and paper. I saw that the Census beurea scrapped PDAs for because pollers did not know how to work computers. Felton probably advised on that one too. Jeez, seems like he is one of those guys that says its is statistically better to have millions of paper ballots being counted by biased pollworkers, than to use “oh my god” computers. Yeah that is more reliable. Well being a professor doesnt pay that well, you need to do something in your spare time to make change.
Anybody got a buggy whip?
It’s hard to judge the quality of the tape printout from that scan. It may well be that the printout simply wasn’t very dark (there’s a tradeoff between darkness and ribbon life) and the scanner was not adjusted for optimal capture. Given that some of the signatures are also ‘broken up’, I’d blame the scan far more than the printer.
That isn’t to say that the program shouldn’t have used double-wide mode for the vote tallies to ensure that they could be faxed cleanly. Indeed, if the printer would allow it, it might have been good to use a graphics font with more distinctively-shaped digits to allow for 100% certain recognition even from a really crummy scan.
I didn’t read all of the notes here. But I did read quite a few. Most were trying to explain what may have happened and others were concerned about the “!” (duh). It doesn’t matter what went wrong people! E-voting is flawed because it can be rigged, it can be mistaken, it can have wrong counts! Yes, I know that the punchcards can have the hanging chad or not punched thru or ripped or whatever and are counted by computers. But, they can be recounted on another machine or even counted by hand, ie: Florida. I have not trusted E-voting and will never trust E-voting. It can be wrong and it is very unlikely that it will be detected. The Sequoia issue may turn out to be some dumb error. The ones you need to worry about are the ones that show no errors. Sequoia, by-the-way, are complete morons for not agreeing to an independent review. There should be multiple independent reviews. There definitely needs an investigation as to how these were “certifed”, but made errors anyway. The US voting system has become a laughingstock in the world. And people like Jimmy Carter and Jessie Jackson have gone to other countries to verify there is no voter fraud. Who will come to our country to verify ours??
Suck4rz, we pwn ur v0t3!! n00bz…
Hello,
This sort of thing is exactly why we don’t want to trust companies with “special” agendas. Who knows what they are up to and who put them up to it. Since “President Bush” ran for election all kinds of accounting and counting problems have arrived. The President has an MBA yet since the S&L scandal to YTD housing (mortgages) scandals we the people have suffered. I know that President Bush has rigged the oil industry to sabatoge us with $5.00+ oil prices once the democrats are in office.
Please find a way to make it right…for the people’s sake.
Claude N.
Regarding the quality of the tape printout:
In 20 years of using various brands of dot-matrix printers, I have never seen such poor quality print from a supposedly brand-new print device. Sure, one or more of the pins in the 9-pin printhead will eventually fail, but not on a system that’s only recently been put into use!
If this is representative of the quality of all the printers they use, what other shortcuts are they taking with the remainder of the voting machines?
The only way if there are discrepancies and Sequoia blocks any neutral investigation: None of the voting machines will be ever used for any kind of voting again. Everything else is anti-democratic.
Just because the voting machine was activated for a specific ballot doesn’t mean the person actually voted at all. It is common for someone to go to the polls, start the ballot and not vote. They may have wanted to vote on a local issue or something else.
I don’t vote in any uncontested elections, therefore it would show that the machine was activated for a vote, but no vote was cast.
One more note about the exclamation point for Giuliani: look at the tape for Cranford, District 27. The serial number, “17534”, is printed at both the top and the bottom. In the bottom version, the “1” looks just like an “!”.
I’m convinced this is just a low-quality printer, which is bad enough, as an 8 could easily be misprinted as a 3 or 0, and 91 votes could become a really emphatic 9!
So, these guys got a nice government contract and pushed the product out the door without any testing. Owned by BUREAUCRACY!
(But who was owned? Seqouia? us? US? Democracy?)
Why not just do what the UK does, and stick to paper voting. It’s simple, reliable, has no security flaws, and we always manage to get the results counted just fine. Computers simply have no place in the voting process.
There is a lot of bickering about the Republican numbers but they don’t really matter. The Democratic numbers are CLEARLY WRONG. They do not add up and there is no denying it! Therefore the machines are not to be trusted.
PS: Wow, I didn’t mean to set off a firestorm around the !.
To the programmers on this thread: Other possibilities include using an “until” loop in one place and a “while” loop in another or updating a counter before a read in one place and after a read in another. Now it beats me how that could happen; you’d think they’d use a common subroutine to do the same counting function for both parties.
Whatever is going on, it looks like we obviously have some questionable quality control on the software. That’s also indicated by the bug Greg H pointed out. Crappy software engineering processes are nothing unusual, and it’s a lot simpler explanation than election-rigging, IMHO.
Ashley,
it is not possible to say that any particular machine worked correctly. If you cannot trust the accuracy of a machine, the fact that the totals match may be down to chance.
The Sequoia document gives a reasonable description of the behaviour of their machines, but the described behaviour is seriously UNreasonable. There are 12 “option switches” or buttons on the operator panel, numbered 1 to 12. Button 6 means Democrat, button 12 means Republican. Pressing button 6 followed by any one of 1-5 or 7-11 means show Democrat on the operator panel and a Republican ballot in the voting booth.
According to the document, this happens because the “Panel Mode” defaults to “Best Fit”. Best fit??!!?? What idiots made this behaviour _possible_, let alone default?!?
According to the document, the expected usage is pressing one, and only one, of the _assigned_ selection buttons, followed by the “Activate” button. _Any_ other sequence of button presses should signal an error and prompt the operator to start afresh, _not_ make a bad guess as to what the operator intended. Good grief…
And, in the amended configuration, it seems that to cancel an entry you press the “Test” button.
No wonder they’re afraid to have their stuff looked at.
This problem would also occur if the machine thought that Giuliani were a Democrat. If Giuliani were accidentally mis-categorized as being on the Democratic option switch, then his 1 vote would fall onto that option switch instead of the Republican one.
This doesn’t get them out of the woods, though. It suggests irregular mis-reconfiguring of the settings on the voting machines, which is an even more serious issue.
Thank you so much for doing this. Keep it up!
Yes, that looks like some vote-shifting attempt gone awry because it didn’t properly take into account the party ballot selection.
I’d expect that early attempts to write code to corrupt voting machines would sometimes have inadequate quality control, allowing bugs to “escape to product delivery” and result in visible anomalies.
The question will be whether these will be enough for the good guys to get proper auditing in place before the bad guys improve their quality control and take the pressure off.
Wow. Well, we have some good consistency here in the error. Every time the Republican vote total is too small and the Democrat total is too big. It’s nearly always by one vote, but as noted, in Crawford, it’s off by 2 for each party. My first thought was that maybe this was an error that happened with every X number of votes, so that the more votes, the biggger the error, but the vote totals on that tape are comparable to the totals on the other tapes. The only other difference I could see immediately was that this was the only machine that recorded any votes for Fred Thompson, though I’m skeptical that this has anything to do with it.
So…. the tapes *appear* to be correctly adding up the total number of votes cast. They are not, however, correctly adding up totals within each party, and the size of the error is not consistent, although the error is always positive in one party and negative by the same amount in the other. If it was always off by one it would look like a very simple error somewhere… but it’s not. Why is Crawford off by two when it otherwise looks very similar to the other results?
Furthermore, the 1 vs ! issue seems easily explained now… these tapes are terrible to read. The last one is nearly illegible. This in itself is an issue… if we can’t tell whether it reads 88 or 80, 30, 33, 38 or whatever, then how are we supposed to use these to authenticate the vote?
Thank you, Ed, for making this information public… our voting process needs to be as transparent as possible, and we’ve had more than enough problems the last few times around without blatant errors like this turning up.
The New tapes.
Not all of these are anomolous per se: some just have more vote attempts than vote casts (e.g. elizabeth 2)
But the first one, (crawford) shows 2 more republican votes cast (57) than attempts(55).
I’ll also remark on an easy mental error one can make. One might be tempted to say something like “well it’s just one vote so It probably ain’t deliberate faud”. There’s counterpoint to that. Suppose that instead a perpetrator were not simply stuffing votes but instead were moving votes from one candiate to another. Or suppose they inserted some initial “negative votes” (as has been shown possible on diebold machines.) to balance out some added positive votes.
Having the registers come up off by just one, might well be a symptom of such activity if the perpetrator made a mistake.
if N[guilinani] > 6:
N[guilinai] -= 5
N[romney] += 5
else:
N[romney] += N[guilinani]+1
N[guiliani] = 1
That bit of code tries to vote shift 5 votes from guilinani as long as guilinai has 6 or more votes, otherwise it just takes all but one.
But I made an error in the code. should have been:
N[romney] += N[guilinani]-1
Sorry, the 404 errors should be fixed now.
Ed, we’re getting a 404 error from the update link.
Something is wrong with that site …. going to that link for the other tape images, the blog tells me the page was not found.
Ed, your new link to the New Jersey voting documents page results in a 404 currently.
The “Gee, isn’t it possible…” sarcasm seems a bit out of place here, and quite unnecessary.
It certainly could matter if that is a “!” or a “1”. If it is a “1”, it does matter that the totals are off, even by one vote. But what matters more than both of these, is Sequoia’s attempts to completely control the voting machine analysis process. If it’s a case of ballots cast without a vote entered, Sequoia needs to explain this and show a clear record of it.
More importantly, why is our voting process being taken over by private companies with so-called trade secrets that they feel they need to safeguard? We should not allow even the appearance of impropriety or incompetence.
More evidence is available — we now have eight more tape images online. See the update at the bottom of the main post.
Marc G,
What happened here is that the Republican ballot was activated 60 times, and 61 Republican votes are recorded. That is a problem, no matter how you slice it.
Gee, isn’t it possible for the machine to be activated 61 times and have one person decide that they didn’t like any candidate and vote for no one so there are only 60 votes cast? This is the same sort of thing the Dems got so worked up about in Florida in 2000 – they seem to believe that every ballot must have a vote cast. Sorry, it doesn’t.
I just want to offer a little programmer insight here. With out more examples it’s hard to say definatively, more print rolls would allow one to extroplate what the machine is doing with more confidence.
But this seems a bit beyond a math problem this is a vote shifting problem. 1 vote is missing from the Democrats and one vote is added to the the Republicans. This type of tracking is core to this machines operation, At first it looked like someone had mistakenly set an initial value to 1 instead of 0 or something but indeed this actually seems worse and needs serious attention.
Perhaps there’s an explanation. This is not to say it was fraud. But the result is whack.
Wild ass guess 0:
The Gnarly thing here is that this is exactly what fraud might look like. Some admin tried to enter a bunch of votes ahead of time. “stuffing the ballot”. But he’s not sharp enough to know how the diagnostice will trip him up.
WILD ASS GUESS 1:
This kind of error has the fingerprints of “unit testing”. When you write code, a good coder will write a test routine that can directly call lower-level subroutines and bypass the upper level. This way they can rapidly detect if a mistake lies in the upper or lower routine. When the Code is done these unit tests are usually left in to also verify that any later changes to a lower level routine don’t have side effects on the upper level.
Perhaps some unit test runs, stuffs some test votes in as a legitimate test, then mistakenly forgets to remove these later or some concurrency issue prevents their removal.
Wild Ass guess 2:
The numbers on the tape mean something different than claimed.
The thing that makes it look like WAG1 (i.e. fraud) is that this was not reported on all machines. The other two WAGS might show up systematically.
After looking at the TIFF, I am still not sure if the mark in Giuliani’s column is a 1 or a !. However, the printout uses a fixed-width font, and the horizontal positioning of the mark is consistent with the other 1s in the printout (notice that the stroke of the 1s are all slightly left of center).
Meanwhile, other punctuation characters seem to be centered (e.g., the period, colon, and slashes at the very top and bottom; the commas in the text). But without seeing what a ! looks like in this font, one still can’t say for sure – the printout is simply too poor quality.
Good grief, check out the full .tif file, where they’ve written in the date –
“do hereby certify that on the _FEB_ day of _5_ _2000_ this board….”
Folks, look at the full tape image available via the link above the other two images. There’s a line in the top section of the tape that reads, “Public Counter” followed by “422”. In addition, underneath the Republican listing, it has a line that reads, “Total” followed again by 422. If we tally up all the votes, INCLUDING a 1 for Giuliani, we get this same total. So, for starters, that seems to lay to rest the ! vs 1 debate… it’s a 1, and the Republican tally is therefore incorrect. Furthermore, this seems to indicate that the machine is tallying total votes for both sides correctly, but not within each party. In this case it has taken a vote away from the Republican total and added it to the Democrat side.
I’d very much like to know just what the heck is going on here. This may be a minor thing in this case, but what if during the general election, every machine switches a couple of votes from one party to another? Furthermore, it makes me start to doubt all the numbers… if it can’t match up its own totals, why should we trust any of the numbers at all? Did it also swap a few votes around between, say, Hillary and Barack? Add a few arbitrarily? How can we know? Just looking at this tape, we can’t. I agree with Ed’s comparison to a calculator that can’t add right… if it can’t add two small numbers, you sure aren’t going to trust it to do your taxes with. There’s always going to be an issue of trust with large scale voting, but at least if everything added up you’d be somewhat confident that things were on the up and up. When they don’t add up, there’s no reason to be confident about anything.
I can’t believe that so many morons can’t understand the simple and exhaustive explanation that has already been given to them: the ‘!’ in Gulliani line is ‘1’ and has always been ‘1’. A clear, well defined 1. The illusion of ‘!’ that you see in the above picture is nothing else than just a scanning/processing/[maybe compression] artifact. I don’t know what kind of crap they used to “scan” it (maybe a cellphone), but the rest of the characters also don’t look much better than this ‘1’.
It’s rediculous. Last time Bush magically won Florida, now this…
Big lough for the rest of the world
Looks like a vote for Guiliani is a vote for the Democratic party!
If the person walked away then why is there an extra vote on the Republican side? If I open a Democrat ballot and walk away the vote gets counted as Republican?
As Robert says, what happens if a voter doesn’t make a selection? That would explain the vote total being lower than the option switch total on the Dem side. And if Giuliani’s “!” is not a 1, then there would be the correct number of votes on the Republican side.
If you don’t have to vote for every office, then to display consistency an ‘abstained’ line count should be shown for every section. That would prevent this sort of confusion.
In these machines you don’t have to vote for every office. It doesn’t have to add up. As long as the switch count is equal or greater, there’s no cause for alarm.
I agree, that looks like a ! in the Guiliani column. If the part of the program that does per-party totals mistakenly has Guiliani down as a Democrat, then the totals would match. I’d suspect a data-entry error in the ballot configuration for Guiliani (e.g. if the candidate’s party is represented in two different ways, and he’s down as D in one and R in the other), but of course we’d need more information to debug what’s really going wrong.
Perhaps it just a case where a person decided not to vote after walking in the booth. The low percentage makes this a plausible conclusion. I know there were times when I did not vote all categories on my county lever style voting machine in past elections. If you tried to match the number of votes to the sign in there would be undercounts in various elections.
What would happen if the voters goes through the line, gets up in front of the machines and say “Ah the hell with any of them” and just walks out or closes the screen without voting.
It would seem to be me that the machine would record the ballot being opened but no votes records. I know for myself there been plenty of elections where I only voted for one or two offices and ignored the others. Of course the old style voting machine my county in PA used wouldn’t record me in the booth only what switches were pulled down when I pulled the lever.
Perhaps this is a case where we now have extra information. As long as the votes don’t exceed the number of times the ballots opened I don’t see this as a problem as long as there is no other glitches. (Like failing to record a cast vote)
When viewing the tiff at high zoom, the Giuliani result contains all the pixels expected of a ! but is missing both the top and bottom serifs present on all other 1’s on the document, and pixels near the bottom of a 1. This is too many to be a coincidence, it looks to me to be a clear !, not 1. This means there are no extra votes in the republican column.
As for the democrat’s column, is it possible to abstain with this machine? Would an abstention go into Personal Choice or would it be unlisted? Ideally, abstentions should be allowed but reported in the tally.
Finally, in the other machines with errors, are the automatic summaries always larger than the manual sum, or are they sometimes smaller?
First, that’s terrifying.
Let’s not make a tempet in a teapot about the “!”… they are using a low-quality printer that has a very small pixel matrix, and a lot of them often don’t print. The “!” is a strict subset of the pixels in the “1”. If you look for other letters missing a similar number of pixels, there are a half dozen or so, including the P that looks like a “D” in “President” of the United Stetes.
Thanks Ed!
I’ll note that our inability to read the “missing dot 1” could be a pre-cursor of our next hanging chad. Probably not as bad, but poor quality control undercuts a selling point for these machines.
The “Guiliani” 1 looks similar to the “Ballot Version” 1 higher up on the scan of the full tape.
All jokes aside, what trade secret is Sequoia claiming they are protecting?
Maybe the USA really needs vote fixing to be able to chose some sane person for a president?
There’s a whole separate reason to distrust Sequoia: they have very obviously gamed the Federal certification authorities and test labs, withholding key pieces of their product line from certification that should have been tested.
As a result, only Sequoia knows exactly how the central tabulator and ballot layout processes work.
Appendix A of this document covers the issue in detail:
http://www.bbvdocs.org/sequoia/Maricopa-County-Elections-Report.pdf
The optimist in me wants to know how many machines worked properly. The pessimist in me wants to know if the error was consistently on one side of the aisle.
Hmm. I would initially suspect that the “personal choice” option is being mishandled. Another possibility might be mishandling of a “correction” option, where the total gets incremented when the user is presented again with the voting choices screen but the vote change itself gets correctly decremented from one candidate and incremented for another. I don’t know what to make of the “!” for Giuliani though.
Todd,
The same problem occurs on other tapes I’ve seen, without any ambiguous-looking 1/! characters. If this tape has a ! for Giuliani, that’s another bug — and the Democratic side still doesn’t add up.
Paul,
Another relevant fact is that the problem happened on some machines and not others. On at least one machine, the discrepancy was two votes rather than one.
For any programmers out there:
Any chance its using a 0 based array, then printing out the count wrong? Ie if you threw 61 votes into a 0 based array, the first vote would be in position 0, the last one in position 60, thus you’d have 61 votes.
Ed writes:
> That’s a numeral 1 (i.e., one vote) for Giuliani.
Is it? It sure doesn’t look like any of the numeral ones on the same page. Perhaps the scanned image is the problem, and it’s clear that it’s a “1” when viewing the actual printout. But consider: an icon containing an exclamation point is commonly used by GUI apps when an error occurs; it’s conceivable that the software prints an exclamation point to indicate an error condition of some sort.
Ed:
Do you happen to know if copies of these tapes are subject to freedom of information requests in NJ? IIRC, Jersey has one of the most byzantine open records laws, but if these could be obtained, it would be potentially very illuminating.
I’ve looked at two tapes so far. The one I reproduced here is off by one. The other one is off by two. We’ll have more analysis of the tapes later.
Well this incident looks really interesting. It would be good if it was investigated little bit close. Let the truth show.
Mike from warm winter coats guide. I hope you will like it there.
Are other eight examples consistently off by one vote (i.e. + 1 for one side/-1 for the other) as Joe says above?
I think you left out something important. (Maybe it was intentional saving it for tomorrow?)
That is, not only did other machines in different counties exhibit this behavior, but in all cases they were off by one vote. (right?) Sequoia’s explanation would at least seem feasible without that… but with each machine displaying only one vote off, that counsels us, to first order, to suspect something more systematic than simply “poll worker error” of the variety Sequoia describes.
The Giuliani-exclamation-point theory doesn’t match the other facts. For example, there are other tapes that show other numbers for Giuliani, with a similar mismatch in the vote totals.
Ah, what’s a lost vote here or there? It’s not like you’re talking a measurable percentage of the total.
Ermm… Nevermind. Sobering. From a political layman’s perspective, I just don’t understand how we’ve come this far. This kind of looseness does not advantage any one party, so it’s not a partisan thing. Every politician should want the system to be as accurate and as safe as possible.
So who’s pushing this stuff? Is it just lobbying and “hey, I paid for that brunch!” from the vendors? Can they really have that much pull?
I really hope there is some simple gear in this machine that I do not see, because the whole process just looks evil to me.
The “!” was a voter surprised that Guiliani actually had the gall to appear on a ballot anywhere near NYC.
But seriously, that’s a bizarre error; no wonder Sequoia doesn’t want their machines looked at. People were incensed when Pentium’s calculations were wrong in a far more subtle way; they should be outraged that this simple an error can happen with their votes.
But we’ve had things that should make them outraged about their votes and machine voting, and so far where are the mass protests? Where’s the outrage?
Admittedly the Tiff is really crappy, but the vote total for Giuliani looks nothing like the other ‘1’ s on the image. Could the ! be an indication that he dropped out of the race before this election was held (on Jan 29) ?
Let me hazard a guess: the vote for Giuliani accidentily got put in Obama’s basket, leading to the results of the Option Switch Totals.
Now, where do I collect the bounty offered for spotting the cause of this bug ;D
Adam,
That’s a numeral 1 (i.e., one vote) for Giuliani.
This may be nitpicky, but I read carefully, and I’m confused, what’s the ! for Giuliani?