The Hoffman files: counting votes in NY23

For IT people what's the lesson from the NY District 23 electoral fiasco? When you buy a hammer, you find yourself compelled to use nails as fasteners - and if a couple of nails don't the do the job the tendency is to drive in a few more until the nail holes themselves weaken the structure to the point of collapse. When that happens, as it seems to have in NY 23, it typically isn't the process that's wrong, it isn't the people, and it certainly isn't the nails or the lumber - the fault ultimately comes back to the decision to buy a hammer.
Written by Paul Murphy, Contributor on

About a week ago Richard Phillips wrote an article for The Gouveneur Times in upper New York reporting that a vote count for the recent District 23 special election provided by the St. Lawrence County Board of Elections contained numerous internal contradictions; including some arithmetically impossible results.

On Wednesday, Dec 2nd, he published a retraction and apology - explaining that the board had sent him the original information on the same day it certified the results, but had since sent him both an explanation of the errors and a set of results from which the contradictory data had been removed.

Here's the full introductory grovel:

This article [the original one] was based upon unofficial results provided to the author by the St. Lawrence County Board of Elections in a .pdf file on the same day that the election results were certified. These were not the certified results, and the author deeply regrets having said that they were.

Five days after the publication of this article, the Board of Elections provided an .xls spreadsheet of the certified results, district by district, in which only the numbers for ballots cast and blank ballots had been changed; and a .pdf file, dated Monday, November 30th, with numbers for blank ballots inserted, district by district. The changes in the numbers for ballots cast are duly noted in this revised article.

The Board of Elections has stated that only the numbers for "blank ballots" were computer generated in the original .pdf file, and that the "whole number" of ballots cast for each election district was entered manually. The data entry program then automatically subtracted the vote counts for each of the candidates and the remainder would appear in the final column as "blank ballots." In these six election districts (and perhaps others), data entry errors were made in the first column, for "whole number" of ballots cast, which resulted in the erroneous numbers in the final column, for "blank ballots."

Read this, and the rest of the article, carefully and you'll see the board cheerfully claiming that it made a lot of data entry errors, but all with respect to one field - and all later corrected in Excel.

The board likes coincident error in its explanations - more from another article by Dr. Phillips:

The Board of Elections has attributed the false initial numbers to human error. Poll workers mistakenly read the wrong line on the computer tape, or so the story goes. But votes were not only denied to Hoffman; they were delivered to Scozzafava. What obviously happened is that vote counts were switched. Hoffman?s tallies on the Conservative Party line were given to Scozzafava, and Scozzafava?s tallies on the Independence Party line were given to Hoffman. If all of Scozzafava?s 36 rightful votes in these three districts were on the Republican Party line, the result would be false tallies of zero votes for Hoffman.

Thus, for the "human error" explanation to be true, poll workers in three different polling places must have made the same two mistakes.

That same December second Gouveneur Times issue contained a related story by Nathan Barker headlined: VIRUS in the VOTING MACHINES: Tainted Results in NY-23. From his introduction:

GOUVERNEUR, NY - The computerized voting machines used by many voters in the 23rd district had a computer virus - tainting the results, not just from those machines known to have been infected, but casting doubt on the accuracy of counts retrieved from any of the machines.

Cathleen Rogers, the Democratic Elections Commissioner in Hamilton County stated that they discovered a problem with their voting machines the week prior to the election and that the "virus" was fixed by a Technical Support representative from Dominion, the manufacturer. The Dominion/Sequoia Voting Systems representative "reprogrammed" their machines in time for them to use in the Nov. 3rd Special Election. None of the machines (from the same manufacturer) used in the other counties within the 23rd district were looked at nor were they recertified after the "reprogramming" that occurred in Hamilton County.

It goes on to suggest a more sinister explanation than typos for the contradictions Dr. Phillips spotted in the original PDF:

Of further note, the models of ImageCast machines used in the districts have a slot through which the paper ballot is deposited into a secure holding tank underneath the machine after the ballot is scanned by the machine. The problem is that the slot is readily accessible to the voter (or poll worker) to stuff manually. 10 voted ballots could be deposited in the slot for every one voter... and if the electronic count was compromised, the "paper backup" would be useless.

The ImageCast machines have one more significant and scary flaw: USB ports. USB ports allow various devices to be attached to a computer in order to input information, connect a device, add wireless network capability and so on. Wireless network devices and USB storage devices can (and are) made small enough to fit into a regular wristwatch or bracelet.

And that, of course, combines with the board's explanation for the impossible numbers in its first report as system artifacts to produce some obvious questions about how those machines were chosen, installed, and supported - here's the applicable bit with the mandatory IT twist from a July 2009 press release announcing the transfer of Sequoia's responsibilities in New York state to its contractor, Dominion.

New York State Board of Elections co-chair Doug Kellner agrees that this transaction is not only good for Sequoia and Dominion, but is good for New York State.

"I am very pleased with Sequoia and Dominion's commitment to complete certification of the ImageCast ballot scanning system which incorporates state-of-the-art scanning technology. New York has the most rigorous testing standards in the nation, including compliance with the 2005 Voluntary Voting System Guidelines as well as additional voting integrity standards. I am also pleased that Dominion has made all of its source code available for review to the New York State Board of Elections and its Citizens? Election Modernization Advisory Committee, and has indicated that it will continue use of New York subcontractors for the manufacture, assembly and support of its system," said Kellner.

Sequoia has been around since the 1980s but has been the subject of numerous conspiracy theories rooted in the claim that it was rescued from terminal financial trouble by two nearly simultaneous events: it's takeover over by a group of Venezuelan businessmen and its return to exuberant solvency through a contract mediated by the Democratic party machine in Cook County Illinois.

Sequoia has been selling its products as open source for some time, but its actual commitment to this is less obvious than its press releases. For example, here's the applicable headline, dated December 1st, 2009 from their site:

Source Code for Sequoia's Frontier Election System Now Available for Public Download!

But this first code release turns out to be less than overwhelming:

This initial release contains source code for generating ballot PDFs within the election management system as well as the internal dependencies required for this functionality.

Overall, it's hard to separate fact from fiction in this mess, but I think some conclusions are pretty clear:

  • it is reasonable to believe that there's no way, now, to know who won that election;

  • it is reasonable to doubt any and all promises of openness and auditability by all concerned, from the systems vendors to the board of elections; and,

  • it is reasonable to think that the question of whether vote cheating took place or not; and, if so, whether the voting system was at the heart of the cheating, is now largely unanswerable from the data available.

From an IT perspective, it's the latter issue that's the killer: these people obviously tried to implement an electronic voting system and, in the aftermath, it's obvious that the system they selected influenced electoral process design decisions before the election, was at the heart of vote collection and counting failures during the election, is broadly unauditable, and is both the basis and the tool for the rationalizations offered after the fact.

There's an old saying about how you see problems when all you have is a hammer - and I think that's exactly what happened here. They bought what is fundamentally a proprietary client-server system with many consequent vulnerabilities that combined with the the processes they designed around its limitations to produce the current mess. Thus the rube goldberg totalizer whose thousand vote mistakes led to Hoffman's premature concession, the user hand holding in the voting booth that violated legal safeguards on voter privacy, the selection of a tainted supplier, and the utter absurdity of using an Excel spreadsheet to massage the results into compliance, were ultimately all the result of someone's inability to see electronic voting as anything beyond a mandate to replace individual voting machines with individual PCs running voting software.

Basically they saw only one possible direction, did their best to make it work, and now face the consequences: millions of dollars wasted, an election whose results are unknowable, and public trust in the system destroyed.

And that's really the bottom line lesson here: we can't fix what happened, but we can learn not to make this kind of mistake ourselves: not to let an unthinking commitment to an inapplicable computer technology get in the way of getting the IT job, whatever it happens to be, done: as cheaply, effectively, and accurately as possible.

Editorial standards