May 03, 2022

In New Hampshire: A continuing plea for hand-counted audits

May 3, 2022

Dear Representatives:

I write in opposition to SB 366-FN, An act requiring an audit of ballots cast in the 2022 primary and general election.

As I have written to Senator Gray in the past, I am opposed to the lack of transparency and public oversight and increased reliance on hackable, riggable computerized election machinery that using scanners and ballot images for audits embodies.

No true election security expert endorses using a scanner to perform an audit, even if it's a new make or model. Manual/hand-counted post-election audits have been recommended since 2009 by both the Secretary of State's own Electronic Ballot Counting Device Advisory Committee [formerly at] and the national League of Women Voters [], as well as the experts who testified before the Kobach commission: Andrew Appel, Harri Hursti, and Ron Rivest. []

Verified Voting, the premier body of election experts, is unequivocal: "Audits require human examination of voter-marked paper ballots... Audits cannot rely on scanned images or machine interpretations of the ballots to accurately reflect voter intent" (p. 7, Principles and Best Practices for… Audits). [P. 7,

Let us stop to ask, Why do audits in the first place? It's because election security experts know there are too many undetectable vulnerabilities with computerized voting, and manual audits of hand-marked paper ballots are our best defense. So why are we turning to computerized scanners and digital images when the experts say we should be doing hand-counted audits of the paper ballots?

The convenience of letting a machine do the work of auditing with digital images is very seductive. Proponents will even argue that people are more error-prone than machines. The problem is that the public can’t oversee the software in a scanner as it computes votes. 

New Hampshire has a long, proud tradition of meticulous hand counts and recounts. The safeguards built into those can be utilized in audits and even strengthened to allow for maximum transparency and public oversight.

Here are concerns about the bill as written:

4. “I. The secretary of state shall randomly select one to 3 percent of AccuVote devices, but not less than 4 such devices, to be used at the 2022 state general election to be audited…. The selection of devices to be audited shall be non-public and made at least 2 weeks prior to the general election, after the AccuVote memory cards have been programed [sic]."  


The whole point of random audits is to make it impossible for *anyone* to know ahead of time which machines will be selected. Allowing the secretary of state to choose the machines in advance negates the protective effect of random selection.

4. "VI.  A random sample of not less than 5 percent of the ballots scanned shall be selected and the images of the ballots selected shall be compared with the voters' choices recorded for the ballot."

So this bill allows as little as 1 percent of the AccuVote machines, but not less than four, to be audited, and then as little as 5 percent of the ballots from those machines to have their images compared with "the voters' choices recorded for the ballot," presumably, the voters’ hand-marked paper ballots. If my husband, an engineer, is not mistaken, that works out to a rate of 0.05 percent of the AccuVote ballots, or 5 in 10,000 paper ballots that will be compared with the corresponding ballot images to try to establish that the digital images are accurate copies of the hand-marked ballots. [Even worse, I'm told that the comparison was between ballot images and the cast vote record (CVR), a computer-generated record based on the ballot images, rather than the paper ballots.] Does anyone see that as sufficient to inspire confidence? In contrast, Massachusetts conducts hand-counted audits of the paper ballots in 3 percent of all precincts when the president is on the ballot.

Here's what Professors Andrew Appel and Philip Stark have to say about this:

"Audit the Digital Images?

"Some vendors are promoting systems that create digital images of ballots. These vendors claim that the images make RLAs easier to perform because fewer (or no) paper ballots need to be inspected. That is incorrect: if a risk-limiting audit relies on images of ballots, it must check that the error in making the images from the voter-verified paper ballots plus the error the system made interpreting those images to make cast-vote records is not large enough to cause the electoral outcome to be wrong. It is a mathematical fact that this requires examining at least as many physical ballots as an audit that compares CVRs to a human reading of the paper ballots, without relying on the digital images." [P. 537,

In other words, as I understand it, an audit that relies on digital images of ballots requires examining at least as many corresponding paper ballots as a hand-counted audit of the paper ballots alone. Which begs the question, wouldn’t it be easier to just audit the paper ballots?

Professor J. Alex Halderman's research also demonstrates the risks of auditing digital images rather than paper ballots: 

"Abstract. As paper ballots and post-election audits gain increased adoption in the United States, election technology vendors are offering products that allow jurisdictions to review ballot images—digital scans produced by optical-scan voting machines—in their post-election audit procedures. Jurisdictions including the state of Maryland rely on such image audits as an alternative to inspecting the physical paper ballots. We show that image audits can be reliably defeated by an attacker who can run malicious code on the voting machines or election management system...." []

Please reject this cave-in to convenience and send a message that an audit requires the same high standard of proof that New Hampshire provides for recounts – nothing less. It must be hand-counted with ample public oversight.


Barbara Glassman

Nashua, NH