The Arizona Senate’s audit of 2.1 million fall 2020 ballots has been extremely controversial since its inception. As recently retired Arizona Republican U.S. Senator Jeff Flake reiterated on May 11, its premise is based on “the ‘big lie’ that the 2020 election was stolen from Donald Trump.”
The audit’s lead contractor, Cyber Ninjas, is a data security firm whose CEO is pro-Trump and has not been certified by federal election administration regulators. The firm has no prior experience vetting vote counts. Its main subcontractor, Wake TSI, took part in a controversial election audit last fall in a tiny rural Pennsylvania county. The auditors have been fighting with Democrats and Republican officials who oversaw Arizona’s 2020 general election about access to voting machinery, computer systems, paper ballots, procedures, transparency, security and more.
But the audit has proceeded at Phoenix’s Veterans Memorial Coliseum. As details emerge from the arena’s floor, it appears its hand count of presidential and U.S. Senate votes from Maricopa County will likely produce results that diverge from the county’s official 2020 results, where Joe Biden beat Donald Trump by 45,109 votes or 2.16 percent. (Statewide, Biden won by 10,457 votes.) The reason for the probable discrepancy is not the hyperpartisanship surrounding the Senate’s audit, but because the hand count is imprecise at key junctures.
Voting Booth’s assessment is based on its role as a floor observer on May 6 and 7, which included strict limits on interviews (for instance, only being allowed to speak with Cyber Ninjas’ attorney or designated technicians). Voting Booth’s conclusion is also based on its review of state and county election procedure rules and guidelines, and consultations with outside lawyers specializing in post-election procedures and other observers allowed in the tightly watched coliseum.
“Proper recount procedure and protocol contain several indispensable components and requirements that must be rigorously adhered to. If any are missing, a manual recount could become inherently flawed. As a result, inexperienced people overseeing the count might not even be aware of errors or be able to correct mistakes,” said Chris Sautter, a veteran recount attorney and co-author of The Recount Primer, a guide to post-election disputes.
The Senate’s hand count
There are two different audits underway at the Phoenix arena. (Both inquiries had to pause and pack up before the weekend of May 15-16 because the coliseum was used for high school graduations. They are expected to resume afterward. A third audit, examining the digital ballot images created by the county’s vote-counting scanners and not associated with Cyber Ninjas, has yet to begin.)
The audit drawing the most attention, and derision, by career election officials including Republicans, is a camera- and microscope-centered examination of returned paper ballots to detect forgeries. This process examines folds and fibers in the ballots, as well as ink markings, and was sparked by unproven allegations that thousands of paper ballots were printed in Asia and smuggled into Maricopa County, according to other audit observers. Even the audit’s liaison, former Arizona Secretary of State Ken Bennett, a Republican, who says the exercise is needed to quell concerns about untrustworthy elections, told Voting Booth that scenario was “crazy.”
But there is another audit going on, a hand count of Maricopa County’s presidential and U.S. senatorial votes. This process is more visible—via video feeds—and is spread out across most of the coliseum’s floor. Seen from afar, the optics resemble recounts in other states. Yet the hand count is not being run by government election officials. Nor is it following Arizona’s Elections Procedures Manual, or using all of the 2020 election records and data that Maricopa County provided to the state Senate (after a court’s February order that the county needed to comply with a subpoena). Instead, it is following procedures set up by Cyber Ninjas.
As a result, the hand count omits key accounting controls at important intervals where discrepancies with the official results can be identified and investigated, and any mistakes related to the hand count can be corrected. Three decision points stood out in this regard. For example, the hand count team was not setting aside problematic individual paper ballots after counting teams used their judgment to interpret a voter’s intent on a sloppily marked ballot.
Nor was Cyber Ninjas’ hand count looking for differences with 2020’s official results based on cross-referencing its subtotals with the county’s data. Notably, it has not reviewed the subtotals from so-called poll tapes on each of the voting machines used on Election Day (when 168,000 people voted). Nor was it using the subtotals from 9,600 batches of early ballots processed before November 3 but counted on Election Day.
(Those subtotals would have to be extracted by Cyber Ninjas from data given to the Senate, which experienced election auditors know how to do. However, as of mid-May, Cyber Ninjas’ team had not examined all of the Senate’s data, according to observers who are familiar with that aspect of the operation. Arizona Senate President Karen Fann, a Republican, wrote a May 12 letter to Maricopa Board of Supervisors Chairman Jack Sellers, a moderate Republican, expressing frustrations that the county’s data was not readily accessible. A legislative hearing has been scheduled for Tuesday, May 18.)
Regardless of these frustrations, Cyber Ninjas’ hand count has proceeded without comparing its step-by-step results to the key baseline of the county’s subtotals. Additionally, as of Friday, May 7, when about 250,000 ballots had been hand-counted, the Senate’s contractors had not begun to copy the thousands of individual tally sheets from the hand count, nor had they begun to look for, and fix, possibly data-entry errors when those subtotals were entered into a spreadsheet to compile the overall vote totals. (The next week, however, video feeds appeared to show tally sheets being scanned.)
Blurry process where precision matters
Politics aside, verifying election results can be seen as a big accounting problem, where all the figures have to add up. Election audits typically are laborious and mind-numbing. When precise records are not created and cross-referenced with underlying data in manageable steps, it becomes harder to find and trace errors, and mistakes by the auditors can become embedded and taint the process’s conclusions. Arizona’s hand count audit is being run by a lead contractor with no prior election experience, which means that controls at key comparison and procedural points were not instituted, or were added after the audit began.
The audit’s first soft spot concerns tagging individual ballots where a voter’s intent is not clear. The hand count starts by counting paper ballots in the order in which they were batched and stacked in the county’s storage boxes. The audits count 50 ballots at a time. A table manager puts the paper ballot on a rotating stand. The three counters view and write down the presidential and U.S. Senate choices on a tally sheet. A supervisor from Wake TSI, a subcontractor, oversees this process. Using green pens, the counters record each vote as a single pen stroke. Five votes are recorded on each line of the tally sheet. After 50 ballots are counted, their slash marks are manually subtotaled. The process is repeated until votes from 100 ballots fill each tally sheet page. The arena floor had 40-plus counting tables.
The counters initial and write the ballot batch number at the top of the tally sheet. If a dispute arises between the counters over a voter’s intent on a single ballot, the vote is assigned to whomever two of the three counters agree upon. A red pen is used to correct the prior green ink mark. Once the hand counting round is finished, the counters’ math is checked by the table manager. A cover sheet is prepared and a runner takes it to a tabulation station.
Cyber Ninjas’ process, which has been fine-tuned as the audit proceeded, differs from the county’s process in several ways, starting with adjudicating voter intent, according to Maricopa County Elections Department documents and the state’s 2019 Elections Procedures Manual. These procedures, which were approved by Arizona’s Republican governor and attorney general, include bipartisan audit boards with members appointed by political parties, specific voter intent standards, and adjudication logs for individual ballots.
The Senate’s hand count lacked bipartisan representation, as Arizona Democrats have boycotted the audit. Counters at each table are using their own judgment about how to count questionable votes. Problematic ballots are not set aside for later review. In addition, stepping back, the audit is not producing a crisp record that can rapidly trace disputes to the originally questioned ballot. Whether this omission will create future issues is unknown, but it is a weakness.
More specifically, the hand count tables are not comparing their subtotals to the county’s subtotals at key vote counting intervals, starting with the voting machines (or their memory cards) used on Election Day when 168,000 residents voted. Comparing the hand count tallies to these cashier-like poll tapes (or the same information on the machinery’s memory cards) would be the most direct way to cross-check the official results in precise intervals, recount experts said. This baseline comparison was not being done.
Nor were the hand count teams comparing their subtotals to the subtotals for the rest of the 2020 general election ballots, which mostly were bundled in numbered batches of 200 ballots each. (In Arizona, these are called early ballots—as they are cast before Election Day. In other states, these ballots are called absentee ballots and in-person early ballots.) Generating those batch subtotals requires extra work, Maricopa County officials said; but they emphasized that these batch subtotals could be extracted and compiled from the data turned over to the Senate.
The Cyber Ninjas’ team also discovered what many election auditors face: ballot inventories are sometimes in less-than-perfect condition. For example, not every storage box has paper dividers between the batches packed inside, Fann’s May 12 letter to the Maricopa County Board of Supervisors chair said. Some batches do not contain precisely 200 ballots, the letter said, citing issues with “handling, organization, and storage of ballots.”
Nonetheless, the subtotals used by the Maricopa County Elections Department to tally its 2020 election results are not the same as the subtotals being created by Cyber Ninjas’ hand count team. Comparing these records before generating a final total is akin to comparing apples to oranges—it is inherently imprecise and lacks cross-referencing.
Not yet tracing data entry errors
These procedural steps are technical. But they are the building blocks and evidence upon which vote totals are created, and upon which accusations of stolen elections are proven or disproven—if accusatory partisans want to heed the facts.
There was another omission in Cyber Ninjas’ procedures closer to the hand count’s finish line that is simpler to follow and could be consequential. This gap is where the results from dozens of counting tables have been entered into a single overall database to compile the audit’s presidential and senatorial results.
Stepping back, the completed tally sheets, which record 100 ballots on each page, are taken by a runner to a line of tables at one end of the arena. There, the sheet’s subtotals are re-checked by a Wake TSI employee, who, in turn, passes the forms and chain of custody cover sheet to other Wake TSI employees. They, in turn, input the subtotals into an Excel spreadsheet.
The individual tally sheets are not dated. Nor, as of May 7, when Voting Booth had observed floor operations for two days, were they copied before their totals were entered into the overall results spreadsheet. By then, about 250,000 ballots had been hand-counted. That meant there were at least 7,500 tally sheets (three per ballot) and possibly 1,000 or more “chain of custody” cover sheets awaiting backup and data-entry verification.
When asked about this omission on May 6 and 7, Cyber Ninjas’ attorney, Bryan Blehm of Phoenix, pointed to unused computers on tables that he said would eventually copy the tally sheets and check if that data was correctly entered into the results spreadsheet. A week later, when 330,000 ballots reportedly had been hand-counted, it appeared that tally sheets were being scanned, according to video feeds. A line of plastic storage bins could be seen near this operation.
The likely outcome
Blehm defended Cyber Ninjas’ process as carefully constructed. It is a determined effort to conduct a massive hand count, even though its procedures, despite being sanctioned by legislators, are not following Arizona’s Elections Procedures Manual. But Cyber Ninjas did not appear to know what inventory and accounting controls its process, said to cost more than $3 million, lacked. Or perhaps it did not care.
For example, after 2020’s Election Day, Maricopa County was required to conduct a hand count of 52 batches of early ballots—which Bennett said was insufficient to attest to the county’s results, where Biden beat Trump by 45,109 votes. That audit was conducted by “ballot boards” where political parties independently appointed the boards’ members, state-issued voter intent standards were followed, and the public could attend—unlike the coliseum audit. That neutral approach, common to most states, guards against partisan favoritism and legitimizes the election’s outcome.
Cyber Ninjas’ hand count process anticipated such an adjudication board, Blehm said. But he said that no voter intent disputes necessitated convening that panel. With Democrats boycotting the hand count, there were fewer disagreements. On the other hand, Voting Booth observed table managers and Wake TSI staffers asking about handling ballots and related issues, such as what to do with food-stained ballots.
The Senate’s audit will not change the legal outcome of the state’s 2020 election. Overcoming a 10,457-vote statewide lead is unheard of, according to recount lawyers. At most, recounts alter outcomes when margins are several hundred votes or fewer, and sloppily marked ballots are fought over—one by one—in lengthy public proceedings.
While it is unclear what results will emerge from the hand count in the coliseum, it is unlikely to mirror Maricopa County’s certified results—and may not be poised to present precise evidence for apparent discrepancies. Where that leaves the public and partisans is an open question. As one floor observer was overheard saying to another observer during a break from counting ballots cast in late October, it appeared many of these early ballots were overwhelmingly for Biden.
“I hope they are fake ballots,” she said, “because I am seeing so many Biden.”
This article was produced by Voting Booth, a project of the Independent Media Institute.
Steven Rosenfeld is the editor and chief correspondent of Voting Booth, a project of the Independent Media Institute. He has reported for National Public Radio, Marketplace, and Christian Science Monitor Radio, as well as a wide range of progressive publications including Salon, AlterNet, the American Prospect, and many others.
