2024 Early Vote Ballots Being Scanned
Do these systems look like they are working properly to you?
2024 Early Vote Ballots being scanned. (Link to video)
Executive Summary:
“Scanning” of 17 batches of Early Vote ballots, was observed in New Castle County (NCC) Delaware. A total of 3,220 ballots were scanned. A video of two machines operating side by side is presented showing ballots being rejected at high rates for a given “pass”. This is not an experimental study. These are actually 2024 ballots. Significant jams, pauses, and mid-pass cleanings were required during processing. The number of “passes” required to process a batch were tracked. The number of “Rejects” for each “pass” was tracked. “First Pass Rejects” (FPR) of ballots through the machines averaged 15% with FPR rates of over 50% observed. Thirty Five percent (35%) of the batches required more than 5 passes, and one batch required 13 passes. The number of “re-made” ballots is reported. It is believed the results do not achieve the EAC VVSG standards required for purchase of voting systems in Delaware. Currently the State of Delaware does not perform risk limiting audits on this part of the system. Based on the performance observed here it is recommended that the State of Delaware implement 100% redundant hand count audit of the Absentee and Early Vote Ballots and do this prior to certification of the 2024 election.
Details:
The scanning room in NCC is shown below. There are 4 machines. For purposes of this article I have identified them left to right as stations 1-4 respectively. It was reported by DOE personnel that they were set-up to scan Absentee Ballots at stations 1 and 2, and early vote ballots on stations 3 and 4. This article will report on a scanning session where stations 3 and 4 were put into action.
I graduated as a Chemical Engineer and long ago worked in industrial processing plants. Actually paper mills. Machines that processed paper at high speeds. As an engineer, as a citizen I was shocked at the process I observed. I watched several batches of ballots not being able to make a complete “pass” with-out getting “jammed” or “paused” several times. Also during each “pass” the machine would reject ballots if it could not read them.
The video shows 2 machines operating simultaneously during a single “pass” of ballots. What you see has been edited down from a 2 minute video. I edited out the portions of the video that showed the operator to respect his or her privacy. This is NOT a hidden camera. This scanning is an open public meeting, and advertised by the State of Delaware. I asked if I could use a camera at the beginning of the session and was approved by DOE personnel.
In the video you will see ballots being automatically fed into the machine going “around” the curved metal guide and then ultimately being either “accepted” or “rejected”. If the scanner “thinks” it knows what the ballot says, it “accepts” the ballot and deposits the ballot on the bottom of the three receiver trays. If the scanner “doesn’t know” what the ballot says it will deposit the ballot in the upper tray. During this video, notice how many ballots get deposited in the upper tray at station 4. You don’t have to count them one by one for detailed data. The primary message will be glaringly obvious. You may be shocked to see so many ballots getting “rejected” in a pass. That is not necessarily the worst case. In many cases, the machines didn’t even get through a “pass”. They, “jam”, or “pause” for some reason. At times you will see an error message like “multi-feed”. At times you will have a “multi-feed” and not see that error message.
Here is a screen shot after a “pass” showing what the machine thinks was the total and what is the number it thinks is in the reject (top) bin.
As you watch the video, and ballots being rejected, and the rest of this article please keep in mind a few things.
1. Logic and Accuracy Tests: (LAT’s) Each of these scanners should have passed an LAT and have been “ready to go”! The “Early In Person” machines that marked these ballots should have been through their own LAT’s and “ready to go”. Typically these “tests” amount to running about 5-6 ballots through each machine. Based on what you observe here how do you feel about the maintenance, testing, and preparation of this combination of machines?
2. NO QC AUDITS: For quality control, the “Day of In Person” voting machines in DE are subject to post election “Risk Limiting Audits”. Now these scanners that process all the Early In Person, and Absentee vote are not! I believe the Delaware Code requires that these machines be audited and have written and informed the Board of Election and the Department of Elections officially on the record in multiple occasions. They do not acknowledge the issue.
So what happens when ballots won’t go through the machine on the first pass? What happens is that the operator presses a few buttons and then the operator takes the “rejects” from the reject tray and puts them back into the feed tray and runs them again, again and again, hoping that there will finally be no rejects. Often between subsequent runs the operator turns the ballots over, or around 180 degrees or does a “cleaning” in hopes that will help. You may be thinking I am exaggerating with the “again, and again, and again” statement. I am not. One of the batches required 13 passes.
At times, no matter how many passes, a ballot just wouldn’t scan, or at least the machine implied it didn’t scan. If there are no audits, how does one really know? In any case, when the machine implies that the ballot didn’t scan, the ballot was “re-made” on a voting machine that is in the room. That new ballot replaces the ballot that didn’t run. The re-made ballot is witnessed and inspected and compared to the first by a second person that did not create it. I saw the original ballot segregated, I did not see it “spoiled”.
Summary Results:
The results of this processing is shown in the table below. A few highlight observations.
Of the total of 17 runs, only 1 batch of ballots made it through on the first pass. The highest number of passes was 13. The average first pass reject rate was on the order 15% with the highest being over 50%. A total of 6/17 or 35% of the batches took more than 5 passes to process.
In addition to “pass rejects” we must also consider that the machines jammed, had, error messages, and required mid system cleanings. Perhaps the best place to get a feel for those events is the chronological event log. Selected portions of my notes from the session are shown below:
Analysis:
I do provide a technical analysis referencing State of Delaware Code, it’s reference to EAC specifications and Voluntary Voting Systems Guidelines (VVSGs) but I thought I would just go with a common sense anecdotal statement and leave that technical stuff in an appendix.
The tasks I measured and recorded are paper feeding and scanning. That portion of the job is a whole lot like a copier. If you had a copier that worked like this I would be willing to bet you would conclude that it is broken, doesn’t work and you would throw it away. So that is what we observed. Unfortunately we as citizens have no idea how this machine tabulates ballots. I would argue that neither does the DOE. We never audit the results, how would you know? In one rare instance, we did have a re-count of an entire RD race in 2022. In the re-count they were not able to find .48% of the ballots that the machines reported. That would have been all the machines, not just the Absentee and Early Vote. As you watched the video and think of the description above do you now have any guesses why they couldn’t reconcile that count and where the problem might be? Let me ask you a question. Let’s say the way the State of Delaware operates and maintains, and tests these scanners, they do the “tabulation” part of the job every bit as good as they do the paper feed and scanning portion of the job. Does that give you confidence in the Early and Absentee portion of our elections or quite the opposite?
Conclusions:
In my opinion the observed performance relative my interpretation of industry standards is alarming. To me the observations bring in to question the maintenance programs, and the logic and accuracy tests that are supposed to prevent performance like this. Still, we are in the middle of an election. To not use the machines is not an option.
It is unconscionable that the State of Delaware, the DOE and the BOE have not performed risk post-election risk limiting audits on the Absentee and EarIy vote. In past articles it was for different reasons. Specifically the lack of audits would provide incentive for a cheater to implement electronic flipping or stuffing of the ballot box. What was observed here adds a strong case for systemic process errors.
Recommendations:
A significant portion of the vote has been processed already. So what should we do.
Once again the recommendation is for the State of Delaware, the DOE and BOE to implement hand count audits of the Absentee and Early Vote ballots. In past articles, I thought it would be reasonable to audit at the sample levels they use under Title 15, 5012A (c) for the “Day of In Person” machines, and that I believe they are required to do for the Early In Person and Absentee Vote. Given the additional observations here, I now believe there should be a redundant hand count audit of ALL the Early In Person and Absentee vote. I believe that this should be done prior to the certification of the election results as opposed to after the certification.
If you believe that is the case, reference this article and drop the Board of Election an email at: DE Board of Elections
Appendix: More Technical Arguments the Process has Failed.
Section § 5001A. (c) of the code states that:
§ 5001A. A voting device or system purchased by the State must be certified by the United States Election Assistance Commission (EAC), or designated federal authority, as meeting or exceeding the voluntary voting systems standards or guidelines (VVSGs) as promulgated by the United States Election Assistance Commission, or designated federal authority, before delivery to and acceptance by the State.
While this language does not specify which VVSG Standard applies the standard for “error rate” provided in VVSG 1.1 will suffice for this analysis. It states:
"All systems shall achieve a report total error rate of no more than one in 125,000 (8×10–6 )". (.0008%)
When trying to ascertain what is an “error” VVSG standards define a failure as an event that results in:
1. loss of one or more functions,
2. degradation of performance resulting in a device that is unable to perform its intended function,
3. automatic reset, restart, or reboot of the voting device, operating system or application software, requiring an unanticipated intervention by a person in the role of election worker or technician before normal operation can continue, or
4. error messages or audit log entries indicating that a failure has occurred. failure rate Ratio of the number of failures that occur to the volume of data processed.
I would argue that when looking at the result from these machines we are looking at the result of 3 systems. First the poll book printer prints the ballot, then the ballot marking device marks the ballot, and finally the scanner “Attempts” to scan the ballots. If we accept that this is three sub-systems then the acceptable error rate stated above could be multiplied by 3. You would have 3/ 125,000 or .0024%.
In the most forgiving argument, we could consider the “re-make” of the ballot the error. The error rate would be 7/ 3220 or .21%. This would be about 100x the acceptable error rate. If we look at item 2 above we might consider “degradation of performance” and the “inability of the machine to perform its intended function on the first pass” as the appropriate rate. That would put us at an error rate of 487/3320 = 15% or over 6,000x the acceptable error rate. When one considers the jams, error messages, and multiple cleanings required that the operator invoked to try to make the situation better either of the above estimates of error rate would be lenient.
You might look the video and jump to the conclusion that the problem was only with the scanner on the right (station 4). While many batches do have “less” rejects, than those for the two batches processed at station 4, the rejects jump right back up with new batches. This would be consistent with, but not necessarily proof of, the problem being with the Early In Person machines that printed the ballots.
Another note. When processing “Absentee Ballots” people often tend to blame the human voter for poor scanning performance. “Oh the voter didn’t fill in the circle right and used check-marks or they spilled coffee on it”. With Early In Person voting, a machine printed the activation card, a machine marked the candidate selections, department personnel, who are a part of “the system” handled the ballots in-between. Very few human touches and yet we see the results in the video.