Critique of MITRE Report:

February 25, 2021 (rev 5-5-21)

Contributors:
Jared Black
Ray Bleharw
John Droz
Alex Kossovsky
Steve Miller
Eric Quinnell
Michael Sheliga
Stan Young
Et. Al

Table of Contents
Executive Summary ………………………………………………………….…..…….. 3

1 – Background and Approach ……………………………………………………. 6

2 – Alleged Ballot Harvesting in Georgia …………………………………..… 8

3 – Election Fingerprinting in Georgia ……………………………….……... 11

4 – Alleged Irregularities in Dominion Machines ………………….…. 15

5 – Voting anomaly in Michigan ………………………………………………... 18

6 – Butler County (PA) Lost Ballots …………………………………………... 20

7 – Appendix B: Duplicate Votes & Elapsed Time …………….………. 23

8 – Appendix C: Benford’s Law and Elections …………………………… 25

9 – Conclusions ……………………………………………………………………….… 28

Appendix …………………………………………………………………………………… 31

Page 2
 Critique on MITRE Report’s Executive Summary
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

The authors of this Critique are unpaid volunteers, whose expertise covers a wide range of fields (Cyber
Security, IT, Statistics, Physics, Economics, etc.). Our main interest is in assuring election integrity when
American citizens legally express their preferences for their representatives. In the last few months we have
generated multiple election-related reports — e.g. see Appendix.

As such we were quite interested when we heard that MITRE people had just released an election Report.
Based on its title, we were hopeful that this would be a thorough and objective analysis of the 2020 election,
as transparency has been desperately needed. Unfortunately, the content was a disappointment — so much
so that we decided that it was worth our time to put together a Critique of the MITRE Report.

Before wading into a Report like MITRE’s, it’s good to get an understanding of who are the authors, and what
are their professed values. We start with MITRE’s stated principles, and then observe whether that public
commitment carries through in practice here. MITRE’s Mission Statement includes such aspirations as:
“MITRE is dedicated to solving problems for a safer world. We work in the public interest to discover new
possibilities, create unexpected opportunities, and lead by pioneering together for the public good to bring
innovative ideas into existence.
“Objective Insight: We provide a conflict-free perspective with trusted access to bring proprietary data to
decisions.
“Technical Know-How: We combine applied science and systems thinking with deep domain expertise to
define the right problems and create solutions.
“Integrity Above all Else. Exercising the highest possible ethical and professional standards is fundamental
to how MITRE operates.”

MITRE’s Mission Statement sounds exemplary, but is it actually accurate in this Report? Of particular
importance in this national matter is objectivity and comprehensiveness. An organization purporting to
present “election data” to enhance “transparency,” would seemingly leave no stone unturned in doing an
objective and comprehensive analysis of this extraordinarily important issue — i.e. for the public good.
Regretfully, as explained in the subsequent chapters, we didn’t find that to be the case.
-----------------

Another general observation before we get into the nitty-gritty: the MITRE Report (“Report”) frequently uses
the word “fraud” in their document. We choose to rarely use that term for two reasons:
1) Analysis of election integrity is a legal matter. The legal definition of “fraud” is not the same as is
understood by the public when they use that word. In other words, the legal word fraud is a subset of the
public’s meaning of the word fraud — so it is disingenuous to use a key term that could be misunderstood.
2) “Fraud” is too narrow a description of the documented election irregularities anyway.
For example, if election officials “correct” an incomplete ballot, is that fraud? They claim to be (and
maybe believe they are) acting within their rights — even though that is disputable and the subject of
some lawsuits. Further, if they truly believed that such corrections were on the up-and-up, to assure
transparency they would have gone to lengths to carefully document all such alterations. Unfortunately,
that rarely happened. Worse, there is ample evidence that government officials blocked investigator’s
access to election systems and deleted logs to prevent effective audits (and transparency). So, again,
were their actions and omissions fraud?
Page 3
 Another example is in Central New York where they likely had the most closely contested election in the
country. In the House of Representatives race, out of 300k± ballots cast there was a difference of 20±
votes. This resulted in one of the extremely rare situations where a judge actually looked closely at the
election process, and how ballots were handled, tabulated, etc. In a damning ruling he determined that
in this one district (NY-22), that election officials had engaged in nine (9) different illegal actions! That
said, he took pains to say that none of them (legally) amounted to fraud.

As a result we try to use broader, more inclusive, non-legal terms, like “irregularities,” “anomalies” and
“malfeasance.” So let’s get onto our Critique…

We are in unison with the first sentence in the Report’s Executive Summary that: the 2020 Presidential
election brought an unprecedented and intense focus on the issue of election integrity. We applaud that
development. One outgrowth of this concern is that there were more election-related reports by independent
experts than ever done in the history of our country. For example, here are over a dozen (12) major reports
done about various aspects of the 2020 Presidential elections.

The Report assures us that: “To help ensure public trust and confidence by providing some additional
transparency, MITRE’s National Election Security Lab gathered and analyzed a wide range of relevant data.”
Inexplicably, not a single one of these well-documented expert reports is utilized in the MITRE Report — or
even mentioned.

Instead of considering the conclusions of independent experts, the Report says: “The team obtained data from
several election and state government websites…” So, it appears that the authors are trying to verify election
integrity by heavily relying on possible guilty parties for information. Additionally, the Report has multiple
references to journalists who purport to “fact check” election anomalies — by using unsupported statements,
again from the parties who may be guilty (e.g. election officials and the companies that provided the election
systems).

What’s missing here are any independent forensic audits, or investigations, pertaining to the 2020 election.
Surprisingly the term “forensic” doesn’t even appear in the Report. As a result, the readers of the MITRE
Report are no more informed about election integrity than if they followed mainstream media.
---------

There were more 2020 election-related lawsuits than ever in our history. The Report (on page 1) claims that
“the team maintained situational awareness of legal and media events occurring between October–December
2020…”. Since their Report was filed in February, why wouldn’t they have the same awareness in January?

Furthermore, as researchers, why would they arbitrarily only start looking at legal events in October, when
several lawsuits pertaining to the 2020 Presidential election were filed before then (e.g. see here). In any case,
using the Report authors’ artificially reduced time period of “legal awareness,” some fifty-five (55) lawsuits
were filed from the beginning of October to the end of December, 2020 (see here). Stunningly, not a single
one of these is discussed or even mentioned!
---------

Both of these unexpected significant omissions would lead a knowledgeable person to conclude that the
MITRE Report has no genuine interest in objectivity, and/or that they lack the technical competence to
comprehensively analyze a national matter of this importance.

Page 4
The rest of the Executive Summary outlines a few areas where the Report authors did a rather superficial
analysis. As mentioned before, in numerous reports by independent experts, there are over a hundred
anomalies listed — so checking out a half-dozen is hardly a scientific basis for dismissing the other 95+%.
Further, the methodology for even dismissing the handful of cases they selectively chose, would not be
considered to be rigorous.

Here's a thought: attorney Jesse Binnall publicly (on TV) testified under oath before Congress about election
improprieties. In that testimony he identified numerous documented examples of major irregularities in
Nevada — from some 130,000 unique voters! These included the same person voting twice (42,000+), the
voting person being officially listed as deceased (1,500±), the voting person not being registered in the state
(19,000+), the voting person not being a citizen (4,000±), the voting person using a non-existent residential
address (8,000+), etc. (Note Donald Trump lost Nevada by less than 35K votes, essentially all in one county.)

Binnall’s testimony was on December 16, 2020, well inside the window where the Report authors were
researching the election integrity issue. His report was about a group of scientists who had carefully
researched hard numbers for significant examples of voting irregularities. Why didn’t the Bald Eagle people
take that data, and double-check it?

Binnall states: “Our data scientists made these calculations not by estimations or statistical sampling, but by
analyzing and comparing the list of actual voters with other lists, most of which are publicly available… Our
determined team verified these irregularities without any of the tools of law enforcement, such as grand jury
subpoenas or FBI agents. Instead, we had less than a month to use critical thinking and elbow grease to
compile our evidence… To put it simply, they explained their methods so that others could check their work.
Our evidence has never been refuted…”

It's a shame that the Bald Eagle authors didn’t follow this stellar example. If they had, their Report could have
done some real good regarding election integrity — rather than be a deflecting diversion as to what actually
transpired.

In short, their assertion (Page ii): “In summary, multiple types of analysis found no evidence of fraud,
manipulation, or uncorrected error in the eight states included in this research” is misleading and
disingenuous, especially considering the MITRE Mission Statement, and the purported objective of this Report.

* * * * * * * * * * *

For any questions about this critique, suggestions for improvements, documented errors, please email
physicist John Droz, who was editor.

Please proceed to our more detailed commentaries of the major parts of the MITRE Report. If you don’t have
the time to read the details about each aspect of their Report, skip to our last chapter: Conclusions.

Page 5
 — Chapter 1 —
Critique of Report Section 1: Background
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

MITRE Summary:
The MITRE report team stated that the Bald Eagle research team analyzed a “wide range of relevant data” to
address “the anomaly allegations” in eight swing states. Its “research focused on the Republican Candidate,
President Donald Trump, and the Democratic Candidate, Joseph Biden, Jr.” and utilized “unfolding public
perceptions and new headlines” to determine its areas of focus. The team stated it analyzed data from all the
swing states however the report’s conclusion of “no fraud or compromise” is based only “on a handful of
anomalies” that are characterized as “five key areas.”

Critique:
The Bald Eagle team’s reliance on media headlines – that typically drive public perception — resulted in a
report of limited value because of its focus on five irregularities from the states of Georgia, Michigan, and
Pennsylvania. The irregularities in those states comprise only a small portion of the wide range of election
anomalies that surfaced after the 2020 election.

The Navarro Report Volume III is one of the more comprehensive documents summarizing alleged voter fraud
and it provides quantitative data for 26 different categories of voting irregularities/illegalities. It is notable
that one of the focuses of the MITRE report was ballot harvesting in Georgia that is not among the allegations
in the Navarro Report. In fact, the MITRE report cited no specific allegation of ballot harvesting in Georgia,
therefore one has to question the merits of including that specific analysis in the report.

Next, by focusing only on the election data related to Trump and Biden, the MITRE research ignored election
data related to Minor Party candidates, write-in candidates, and down-ballot candidates in state-wide
races. Other research conducted after the 2020 election pertaining to those categories of candidates
revealed significant anomalies in voting results that cannot be explained by chance and are strong indicators,
if not proof in some instances, of wide-spread vote manipulation.

Aside from the data source deficiencies noted above, the MITRE analysis did not utilize election time-series
data from Edison Research in its analysis of vote switching. The report references a tweet from former
President Donald Trump, but did not subsequently investigate the source of Trump’s tweet that would have
led them to the time-series data. The vast majority of data analysts who have alleged vote switching and
other anomalies have done so based on the Edison Research data. It is important to note that the Edison
Research data is public, and matches the certified election results in almost every battleground state. (Note:
Pennsylvania’s certified result was slightly different due to a correction made in Allegheny County.)

Although two of the Bald Eagle team members possess extensive experience in cyber-security, the Report did
not address repeated post-election allegations of hacking and/or vulnerabilities of the 2020 election system.
The Report team could have covered this topic in its analysis of the Antrim County, Michigan anomaly or in its
analysis of Dominion voting systems.

Page 6
The Report cited five legitimate data sources of election related data from the Pennsylvania, Michigan, and
Georgia state governments, however the vast majority of sources used in the Report are unreliable (biased)
and often irrelevant (e.g. from the media and government officials). For example, the sources used in the
analysis of Butler County, Pennsylvania included one legitimate government data source while the remainder
were irrelevant and questionable media reports that included a reddit.com bulletin board that contained
memes and member comments, an irrelevant media report alleging the problem was related to mail delivery,
and a misleading media report on the cause of the delay in mailing ballots.

While the Report states that analyses of voting/election data were performed in eight battleground states,
there are no summary tables, graphics, or narratives for four of the five key themes in the Report. For
example, summary graphics could have been included for the analysis for ballot harvesting, vote spikes, vote
switching, and delays in processing of mail-in ballots, for the eight battleground states. The exception to the
rule was the analysis of vote switching by Dominion systems that utilized results from all eight states.

In summary, the MITRE Report was narrowly focused on a few anomalies and just the major party candidates
in the 2020 election and therefore it relied on a very limited set of data and analyses.

Rewrite Suggestions:
The main suggestion is that the authors recognize and address the numerous allegations of voter fraud and,
necessarily, expand their sources to include Edison Research time series data, government and non-profit
sources of election related data (e.g., tracking of mail-in ballots, changes to election deadlines, etc.) and
numerous affidavits alleging voting irregularities with election systems used in the 2020 election.

Page 7
 — Chapter 2 —
Critique of Report Section 2.1:
Alleged Ballot Harvesting in Georgia
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

This is a brief review of the MITRE Report claim that:

no unusual mail-in-ballot manipulations were found in the Georgia vote count.

The MITRE team focused on Georgia mail-in-ballots (evidently as a representative test) to assess whether
ballot harvesting occurred. This assessment was made by plotting the logarithm of the requested ballots
versus the logarithm of the returned ballots for each of the 159 Georgia counties. These values will plot as a
straight line if there are no unusual addition or reduction of ballots. The following plot is Figure 1 from the
MITRE Report.

The county data generally falls on or near the blue trend line, which the MITRE team interprets to mean that
no unusual ballot harvesting took place. They note that if a consequential amount of county’s ballots had
been deleted, the data point for that county would fall well below the trend line. Conversely, if a sizeable
number of ballots had been added (harvested), the data point would place well above the trend line.

In general, the data along the trend line indicates that about 72% of the mailed ballots were returned. But
there are four outliers shown at the far upper right of the plot. These values are from the highly populated
counties of Fulton, DeKalb, Cobb and Gwinnett.
Page 8
A closer examination of these four counties shows that two had ballot returns near the expected average
return of all counties. But of the other two, one had a remarkably high ballot return of 82% while the second
had a relatively low ballot return of 67%. These two outliers represent significant deviations in the number of
expected ballots. Numerically, one represents 9,800± more ballots returned than expected. The other
represents 19,000± fewer returned ballots than the trend would predict. To put these amounts into
perspective, note that the Biden statewide margin of victory in Georgia was 11,800± votes.

These differences can be seen graphically by rotating the MITRE Report’s Figure 1 clockwise 45 degrees and
then flipping the y-axis top to the bottom. The four highly populated counties are outlined in red:

Fulton County especially stands out as having a large discrepancy between requested and returned ballots
when compared with other counties of similar size (note both axes are still on a log scale). Points here are
colored by the fraction of mail votes Biden won in the election. Contrary to the MITRE Report’s quick
conclusion, the same plot when viewed with increased resolution leaves open the clear possibility that
harvesting did in fact occur — and with sufficient magnitude to change the Georgia 2020 Presidential election
results.

To summarize, the MITRE team concludes that since county mail-ballot data fall on or near the trend line
based on simple visual inspection, the ballots were not mis-handled. A more detailed analysis shows there are
in fact outliers of sufficient magnitude to influence the election outcome. How those occurred justifies further
investigation.

Page 9
While reviewing the Georgia voter data posted on various sites, an additional unsettling concern arose
regarding the four most populous counties. 1,654,500± absentee ballots were mailed to voters in those four
counties. Of those receiving a mail-in-ballot, 922,000± voters chose to vote in person. There is no mention of
how those 732,000± purportedly unused mailed-out ballots were handled.

Another aspect noticed, was the rejection rate of mail-in-ballots in those four counties. Voter data shows
three counties rejected about 15% of the mail-in-ballots. Cobb County, however, rejected just 5%. For techies,
we created the following three tables to provide additional details for the preceding data analysis.

Trend Line equation, ln(y) = 1.0126*ln(x) - 0.4239

Ballot Ballot
Ballots Ballots
sent return % return
sent X Return Y
ln(X) ln(Y)
6 5.652 403 285 70.6
7 6.664 1,097 784 71.5
8 7.677 2,981 2,158 72.4
9 8.690 8,103 5,940 73.3
10 9.702 22,026 16,352 74.2
11 10.715 59,874 45,013 75.2
12 11.727 162,755 123,909 76.1

The table above shows the general trend of ballots send and returned as defined by MITRE. It is noted that
the MITRE Report doesn’t note how many returned ballots were superseded by the citizen voting in person.
Nor is there any mention of the 25%± of mailed ballots not returned.

The next table shows how accepted mail-in-ballots voted. The average return, as noted in the prior table, is
about 74%. By comparison, the Fulton County return is low at 67.5%, and the Cobb County return is a high
81.6%. This result doesn’t support the MITRE claim of negligible vote harvesting.

The table below shows the deviation from the expected ballot return for the four counties. Fulton County
shows 18,995 fewer ballots than expected, while the Cobb ballot return is 9,794 more than expected.

Page 10
 — Chapter 3 —
Critique of Report Section 2.2/A.1/A.2:
Election Fingerprinting
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

The Report cites and uses a method by Klimek et al. The method is conceptually simple. It assumes that the
vote count ratios across voting units, e.g. counties, districts, after suitable mathematical transformation follow
a standard normal distribution. See MITRE Report Figure 3 below.

The shape of the Georgia distribution (blue) is decidedly unlike the normal distribution (red). There are three
obvious shoulders/bumps indicating a four-component mixture. The Klimek methods assumes the data comes
from a single population, each geographic unit is made up of a similar mixture of people. Here we have a
mixture. Each shoulder/hump on the right of the Georgia distribution reflects a component of the mixture.

Klimek noted that voting in Canada (Figure 21 in Report, below), exhibited what looked like fraud, but then
noted that the voting was a mixture of English-speaking and French-speaking regions, so fraud was dismissed.
MITRE instead cited the picture as what a “fair election should look like” in their appendix – rather than the
oddity it represented to the original authors requiring further explanation.

Page 11
Now we apply Klimek, et al, regarding the MITRE graphs. Looking at the statewide fingerprint shown in Figure
4, we see two clusters: one is centered around 20% of votes for Biden, with 75% turnout; the other cluster is
spread along a line at 95% of votes for Biden, with a center of around 60% turnout. The Figure 4 from the
Report is shown below, circling the 95% cluster:

One could use the two groups to say, serious voters (75% turnout), favor Trump (80/20) and more indifferent
voters (60% turnout), favor Biden (95/5). Now a 95%-100% vote for any candidate can be taken as prima
facia evidence of election malfeasance, and is a keystone of the Klimek analysis. Jewish Americans vote
typically 80/20 Democrat/Republican. Black Americans vote typically 90/10 Dem/Rep. In 2020, the ratios are
said to have moved more to Trump’s favor: 70/30 and 85/15 respectively. The ratio 95/5 is a highly unusual
anomaly. The MITRE Report should have noted it.

Continuing the Klimek analysis: “We investigated the cluster with the high percentage of votes for Biden and
found that all of the precincts were in areas where conventional wisdom indicates there are large numbers of
Democrats, thus Biden’s percentage of the total vote would be high at any turnout level.”

To showcase this specifically using the same method as MITRE, the missing fingerprint from their analysis is
Fulton County shown below – where MITRE decided to exclude any figure and not report on any statistical
analysis. Does the circled mass in Fulton County prove that fraud occurred using the fingerprint method then?
As always, the devil is in the details. Those precincts, are South Fulton precincts, which demographically and
historically are overwhelmingly Democratic – and can be seen in 2020. A similar odd pattern was seen in 2016
(see figure below, on the right).

Page 12
Using the logic above, any level of vote irregularities in a Democratic geographic unit can be dismissed as the
expected voting pattern. The Klimek test is not definitive for the question at hand so the MITRE analysis
should be dismissed as unable to address the question or allegations.

Finally, the Report cited an analysis using a distribution of votes against the winner, divided by the registered
voters, then normalized logarithmically. One particular grievance against this equation is the denominator –
which is registered Georgia voters. Georgia passed the motor-voter law after 2016, so the denominator’s
magnitude is significantly inflated – and any references to regular turnout is false by definition. This is the first
real data point with such a registration pool. The turnout in Georgia was incredibly higher (in raw numbers of
voters) over 2016, even if the turnout as a percent of registered is unremarkable.

Moving forward, to quote the Report authors:

“A value 𝑣𝑖=1 occurs when exactly half of the registered voters in a precinct voted for the winner; since the
logarithm of 1 is 0 (𝐥𝐧(1)=0), this is the boundary between precincts where the winner of the election
received votes from more than half of all registered voters 𝐥𝐧(𝑣𝑖)<0, and the precincts where the winner
did not receive more than half of all registered voters 𝐥𝐧(𝑣𝑖)>0.”

Then the Report shows the four moments of distribution as follows, in their Table 3:

With the lack of labels, we must carefully study if this table is ln(v) or simply “v”. Careful reading suggests it is
ln(v). Therefore, as correctly shown by the graph, the mean and skew are in excess of ln(v) of 0, meaning that
both the mass of the tails and the distribution itself describe that the majority of all precincts voted against
Biden by MITRE’s own metrics. How then did Biden win the state? Such superficial analysis merely scratches
the surface of such a question, and provides no clear way to test or validate any statistical findings.

What should have been reported in Georgia

What can lead to a 95/5 ratio? A court-approved forensic examination of the specific Antrim County
(Michigan) situation found that the voting machine placed well over 50% of the votes into an adjudication file.
This file allocated votes (likely via an algorithm), without adequate bipartisan supervision.

The “hand recounts” that MITRE cites are alleged to be merely printouts of the post-adjudicated votes – not
the originals. The Antrim forensic report was dismissed out of hand by MITRE. Any machine/computer fraud
can scale, so it can be used anywhere — e.g. Georgia. To a skeptical researcher, the Antrim, Michigan results
should have pointed the way to request additional forensic examinations.

Many claims of fraud in Georgia were simply not covered by the MITRE Report: out of state voters, dead
voters, under-age voters, voters registered after the registration deadline, etc. The video of egregious, mid-
night unsupervised ballot counting (at Fulton County, Georgia) was not mentioned in the MITRE Report.

Page 13
Most egregiously, no analysis was provided on time-series data at all, which at the least begins to identify
places to look to validate or dismiss any of the anomalies that were identified. As an example, they could use
time series differential analysis (linear or logarithmic, both work) on public NYT Edison data (such as the
below) to help identify why a mean and skew ln(v) plot in Trump’s favor resulted in a Biden win. The answer
may possibly lie in data points well outside the histogram such as this, seen below:

------------------------

In short, the MITRE Report applied the Klimek method where it is obviously inappropriate — which we
confirmed by careful reading of Klimek, et al. Additionally, the logarithmic histograms are not easily parsed at
best, as well as suggesting the opposite of what is claimed. In general, the MITRE Report was superficial in that
it did not cover the many known problematic aspects of the 2020 presidential election.

Page 14
 — Chapter 4 —
Critique of Report Section 2.3:
Alleged Irregularities in Dominion Machines
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

The Report begins their counter to the multiple assertions against Dominion, by saying that “no evidence has
been provided to national media outlets”. It is useful to pause and consider that such a reply is egregiously
sophomoric, as the media is not the arbitrator for evidence-based decision-making (and note that Dominion
has been threatening large media outlets for even mentioning irregularities and evidence such as the Antrim
County [Michigan] report.) The authors of this section engage in a logical fallacy by using an appeal to
authority, while explicitly choosing to not investigate actual primary data sources of allegations – namely
thousands of affidavits; including eye-witness testimonies, confessions, videos, forensic audits, judicial
hearings, and court proceedings. Statistics are precursors to look for evidence, and are of little relevance when
appearing after actual court evidence is available from investigation.

Citing sources of allegations from “social media” and served up “media outlets” is by far the most outrageous
part of the MITRE Report. An academic journal would instantly reject this review with such willful ignorance
outright, as would an undergraduate professor in an entry level course. Yet the section will be given credence
because it satisfies a selection bias to feed a narrative. So be it, we will rebuke.

We respond first to the flawed premise the statistical analysis is based upon, and follow up with an
acknowledgement of the real source of allegations – actual evidence. In actual allegations from real primary
sources, the claims are not against just Dominion, but all computerized voting machines in the system. This
alone already undercuts the assumption from the analytical results presented.

Statistical Analysis
The MITRE analysis is fundamentally flawed because it centers around subtracting percentages that were
based on a different number of votes. This is an elementary fatal flaw which would be caught by any external
reviews of mathematical rigor. A second major flaw is the lack of citing omitted variables. Omitted variables
bias is a central issue in the social sciences. These controls are vital because areas that tend to vote for Biden
differ in many ways from those that vote for Trump and are less likely to have Dominion machines. The MITRE
Report authors do not even mention the presence of omitted variables or endogeneity, which implies that
their team either lacks the technical know-how or is intentionally deceiving readers

To be complete, we begin our analysis with the cross-sectional evidence and present two diagnostics. First,
using 3,108 counties, we regress the Biden to Trump vote margin on an indicator for whether the county has
Dominion, controlling for county demographics, including: population, median household income, the share
that are male, the age distribution (under age 5, age 5-9, 10-14, etc., 85+), the share that are White, the share
that Black, the share that are married, and the education distribution. This produces a coefficient of 2.68 (p-
value = 0.001). (Importantly, this same analysis, when failing to control for demographic differences produces
a coefficient of -3 (p-value = 0.169). MITRE omitted many important variables.)
Note: These p-values are integral for making statistical claims with any degree of confidence. For example,
a p-value under 0.05 effectively means that, if you were to resample the same specification 100 times, 95%
of the time you would obtain a result that is within the previously obtained number. Best practices in social
sciences are that p-values above 0.10 imply that the estimated coefficient is not statistically significant.

Page 15
Second, to examine whether our results are driven by states that are not in contest, we restrict our sample to
the set of battleground states (AZ, MI, WI, PA, GA, NV). We found an even higher coefficient on the Dominion
indicator of 6.08 (p-value = 0.007). Moreover, these results are qualitatively similar even if we include larger
counties more (e.g., weighting by population). This produces a coefficient of 3.92 (p-value = 0.092).

We then replicate the Report’s analysis as closely as we can based on their description. In particular, we
collect Democratic votes for 2016 and 2020 at the county-level. We subsequently regress the growth in these
votes between 2016 and 2020 on an indicator for whether the county in 2020 has Dominion, controlling for
the usual demographic characteristics. We find a coefficient of 0.025 (p-value = 0.037) on the Dominion
indicator, implying that counties that have Dominion experienced a 2.5 percentage point increase in
Democratic votes, relative to their counterparts. Moreover, that coefficient is obtained from regressions on
the restricted sample of battleground states. (However, if we do not control for demographics, then we obtain
a coefficient of -0.0019 (p-value = 0.86). No political scientist would ever ignore a demographics control or
similar confounding factors, so why did MITRE?)

In light of our results that there is an economically and statistically significant difference in votes for Biden for
the counties with Dominion machines, we now comment on a misleading presentation of the Report’s results
in Table 7. There, they calculate the overall average, rather than the change between 2016 and 2020, in votes
for a given county. That approach is not only inconsistent with their presentation of within-county changes in
their following plots, but also overlooks the fact that there is a great deal of composition effects across
counties that they are failing to control for when they average across all counties to produce the national
average. They also provide no discussion of the weighting approach. These are not mere details: we have just
shown that controlling for demographic characteristics is important and that failure to do so produces
statistically insignificant results.

We have far more analysis on this topic – but in the interest of brevity and addressing the actual allegations
via real evidence we move forward.

Acknowledgement of Evidence
This part of the rebuttal is, like the statistical section, cut down significantly for brevity. Expansion of the full
response is available, and all affidavits are available in public court dockets – of which there are thousands,
but we’ll keep it to the machines specifically for this particular section.

In the 2020 General Election, forensic, statistical and affidavit-based evidence demonstrate all three sources of
vulnerabilities were accessed by unknown actors, machine programming errors were introduced from sources
unknown, and network security vulnerabilities exploited by sources domestic and foreign. In short, there
were widespread problems of machine-altered results across multiple voting company’s platforms.

While citing Dominion to explain the basics of machine vote manipulations, it must be recognized that ES&S,
Hart, Smartmatic and others face similar problems and are equally compromised. They all share a common
GEMS–based software core that allows, among other things, for algorithms to distribute votes resulting in
fractionalized voting as opposed to simply counting whole votes. This opens the door to enormous capabilities
to shift votes using various programmed algorithms by either inside or outside threat actors. And the
Dominion logs from the already cited ASOG Antrim MI forensic investigation shows such an algorithm was
turned on.

Page 16
The adjudication rates found by the same Antrim investigation showed enormous rate of errors (68%)
generated by the machines, allegedly by “human error” but clearly intrinsic to the machine software. In
Georgia, the Secretary of State Brad Raffensperger was forced to admit during a Georgia Senate election
hearing on December 30, 2020 that of 137,134 ballots, 132,272 (96.45%) of them had to be adjudicated. At
the same hearing, Kathy Latham (a Republican County chair), testified that “election officials discovered that
ballots could be altered during the adjudication process.” Latham added that: “officials couldn’t see from their
end who had adjudicated which ballots”.

This confirms that the problems uncovered in Antrim (Michigan) also existed in Georgia, where massive
numbers of ballots were sent to adjudication from whence unknown operators voted them as they pleased.
As elsewhere all over the country, strenuous efforts have been made to keep truly independent teams from
doing forensics and paper ballot examinations as exemplified by Fulton Co., Ga. and Maricopa Co., Az. where
Legislative subpoenas have been issued to the counties to accomplish such and the counties have resorted to
the courts to stop them. The AZ subpoenas have roots in the countless eye-witness affidavits filed in the state.

In Texas, based on a tip from cyber experts who examined and were alarmed by the Dallas County central
server logs from the 2018 election, in 2020 cyber experts from Openrecords.org captured a computer hack
(they believe there were at least 3 hacks) in real time in the Dallas General Election. They did it by capturing
the daily downloads of the Daily Vote Roster posted by Dallas County that uses ES&S equipment, managed by
Barcelona-based Scytl and tied into Clarity Elections/Scytl election reporting network (Dominion, Smartmatic
and Hart are also tied into Clarity Elections/Scytl). The Open Records people had a simple methodology that
proved beyond any doubt that a massive computer hack(s) with vote tampering took place in Dallas.

Between October 6th and October 30th, Open Records saw the county actually purge 56,974 votes (absentee
and early voting in-person) after they were cast and then create 50,529 new votes using previously purged
state voter id numbers. In one case, 10 full blocks of a street in Highland Park had their votes purged and then
selectively replaced at random over the following days. Overall 5,690 votes with state voter id numbers were
purged and never re-appeared.

Cybersecurity evidence is filed in SCOTUS dockets on the vulnerabilities inherent in these machines, showing
that Edison Research used an unencrypted VPN and that their platform was accessed by foreign adversaries,
namely the Chinese Communist Party (CCP) and Iran. The statistical evidence is also corroborated by
whistleblowers, including Eric Coomer who unintentionally admitted on video rigging the election for Biden –
perhaps the catalyst for getting Dominion more attention above the other platforms.

This goes on and on, yet the MITRE authors cited “social media” as the source of machine voting allegations.
Perhaps this decision was reinforced by their cited media outlets claiming no evidence has been supplied due
to the lawsuits dismissed, yet we have data showing that is also a myth.

Ultimately, any efforts to use data analysis to improve the transparency or integrity of elections should at the
very basic level address and investigate the primary sources of the allegations – especially if they are rooted in
thousands of fellow countrymen risking perjury charges in courts of law. Such a source carries far more weight
than scanning tweets.

Page 17
 — Chapter 5 —
Critique of Report Section 2.4:
Voting Anomaly in Michigan
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

MITRE Report 2.4.1:

On the morning after the election, a SQUINT™ report described allegations that more than 100,000 votes had
been fraudulently dumped for Biden in Michigan in the middle of the night. The Bald Eagle team downloaded
preliminary results from the Michigan Secretary of State’s website and determined Antrim County was an
outlier. After investigating Antrim County in Michigan for data anomalies that created temporary unusual
increases in votes for Mr. Biden, the team found that the issues were caused by user error, which was
discovered and corrected quickly and did not have an impact on the official election results.
Comments:
The vote spike of 133,339 for Biden observed by social media users at 5:04 AM EST on November 4, 2020 was
not related to Antrim County, Michigan therefore the analysis related to Antrim County’s election tabulation
error is irrelevant (though it is considered relevant for other reasons: see below).
A team of experts that analyzed the Michigan election results could not find this alleged spike in the time-
series data reported by Edison Research. To be clear, social media and mainstream media reports of this vote
spike were not based on the actual election data. It is almost certain that Decision Desk made a transcription
error in its Twitter post and blamed the error on Michigan election officials. Michigan Secretary of State
Jocelyn Benson did not confirm the alleged error, or the alleged correction.
In summary, the MITRE team appears to have relied upon a media report from Politifact to support the
analysis that the vote spike existed, was the fault of Michigan election officials, and was subsequently fixed.
None of those points are supported by the evidence. (Based on our personal experience with Politifact, they
are a political agenda-promoting source disguised as a fact-checker.)
Then there was this major Michigan spike that the MITRE Report inexplicably did not address:

Page 18
That was a vote spike of 149,772 votes that occurred on November 4, 2020 at 6:30 AM EST. Biden received an
abnormally high 94% of the votes (141,258) while Trump received just 4% (5,968), for a net gain of 135,290
votes for Biden. This result is clearly anomalous and that particular vote spike was never corrected and
remained as part of the official election results.
The MITRE Report’s conclusion on page 25 that vote spikes didn’t impact the outcome is based on an event
that didn’t occur, but worse yet, ignores eight (8) other instances of vote spikes of over 100,000 votes in the
battleground states (see red items in following table, and go here for full report).

Although the MITRE team performed numerous analyses regarding voting patterns in Antrim County,
Michigan (and correctly found that there were not enough votes in the county to account for the fictional data
spike), the Report did not address the serious allegations of voting system errors found in a forensic
investigation. The forensic investigation found that 10,667 of 15,676 ballots (68%) could not be automatically
tabulated by the election system and required adjudication.
The investigators stated that ballots sent to adjudication “can be altered by administrators…with no audit trail
of which administrator actually adjudicates (i.e. votes) the ballot batch. This demonstrated a significant and
fatal error in security and election integrity because it provides no meaningful observation of the adjudication
process or audit trail of which administrator actually adjudicated the ballots.”
In summary, MITREs analysis of Michigan voting irregularities is misinformed and woefully incomplete as it
does not address the key issues. Worse yet, the Report’s conclusion that vote spikes didn’t impact the
outcome of the election was based on an event that didn’t happen.

Rewrite Suggestions
The main suggestion is that the authors investigate not only the eight other vote spikes of over 100,000 in the
battleground states, but also examine the multitude of anomalous vote spikes that occurred in non-
battleground states.

Page 19
 — Chapter 6 —
Critique of Report Section 2.5:
Butler County (PA) Lost Ballots
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

MITRE Summary: “…the mail-in ballot return rate for Butler County, PA was significantly lower than the other
counties in Pennsylvania. This can be partially explained by a computer glitch…The team conducted iterative
and daily analysis of the return rates, and although many requested mail-in ballots were lost in Butler County,
PA, the overall return rate of ballots leading up to Election Day fell in the expected range of all other counties
in Pennsylvania. No evidence of nefarious or fraudulent activities was identified.”

Rebuttal: The MITRE Report conclusions that there was no evidence of nefarious activities and that Butler
County fell within the range of all other Pennsylvania County is incorrect. The late ballot issue was caused by
human “error,” not a computer glitch, according to government officials. Butler County Commissioner Leslie
Osche stated the opposite (my emphasis added):
“…someone at the state level entered the wrong information into the SURE system last Thursday [October
8], changing the ballot status for voters across Pennsylvania. It tells you that your ballot was mailed back
in mid or early September before we even had a certified ballot…there was a challenge with the Green party
and until that challenge was resolved in the courts, we couldn’t even print ballots…the issue is widespread.”

In the Report, page 24, Butler County (orange dot) was identified as an outlier or in the first quartile (17
counties) in 7 of the 9 box and whisker plots. On Election Day, Butler County was among those counties in the
first quartile or otherwise at the lower end of counties that returned mail-in ballots.

The MITRE Report did not identify the other outlier counties, nor did it identify the counties that were in the
lower (first) quartile. A closer examination of the lower quartile and outliers (in yellow, below) reveals that all
of the outliers (except Bucks County and Montgomery County) are counties that Trump won easily in 2016.
Republican Counties also consistently dominate the lower quartile.
Pct. Pct. Pct. Pct. Pct. Pct. Pct.
16-Oct Return 23-Oct Return 28-Oct Return 30-Oct Return 31-Oct Return 1-Nov Return 3-Nov Return
WESTMORELAND 6.80% ERIE 48.40% WESTMORELAND 63.70% FRANKLIN 66.90% WESTMORELAND 73.04% LYCOMING 74.00% BUCKS 79.30%
LEHIGH 6.50% CENTRE 47.00% MONTGOMERY 63.40% LEHIGH 66.50% POTTER 72.67% MERCER 73.90% CLINTON 79.20%
MERCER 5.60% WESTMORELAND 45.90% HUNTINGDON 63.20% PHILADELPHIA 66.40% FRANKLIN 72.60% YORK 73.80% PHILADELPHIA 79.20%
CENTRE 2.90% FULTON 43.40% NORTHUMBERLAND 62.30% LEBANON 66.00% ERIE 72.30% BUCKS 73.50% WESTMORELAND 79.20%
BUTLER 2.80% YORK 39.90% FULTON 61.30% ARMSTRONG 65.90% CAMBRIA 71.81% CLINTON 73.20% YORK 79.00%
NORTHUMBERLAND 2.60% FAYETTE 37.20% LEBANON 60.80% COLUMBIA 65.90% WASHINGTON 71.12% ARMSTRONG 73.00% NORTHUMBERLAND 78.80%
CAMERON 2.20% LEHIGH 37.10% MONROE 58.40% PERRY 65.40% CUMBERLAND 70.96% BERKS 72.80% ARMSTRONG 78.70%
YORK 1.90% DELAWARE 36.90% PERRY 57.60% NORTHUMBERLAND 65.00% BLAIR 70.68% PERRY 72.60% LYCOMING 78.40%
FULTON 1.80% MONTGOMERY 36.00% ARMSTRONG 57.30% FULTON 64.10% CLARION 70.36% FULTON 72.40% BUTLER 77.70%
MONROE 1.80% MONROE 35.70% LUZERNE 55.00% LUZERNE 63.60% CAMERON 69.82% WESTMORELAND 72.10% FULTON 77.40%
PERRY 1.70% PERRY 35.40% LEHIGH 54.90% MONROE 62.90% BRADFORD 69.27% NORTHUMBERLAND 71.80% LUZERNE 76.60%
DELAWARE 1.50% ARMSTRONG 29.80% YORK 54.80% DELAWARE 62.10% MIFFLIN 68.82% LUZERNE 71.40% MERCER 76.50%
MONTGOMERY 1.00% LEBANON 29.60% SCHUYLKILL 53.80% YORK 60.50% DELAWARE 68.79% MONROE 70.80% MIFFLIN 76.20%
ARMSTRONG 0.70% LUZERNE 29.20% BUCKS 52.00% SCHUYLKILL 60.20% YORK 67.16% SCHUYLKILL 69.30% DELAWARE 73.10%
BUCKS 0.60% SCHUYLKILL 28.10% DELAWARE 51.50% BUCKS 59.70% SOMERSET 66.63% DELAWARE 67.30% BERKS 72.90%
LUZERNE 0.30% BUCKS 19.60% FAYETTE 49.60% FAYETTE 56.70% VENANGO 62.68% BUTLER 66.40% SCHUYLKILL 71.80%
CLARION 0.00% BUTLER 15.70% BUTLER 24.20% BUTLER 24.50% WARREN 46.57% FAYETTE 65.00% FAYETTE 71.70%

Page 20
Return rates by political party (by Election Day) show that Democrat return rates were approximately 10%
higher than Republicans. That translates to approximately 78,800 votes in a state that was decided by
80,555 votes.

While this evidence doesn’t prove something nefarious took place, it adds to other evidence and numerical
coincidences that raise serious questions about the integrity of mail-in voting in Pennsylvania.

For example, we compared Trump’s Election Day performance to his mail-in vote performance with his GOP
associates in statewide races. One would reasonably expect that GOP voters would vote in similar patterns in-
person as they would by mail. Interestingly, that was not the case – by 64,500± votes — to a not very well-
known Auditor General candidate, Timothy DeFoor.

GOP Cohort President Attorney General Auditor General Treasurer Margin

Election Day 2020 2,731,230 2,530,207 2,626,111 2,603,048 105,119
Mail In Votes 2020 595,538 574,052 660,030 636,780 -64,492

This pattern cannot be explained by chance. Our analysis of the time series data confirmed that 64,492 votes
were removed from Trump, laundered through the “Third Party,” and then eventually moved to Biden.

VOTE SWITCHING DETAILS

Trump to Trump to Biden to Adjudicated Third to Third to
Biden Third Third Biden to Trump Votes Biden Trump
Prior to 3AM -17,877 -2,141 -9,675 -2481 49,623 26,835 24,871
After 3AM -13,820 -62,352 -19,694 0 110,363 176,626 5,284
Total -31,697 -64,492 -29,369 -2,481 159,986 203,461 30,155

After performing computations of the puts and takes, Biden’s net gain from vote switching was 203,508 votes
and Trump’s net loss was 63,553 votes. Obviously, Biden received votes from other sources and those
sources most definitely included write-in votes. Interestingly, there was an inexplicably low number of write-
in votes cast in the 2020 election (7,672), and only 452 of them were by mail-in ballot.

Our review of the official election results found that absentee write-in votes were counted in just 9 of 67
counties -- Adams, Allegheny, Chester, Clearfield, Delaware, Lancaster, Lebanon, Montour, Northampton, and
Westmoreland counties. Philadelphia County is conspicuously absent from this list.

We agree with the MITRE Report that “…there were an unknown number of mail-in ballots lost.” This
admission of an “unknown number of “lost ballots” is a serious indictment in the way mail-in voting was
conducted in Pennsylvania and other states, with no chain of custody on the ballots. Tens of thousands of
ballots for Trump and other candidates could have easily been destroyed with no direct way of determining
that it ever occurred. In addition, the lack of meaningful observation of ballot adjudication also could have
resulted in “lost” votes – or more correctly – switched votes.

The plot on the next page provides statistical evidence that something along these lines happened in
Philadelphia County, Pennsylvania.

Each point in this plot is a Philadelphia County precinct with its size proportional to the total number of votes,
and its color scale by reported voter turnout. The y-axis is the fraction of votes reported for Biden and the x-
axis is the fraction of registered Democrats in that precinct. The circled points along the top are precincts
exhibiting a disproportionally high fraction for Biden compared to what is expected based on voter registration.

Page 21
A rough estimate for the number of votes
transferred to Biden from this analysis is
115,000. This was computed by fitting a
spline to only the main cluster of data below the
45-degree line, predicting what the Biden
fraction should have been in the suspicious
points in the upper and left part of the graph,
and then taking the difference in predicted
votes for Biden from what was actually
reported.

To connect this back to those from Chapter 2,

here is the same data shown as fingerprint
graph, changing the x-axis from fraction of
registered democrats to turnout fraction.

This is an actual example where the Klimek

analysis should apply. The numerous points
showing 95-100% indicate fraud.
Page 22
 — Chapter 7 —
Critique of Report Appendices B.1 & B.2:
Duplicate Votes & Elapsed Time to Send Ballots
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

Appendix B Analysis:
This analysis concerns: (1) double voting in Georgia, and (2) the elapsed time to send out mail-in ballots. We
find the basic analysis of double voting in Georgia accurate, although not well sourced. We also find the
minimal analysis of main-in ballot response plausible, although not sourced.

On the other hand, it is felt that the Report's lack of analysis of related issues suggested by the Report's data,
is disappointing. In particular, the Report's data suggests that further analysis of “advanced voting sites” (AVS
— also known as early in-person voting), is warranted. A detailed analysis of data obtained from the
Department of State and individual counties reveals a many to one ratio in the number of AVS were created
using substantial amounts of money from CTCL funded Mark Zuckerberg. Notably heavily Democratic Fulton
County had 39 AVS (7 outreach, 1 300 table mega, 29 normal and 2 mobile in large RVs) while most rural
Republican counties had one.

Appendix B.1 Coverage of Double Voting:

The first section of Appendix B concerns double voting as shown in Table 9 of the Report. The section's main
claim that “In the Georgia data, we detected a total of seventeen voters who appeared to have multiple
ballots accepted” does not seem to be referenced. Further, the sources that the Report does reference in this
Appendix include media commentary — not conclusions from independent parties doing a thorough and
objective investigation (e.g. #34 is a Time news story). While it would be reassuring to believe that this
extraordinarily low number is accurate, the results from other independent investigations — that did not rely
on election officials for data — makes it extremely unlikely to be even remotely accurate.

For example, consider the public sworn testimony of attorney Jesse Binnall about the independent
investigation done in Nevada, following the 2020 elections. He states (under oath):
“Over 42,000 people voted more than once. Our experts were able to make this determination by
reviewing the list of actual voters and comparing it to other voters with the same name, address,
and date of birth. This method was also able to catch people using different first name variations,
such as William and Bill, and individuals who were registered both under a married name and a
maiden name.”

Note 1: When a true forensic examination is done (as with Binnall’s team), they do not rely on state election
officials to give them data of voting irregularities. As stated above, they get data independently. It would seem
to be obvious that state election officials have a vested interest in claiming that there is no consequential
malfeasance, as otherwise it would directly reflect on them and their job performance.

Note 2: Nevada has 2.0± million registered voters, while Georgia has 7.6± million. If there were the same
percentage of people in Georgia that voted twice (as apparently did in Nevada), that would amount to
150,000± people who voted more than once in Georgia. Clearly, saying that there are only 17 strains credulity.

Note 3: The official Georgia election count says that Biden narrowly defeated Trump, by about 12,500 votes.

Note 4: That the Report spent 2+ pages on discussing 17 duplicate Georgia votes, is quite telling.
Page 23
Appendix B.2 Coverage of Time to Send Mail-In Ballots
Appendix B.2 is a hefty two paragraphs that analyzes the elapsed time to send out mail-in ballots by local
Georgia election bureaus. This section suffers from the same deficiency as the prior, as it also fails to provide a
documented source, or a method to duplicate the alleged results.

As we stated above, when a true objective examination is done, researchers do not rely on state election
officials to give them election process data — they get such data independently. It would seem to be intuitively
obvious that state election officials have a vested interest in claiming that they are responsive to voter
requests for absentee ballots, as otherwise it would directly reflect on them and their job performance.

With that said, it is pleasing to find that the claimed time to send out mail-in ballots was only roughly 1.9 days.
Of course, this short time (which clerks should strive to achieve), generally helps mail-in voters who broke
roughly 3 to 1 for President Biden (see here). On the other hand, we submit (aka conjecture without proof)
that no matter how soon tax bills (for example) are sent out, roughly 6% will be returned after the deadline
because procrastination is a common human trait.

Page 24
 — Chapter 8 —
Critique of Report Appendix C:
Benford’s Law and Elections
Ref MITRE Report # MP210086: Data Analytics to Enhance Election Transparency (February 2021)

We have decades of experience in the theory and application of Benford’s law, and don’t have any major
disagreement with the Report’s comments on Benford’s law application to elections.
Benford's law has been used to indicate the possible presence of fraud in many cases, and these indications
then lead to more detailed analysis being performed. Before using a test based on Benford's law, however,
one must first justify the assumption that the output should be Benford. There is extensive literature on why
first digit tests to Benford's law are not valid for election data, though second digit tests may be. We discuss
these and other issues below, as well as a new test, looking at the data in base 3 instead of base 10.
The advantage of switching to base 3 is that the data is spread over many more orders of magnitude (a factor
of 81 is not even two orders of magnitude base 10, but corresponds to 10000 base 3). We looked at some
statewide and some precinct level data from the 2020 presidential election. A preliminary analysis using base
3 showed good fits to Benford. A more detailed analysis, as well as theoretical analysis of both base 3 and
other metrics to replace the standard chi-square tests, are planned for Summer '21.
Benford’s Law is a mathematical result that describes many data sets and is often used to detect if a data set
has been modified. As there has been a lot of mixed discussion on the internet as to whether or not Benford’s
law is applicable to detect possible fraud in the 2020 election, we analyzed some of the data. The standard
base 10 test, as well as a new base 3 test of Benford's Law, do not appear to have the ability to identify
fraud in the 2020 election results for the states, counties, and precincts that we examined.
Benford’s law (originally conjectured by Newcomb) states there is a tendency in many data sets to have more
numbers with low rather than high leading digit (the leading digit of 2020 is 2, of .0341 is 3). Specifically

Page 25
Not all data sets should follow Benford’s law; for example, if most of the precincts have approximately the
same population and each candidate’s support is the same in each precinct, there will be a clustering of
leading digits. One solution is to look at second digits. Our new approach is to write the number of votes in
base 3 instead of base 10; the advantage of this is that our numbers are now spread out over several more
magnitudes (81 is a two-digit number in base 10, but in base 3 it is 100003, five digits, and this spreads out
clumped data).
Our goal was to look at a variety of statistics and the results of the two major candidate (Biden and Trump) in
several settings. This was done to see if it is reasonable to expect Benford’s law to hold, and if so if the data in
Pennsylvania follows Benford’s law. In the table below the number in parentheses by the candidate indicates
the base used for the comparison. Below the data by county (or whatever the grouping is called). The higher
the chi-square value, the further the observed distribution is from Benford’s law.
(---------------------------QUICK DIGRESSION ON CHI-SQUARE VALUES---------------------------)

(---------------------------END OF QUICK DIGRESSION ON CHI-SQUARE VALUES---------------------------)

Below are the results for several states; similar results were found at the county and precinct level.

Page 26
Our data was drawn from not just PA but also several battleground states, as well as a few uncontested states.
The data is, for the most part, consistent with Benford’s law. The first digits base 10 typically had values below
the 95% threshold. As expected, the fit to Benford was better when we shifted to base 3, as there the data
covers more orders of magnitude.
The analysis supports election data is consistent with Benford at this scale; however, if a party were to modify
only a few precincts, that would not be detectable by such analysis. Additionally, if a small fixed number of
votes were added across precincts, that also would not be detected (as it would almost surely not change the
first two digits).
There was only one place above the 95% threshold for both the base 10 and base 3 test: the Biden PA mail in
vote, but the two chi-square values are below the 99% cutoff threshold. The largest chi-square base 10 was
Trump in Kentucky, the largest base 3 was Trump in Illinois. We have similar data at the precinct level for
many of these states, indicating similar behavior (though the clustering effects are a bit stronger base in 0).
For example, below are data for Philadelphia. Note that before one uses Benford’s law one must prove that
Benford’s law is applicable. As remarked by many, there is good reason to believe data at the precinct level
should not follow Benford’s law. If most precincts are between 1000 and 2000 people and between 70% and
80% of the people vote and one candidate gets between 75% and 85% of the vote, then their vote totals range
from 525 to 1360, never having a first digit of 2, 3, or 4 (among other issues). To deal with such issues, people
often look at the second digits, or our new trick is to look at the totals base three (base 10 there is barely a
factor of 2 between the low and the high, while base 3 it is almost a full order of magnitude).

REFERENCES:
1. A. Berger and T. P. Hill: Benford Online Bibliography.
2. Chi-square calculator.
3. Chi-square test statistic.
4. A. E. Kossovsky: BENFORDS LAW: Theory, the General Law of Relative Quantities, and Forensic Fraud
Detection Applications. World Scientific Publishing Company, 2014.
5. J. C. McGinty, Can an Accounting Tool Detect Election Fraud? WSJ December 4, 2020.
6. S. J. Miller (editor): Theory and Applications of Benford’s Law Princeton University Press, 2015.
7. Wikipedia re Benford.

Page 27
 — Chapter 9 —
Conclusions about the MITRE Report
The MITRE Report is an incomplete, uninformed, and flawed analysis of the 2020 Presidential election. While
various experts identified and quantified nearly thirty categories of 2020 election irregularities (e.g. here), the
MITRE Report discussed only five. As such, it is clearly deficient. How could any serious analysis about election
malfeasance, not address that nearly 4 million votes were deleted from the system during the 2020 election
(see here and table to right)?

Our review of MITRE’s statistical analyses found some instances

of correct application, but in those cases the conclusions were
incorrect and/or the analysis was incomplete. The remaining
analyses by the MITRE team were either fundamentally flawed
or misapplied.

It is our opinion that the MITRE Report does not reflect a

serious attempt at evaluating 2020 voting integrity. The paper
lacks the basic necessary details, research, and citations
required for such a complex and important issue. There were
no apparent attempts for consulting experts in statistical fields.

Furthermore, there is not even an acknowledgement in the MITRE Report of the thousands of eye-witness,
forensic, and expert affidavits filed in court – or whether such evidence was even heard by a judge or
explained. Any “proof” of no consequential irregularities requires more than statistics. So far, the “fraud”
column has all the testimonies and forensics – and procedural dismissals don’t nullify the truth of evidence.

The MITRE effort seems non-genuine in the pursuit of what actually transpired in 2020, but rather appears to
be an effort to support the prevailing media narrative by providing “an opposing statistical study” that the
2020 election was legitimately conducted, with inconsequential irregularities.

We can expect more voting anomalies in 2022, because without meaningful reforms, this will happen again.
Without verifiable, transparent vote integrity, citizens become untethered from their government.

Brief summaries of each MITRE Report topic, and our commentaries, follow:
1. Ballot harvesting in Georgia. This topic was seemingly chosen out of thin air. Unlike other sections of the
MITRE Report, there are no references to legal cases, social media posts, and/or media coverage about
ballot harvest allegations in the Peach state. While the statistical technique was applied correctly, the
interpretation of the data was not entirely correct.

MITRE Analysis: The MITRE analysis determined the average for mail-in ballot return rates was 74% and
there were no anomalous counties.

Our Analysis: Our review of the same situation found that there were two outlier counties, one with a
68.8% return rate and the other with a rather high 81.3% return rate. Those values translate to 11,733
fewer returns from the former county and 13,371 more ballots from the latter. The margin of victory in
the Georgia 2020 presidential contest was only 12,670 votes, therefore the combination of the two
outliers could possibly have swung the Georgia election.
Page 28
2. Election Fingerprinting/Fraud Detection in Georgia. The MITRE Report cites and uses a method by
Klimek, et al, that assumes that the vote count ratios across voting units, e.g. counties, districts, after
suitable mathematical transformation follow a bell curve and homogeneous demographic. The Report
utilizes a related vote-cluster analysis (or fingerprinting) to determine if there were indications of fraud.

MITRE Analysis: The group applied a fingerprint of GA in whole, a second fingerprint using 6 selected
counties, and finally an analysis of the GA vote-distribution curve. MITRE found no evidence of fraud or
abnormalities.

Our Analysis: The MITRE group bypassed identifying or using the required assumptions of the Klimek
analysis in both clustering of groups as well as investigating repeated datapoints pushing 100% for a single
candidate. Our group followed the paper’s methodology, investigated the details as suggested, and found
that demographics explains some clustering. This renders the Georgia fingerprint graphic indeterminate
for detecting (or in MITRE’S case proving lack of-) fraud. Finally, our evaluation of the last technique found
that MITRE misinterpreted their own ln(v) analysis, which actually proved the majority of precincts in GA
voted against Biden. This latter point is rather well established – Biden “won” the 2020 election by
winning fewer counties (and therefore precincts) than any other candidate in recent history.

3. Dominion Voting Systems. The MITRE Report analyzed the presence of vote switching/manipulation by
comparing counties that used Dominion systems counties to those that used non-Dominion systems. The
primary hypothesis was that vote switches would be readily identified as spikes in heavily Democrat
counties.

MITRE Analysis: To determine the spike counties following the hypothesis, MITRE identified the counties
that had the greatest percent difference between 2016 and 2020 based on Democrat percentage of vote.
The formula subtracted Clinton’s 2016 percent of vote from Biden’s 2020 percent of vote. MITRE
performed a follow-on analysis of sixteen (16) “outlier” counties to determine if any of them fell outside
two standard deviations of the mean. MITRE determined that neither technique identified vote
manipulation by Dominion machines.

Our Analysis: MITRE incorrectly assumed Dominion was the only system being used to manipulate votes.
Legal filings after the 2020 election alleged all systems were used to manipulate votes, not just Dominion.
All of the voting systems used in the United States possess an adjudication function that can be utilized to
fraudulently reassign votes. Forensic and eye-witness testimonies are prolific and didn’t seem worthy of
mention?

Next, MITRE’s analysis had a flaw from the start as it is mathematically incorrect to subtract percentages
that are based on different populations – a peer review of such a calculation in a mathematical
community outside the group would instantly point out this error. Furthermore, no social science controls
for the dataset seem to be present at all.

Our analysis rebuts the MITRE conclusion as there is statistically and economically significant evidence that
counties that used Dominion machines had systematically higher margins for Biden—margins that easily
would have turned the results of the election, particularly given the swing states (and counties) that they
were concentrated in. The statistical evidence is corroborated by cybersecurity evidence, eyewitness
testimonies, forensic evidence, video evidence, and the thousands of affidavits filed under penalty of
perjury in a court of law. Such sources are far superior to MITRE’s source of allegations from “social
media”.

Page 29
4. Voting Anomaly in Michigan. The MITRE Report focused on a spike that advantaged Joe Biden by 138,339
votes in Michigan. This spike was alleged to have occurred at 5:04 AM EST on November 4, 2020 based on
a tweet by Decision Desk HQ. This section of the Report also performed a historical comparison of election
results in Antrim County, Michigan in response to reports of the flipping of votes in that county.

MITRE Analysis: The MITRE team’s conclusions were that both the Decision Desk HQ tweet and Antrim
vote flipping instances were caused by human error, corrected, and didn’t impact 2020 election results.

Our Analysis: Using official election results, there was not a spike in Michigan that advantaged Biden by
138,339 votes at 5:04 AM EST on November 4, 2020. It logically follows that there was not a correction to
that spike in the official vote total. This spike was not confirmed by any Michigan election official. It is
almost certain that this “spike” was the result of a typographical error by Decision Desk HQ.

Meantime, our team found a vote spike in Edison data that advantaged Biden by 135,290 votes that
occurred at 6:31 AM EST on November 4, 2020. This vote spike was never corrected and is nearly the
margin of difference in the 2020 Michigan election. In addition, our team found eight other (8) spikes of
over 100,000 votes in the key battleground states. None of those spikes were corrected — or logically
explained — and clearly impact the outcomes in those states.

The MITRE analysis of the Antrim County vote switching confirmed what everyone already understood –
that an error was found and corrected. However, the MITRE analysis ignored the very significant findings
from that very audit that determined nearly 68% of ballots were sent to adjudication and that fraudulent
reassignment of votes was the cause of the original flip of votes in Antrim County.

MITRE provided no evidence that such adjudication isn’t widespread. Without more well-chosen forensic
audits, no one really knows the extent of this disturbing, non-transparent manipulation of votes.

5. Missing Mail-in Ballots in Butler County, PA. The MITRE team analyzed data pertaining to the lag in
returning mail-in ballots and potential ballots lost in the mail from Butler County, PA. The analysis
determined that the late returns originally were the result of a computer glitch. The follow-on analysis
used a box and whisker technique to track the ballot return percentages in all 67 PA Counties.

MITRE Analysis: While the statistical analysis identified outliers during several of the iterations, the MITRE
team concluded there was no evidence of nefarious or fraudulent activities. The analysis relies on media
reports to conclude the low return rates in Butler County in mid-to-late October were caused by a
computer glitch.

Our Analysis: Election officials in Butler County stated that an individual had wrongly entered data into
Pennsylvania’s SURE system and that the ballots in question were not mailed out to voters – rendering the
assumption that the cause was a computer glitch provably false. The MITRE analysis of the rates of return
over time did not go into detail regarding the counties that had lagging rates of return.

Our analysis determined that approximately 13 of the 17 laggard counties were predominantly GOP
counties. Pennsylvania records show that Democrat return rates were 10 percentage points higher than
GOP rates – that translates to 77,800 votes or nearly the margin of victory in Pennsylvania. Our additional
analysis of the mail-in voting patterns in Pennsylvania found evidence that over 64,000 votes were
removed from Trump. In summary, not only was there sufficient evidence of probable nefarious activities
shown in MITRE’s box and whisker analysis of mail-in votes, but the evidence of mail-in voting irregularities
was everywhere our group looked.
Page 30
 — Appendix —

Some other 2020 election integrity reports, etc. our team of independent experts have put together are:
a) Pennsylvania Report,
b) Michigan Report,
c) Election Spikes Report,
d) Edison Timeseries Distribution Analysis,
e) List of 2020 Presidential-related Election Lawsuits,
f) Claudia Tenney Report (NY-22),
plus other less formal reports on Georgia, Wisconsin, etc.

Since the media hasn’t done a good job of publicizing material they consider “contrarian,” we often get asked:
what are some other quality 2020 election-related reports? Although we haven’t done exhaustive research on
other reports, here are some of our favorites:
1 - The Nevada Report by attorney Jesse Binnall.
2 - The Antrim Michigan Report done by Russ Ramsland.
3 - Three reports by Peter Navarro: Immaculate Deception, Art of the Steal and The Case, Evidence, etc.
4 - Three technical studies by Edward Solomon: Georgia, Pennsylvania, and Michigan.
5 - A study about absentee ballots by Dr. John Lott.
6 - Election Fraud and Foreign Interference report by Colonel Phil Waldron.
7 - Mathematician Bobby Piton’s Arizona and Pennsylvania analyses.
8 - A Judicial Watch Study that concluded that 353 US counties have more registered voters than people
eligible to vote.
9 - Interesting compendiums of 2020 election-related actions and claims: here, here, and here.
10-Lastly you might want to do searches over the Energy and Environmental Newsletter 2020 archives,
where over a hundred articles and reports about the election situation are referenced.

If you are aware any other quality 2020 election studies and reports, feel free to email us here.

Page 31