Shame on the charlatan who put this garbage together.

He poses a parameterized model of election results, and then seeks parameter values that reproduce the result.

He then tries to claim that the best fit parameter values indicate that 130% of Democrats voted for Biden and -30% of Democrats voted for Trump.

Since those values are outside the valid range, he overlays an interpretation of those values as the result of votes being "swapped" from Trump to Biden.

Models are always simplifications of reality. When using them, it is of critical importance to keep track of the simplifying assumptions made, and the distortions to be expected because of it.

The most glaring simplification embedded in the presented model is that is presumes that voting distributions by party will be the same in every precinct in the study.

When he tunes his model with an 80-20 Biden over Trump split from Democrats, he imposes that constraint in every precinct, rural or urban, rich or poor, of every racial make-up. No variations other than party registration can influence the model results.

My experience tells me that constraint is not borne out by the recorded reality in any election. Rural Democrats have a different statistical preference in the election than urban. White different from Black different from Hispanic.

The model will not even permit that variation to be represented. It is literally blind to even considering that possibility.

So later, when he tells you that the "only" possibility that matched the election results was one with the 130% vs -30% parameter values, he is not informing the audience of the vast universe of outcomes his model does not even consider.

Because his model is incomplete -- because the space of possible outcomes is not covered by his constrained viewpoint -- he cannot logically draw the conclusion that he does.

I've never head of this guy (which gives me great reason to doubt his headline claim of being the "inventor of email") but if he has the credentials claimed, he knows everything I just said.

He does leave himself one semi-graceful exit opportunity...

"One possibility there's a demographic within the Independent voters, not visible in our model, who voted for Biden"

So he confesses that his model has blind spots, and that they leave room for explanations of the vote result that do not require outrageous suggestions of tens of thousands of corruptions in the vote tabulations.

But he unforgivably undersells that possibility. It's not merely some esoteric subpopulation of the un-partied that could explain the results. Vast spaces of perfectly plausible normal results -- equally invisible to his model -- can explain it too.

I can only understand the failure to disclose that as an intentional attempt to deceive an audience with less analytical sophistication.

It gets worse.

Set aside the weakness of the model and the reasoning he builds upon it and focus just on the model parameters that are claimed to perfectly predict the election outcome.

If I can read his figures, the claim is that the election results, when precincts are arranged from smallest to largest, are nearly perfectly reproduced when Biden receives votes of 60% of Green, 36% of Independent, and 130% of Democrats, while...

Trump received votes of 60% of Libertarian, 58% of Independent, 100% of Republican, and (somehow) receives a deduction of 30% of the Democrat registration.

He waves his hands and calls this "vote swapping", but there's no swapping mechanism that produces that distribution.

In order to swap a vote from Trump to Biden, the vote for Trump has to be there in the first place. Within the constrained model, where is that?

Maybe it's clearer if we start from the closest parameter values that are valid. Imagine the vote from Democrat registered voters was 100% for Biden.

He doesn't present that case, but he presents a case close to it at 7:15 in the video. At that timestamp we have 100% Dem vote for Biden, but we also have GOP split 80-20 Trump over Biden.

Anyhow, the curve fit there is not as good, so we might seek changes to the parameters to make it better and notice that a shift in the Dem vote to 101% Biden and -1% Trump would fit better.

But what change in the ballot counting does that correspond to?

There isn't one. He wants you to think that ballots are moving from Trump to Biden, but you cannot bring ballots into existence by moving zero of them.

By the terms of his own model, he is claiming events that the model does not support. His interpretation is ill-founded.

To raise the sophistication just a step, consider for a moment what these "curves" are that are the target for parameter matching. They are cumulative percent when precincts are ordered smallest vote to largest vote.

At the far right, the curves end in the percentage results for the entire state. At the far left, the curves start with the percentage result for the smallest precinct.

There's not much in the way of features in the curves. Two point make a line, and that simple line would be a decent curve match.

If we look at the curves at 7:15 in the video (where we assume 100% Dem vote for Biden) notice they are close to correct on the right. They fail to match on the left.

So the bulk of the evidence of rejecting the valid assumption of 100% Dem vote for Biden and insisting on a need for a 130% Dem vote for Biden comes from the need to match the result in AZ's smallest precinct.

Eyeballing the graph, it looks like the outcome in that precinct was about 67% for Trump and 32% for Biden.

Eyeballing the model curves at 7:15 I see about a 64% vote for Trump and a 36% vote for Biden. Neglect the independents and minor parties to get
0.2 x GOP + 1.0 x DEM = .36 and
0.8 x GOP = .64

Solve and I get an estimate of two party registrations in this precinct as 80% GOP and 20% Dem.

This is probably wrong because it neglects Independents. But if the Independents in this precinct voted in proportions roughly equal to the final result, the distortions will be small.

I'd be interested to see the model output curves for the parameters of a GOP split 84 - 16 Trump over Biden and a 100% Dem vote for Biden. I have confidence it would better match the left side of the election outcome.

Further, I'd predict the match would not be meaningfully worse that the one produced by the 130 v -30 wishcasting.

In fact, it would be far better because it would not venture into parameter values outside the validity of the model.

That's not to say I think those figures represent what happened in the election. As I started saying, the model is far too crude to capture what actually happened.

But if we're going to pretend curve fits are compelling evidence, we have to at least examine both the fits that feed our fantasies and those that starve them.

The hearing moves to Q&A and it goes from misleading tall tales to absolute gobbledy-gook. He complains that vote totals are stored as "decimal" values rather than whole numbers, and then makes a fanciful claim that this implementation detail enables remote manipulation. (16:50)

Overall I think this appearance was misleading and shameful. That said, I think his final comments about the need for balloting and tabulation systems that are more transparently accountable is a fine thing.

If one positive thing can come from all these post-election shenanigans, I hope it is a demonstration of a need for balloting and tabulation systems that are not only accurate, but also transparently trustworthy to the broad population.