I claim no particular expertise in statistical modeling, but I am trained in statistics, have taught undergraduate-level probabilistic modeling (computer science), and have a wickedly refined sense of the statistical nature things.
That said, if what the witness has testified is in fact the truth, I must agree with his assessment that the observation is strong indication of something very rotten in Denmark.
The larger a sample, the closer to the mean will be its character. Six hundred thousand is a very large sample. If the mean separation between the candidates in the original official ballot count was even as high as twenty percent, indicating an extraordinary landslide, there is basically no statistically valid chance that one would see so freakishly lopsided a batch tally as that reported. Simply put, it is statistically not possible to see so vast a variance from the mean in a sample representing nearly 10% of the whole.
The actual stated numbers lead to a non-credible proportion that runs wildly from the mean. 1 - (3200/604000) = 0.99947. That Biden held 99.947% over so large a block of counts is simply not credible in any way or degree. It is statistically impossible. Even if the batch were taken from inner-city über-ghetto parasite-infested Philly, so absurd a skew from the mean is not possible. It wouldn't even be possible in NYC, which almost certainly runs farther left and parasite-skewed.
A single such variance in tallying would be close enough to impossible, but there is something else that would have had to have happened that was not mentioned in the video. If Biden had truly won as per the official story, then there would
perforce have had to have been at least one equally anomalous instance of a batch as violently skewed in Trump's favor in order to put the final result at the thin final margin. A single such anomaly would make the overall probability at zero, which is to say there is no possible way that an "organic" process would yield two huge seismic waves of variance from the mean during counting. It is simply not possible, PERIOD.
Equally unlikely, even if it appear otherwise to an untrained observer, would be a larger number of less-skewed batch entries, all in Trump's favor, to drag the final result back to the percent or two final result difference. That, too, is vanishingly small in terms of possibility and when coupled with the single enormous variance once again constitutes a practical impossibility.
The actual numbers, taken from Wikipedia (I know, but it's 4 AM) are as follow:
Trump 3378263 votes, Biden 3459923. Biden won by just over one percent - nowhere nearly a landslide. Yet, in nearly nine percent of the total tally of 6838186 votes - the batch in question in the OP - he won by 99.947%. Ç'est impossible.
What we typically find in elections in terms of overall result v. time is an initial period with large spikes (proportionally speaking, with the absolute numbers being LOW) where the variance is large - this happens ALL THE TIME. But as more precincts report, the variances become proportionally ever smaller in relation to the total tally, which is why the curve normalizes with time, becoming smoother as the percent-reported grows. To have so violent a spike at any time during the counting demands equal compensation in the other direction, i.e. in Trump's favor. So if Trump lost by, say, one percent, he will have had to have done freakishly well in other quarters of the state in order to regain the devastating losses purportedly sustained in the precinct(s) that added up to 99.947%. This is pure fiction of the Twilight Zone variety.
Now, if we toss all those anomalous votes, which is the solution least injurious to the "dignity" and credibility of the state of Pennsylvania, the numbers come thusly:
Trump: 3375063, Biden: 2851923 -> Trump takes PA by between 16% and 19%, depending on how you calculate the ratio. That, my friends, is a landslide.
Anyone denying fraud in such a case would be prima facie either a liar, blitheringly ignorant of statistics, non-trivially brain-damaged, or some combination of all three.
Accepting the PA result, again assuming the witness speaks truthfully and accurately, would be analogous to believing a claim that they saw a 300 year old anvil in controlled flight from NYC to Los Angeles without the benefit of any aeronautical means whatsoever. Not possible under normal terrestrial conditions.
So either there was in fact massive fraud in PA, or the witness is a liar and should be imprisoned for life. My vote leans toward the former.