Thread by @rising_serpent

I cannot overstate the importance of this thread. It's technical but well worth reading and rereading to understand the gravity of what's being said.
Large orders of magnitude data tends to follow a normal distribution.
Observing anomalies in distribution gives crucial insight.

https://twitter.com/APhilosophae/status/1325592112428163072

1. To simplify it the best I can:
In person voting tends to not have a simplistic Democrat versus Republican linear distribution along a mean when you plot all votes. Because there are regional variations, families work together, as to friends with similar political persuasions.

2. On the other hand mailing ballots tend to have a fairly homogenous and almost linear distribution when plotted on a Democrat versus Republican X and Y axis distribution

3. This is because mail and ballots are essentially being shuffled like a deck of cards and a vote for Democrat versus that for a Republican essentially act s like a coin toss most of the time with occasional spikes in one direction or the other. However linearity is unmistakable

4. Hopefully this much is clear thus far.
Now imagine you have a large dump of mail-in ballots that coming later and are unshuffled and essentially unrandomized. That is when you start seeing weird spikes. These tend to unsurprisingly favor the Democrat candidate.

5. What you're using here is the basic principle of random chance to determine what is not random and can be measured by observing how data points behave when plotted over time.

6. The visualizations provided and based upon exquisite granularity data sets are the most sophisticated and most convincing evidence I have seen so far.

7. To oversimplify. First establish your data follows more or less random chance. Then whatever doesn't follow random chance is the anomalous data and points to changes in the way the data was being collected. Pinpoint what was happening at that time and you have your answer.