If we don't want another J6, how about we give Americans CONFIDENCE in our electoral process?
We are the laughing stock of the WORLD right now. The Iraqis with their purple fingers have better election security than we do.
Do you realize, the very FIRST thing that happens with our elections results is they're given to a foreign company with foreign ownership? Scytl in Spain, look it up. There is no chain of custody after that, from there the numbers go to Edison Research and from there to the news media. Every news channel gets the same feed, they're all the same.
The central issue is CAUSALITY. We would like to know if the anomalies in the various voting curves are sufficient to be causal to the outcome. This is quantifiable, we can quantify "how" causal with how much confidence.
Taking apart datasets is not easy. One must be open minded, and relentlessly pursue the truth. In this particular case the dataset has been WELL studied.
To verify the integrity of the electoral PROCESS, the very first thing we can do is correlate the submission ticks with the registration information from each location. This way we generate additional data: fraction of registered voters voting and etc. But we don't stop there, we verify that the model we have is correct, by running a factor analysis on all the submissions. We curve fit the submission spectra to power series, that way we can extract regularity and long range interactions. We look for fractal structure, and evidence of criticality.
Then, returning to the original dataset with our newfound structural information, we start looking INSIDE each individual submission. Every marketing firm has software that shows data points against geography. (I have it, I use it every day).
The problem with our data is, we are not confident that the data is correct. No one can ascertain with confidence that the person submitting the ballot is the same person who registered. Therefore, we test two hypotheses: they are the same, or they aren't the same.
The way we actually test this, requires some cleverness on our part. Because if data points are created together, there are correlations between them - and these correlations can be made visible, with the right kind of analysis.
But here is the deeper point, moving forward - anyone who actually needs to perform forensic analysis, knows that the data has to be structured a certain way IN THE FIRST PLACE, to make it work, to make it possible. And ours, is not. Recounts are of no value whatsoever, the problem is in an entirely different place.
What we have to, is PARAMETRIZE the statistics, to make the underlying generators visible. If all data points are independent then the spectra look one way, and if they're not then they look different. The "batching" of results is not a problem, Wall Street does the exact same thing with stock tickers. That's why we use neural networks, cause there's no faster way of adjusting the mesh size.
Once you can parametrize the moments you're home free, them you can run Monte Carlo's to generate examples of representative outcomes, and you can VERIFY that the predicted 6-7 standard deviations actually occurs. Without this kind of analysis you can't tell, there's too much data and it's too confusing. With the proper analysis it becomes not only tractable but you can visualize it as well.
Okay, so, the statistics say that "one" event is not all that unlikely. It's "pretty likely" that if you have 50 experiments (50 states), something wierd will happen in at least one of them. However the NEXT question tells us a lot: simultaneity. If you have TWO events in two different states, how likely is it that they will be "simultaneous", or occur within a certain interval of each other? If ticks deviate from poisson statistics it tells us a lot.
Then ultimately, how likely is it that discontinuities will occur in FIVE states simultaneously, all of which happen to be important swing states?
Intuitively, if you have 5 pennies and you flip them all at the same time, how likely is it that they'll ALL end up heads? And if you keep flipping them at random intervals (which is like a stream of votes), how likely is it that you'll ever see 5 simultaneous heads?