Hi Lilli, hope this helps with what I was saying, and what I feel Mark Lindeman was talking about.
The two simulations attempt to show that similar outputs of an election (focus on the scatter plot) can be achieved by either a hack or a different, more nuanced voter distribution. Since an election is a black box, without audits (ballots, software) or additional data (demographics, polling data), additional data is needed to determine statistical likelihood of the explanations.
By accident, these two examples also show what I was talking about with Russian tail being a potentially ineffective classifier. In the non-hack version there appears a distinct Russian tail, whereas with the hacked simulation, the top chart appears to fit a bell curve.
Uses two distinct hypothetical voting districts with different turn-out, ballots per tabulator, and blue/red choice.
Uses a single hypothetical district modeled by a (naive) normal distribution of votes, with a power function vote swap hack.