Translation (via http://www.deepl.com/Translator http://www.deepl.com/Translator): > Well, here's the picture. Data from 96840 polling stations (107.9 million

Translation (via http://www.deepl.com/Translator http://www.deepl.com/Translator): > Well, here's the picture. Data from 96840 polling stations (107.9 million registered voters out of 109.2 according to the CEC). All of Russia > On the left is the histogram of the votes for the parties in the 1% turnout intervals, the thin line is the histogram of the votes for all the parties except for ER, scaled to fit the histogram for ER in the initial polling station; assuming (strictly speaking incorrectly) that all the fraud for ER is pure stuffing, the shaded area gives the size of the stuffing. > To the right are party results as a function of turnout by precinct, with each precinct a set of different coloured dots for the respective parties. The dense core is presumably non-fraudulent precincts; the centre of the core for ER is determined by the Minimum Covariance Determinant method with weighting of precincts by the number of voters and a cut-off threshold of 50% (the points in the cut-off are bounded by an ellipse). Accordingly, the kernel centre coordinates for ER give an estimate of the real turnout and the real result of ER. > In non-fraudulent precincts, turnout averages 38% (as in 2016), the ER gets 31% to 33% (as in 2011) by the two estimates. What is separately remarkable is that the peaks in the distribution of votes in the integer percentages have decreased (but remain well above the threshold of significance, as D. Kobak has calculated). > "Tooth" on the 95% turnout is the Moscow e-voting.