I don't think you grasped the statistical fallacy. Bolsonaro didn't gain "1 voter"; he gained the votes of an undetermined number of voters, and lost an (undetermined number - 1) of voters. The net change happens to be +1. This can be supported by a very wide range of plausible scenarios. (and here again we need to avoid the Texas Sharpshooter fallacy and actually integrate over the probabilities of *all* plausible scenarios) The posts you're citing ruled out every single plausible scenario except one: it calculated the single scenario where "no voter changes their mind; all first round voters also voted in the second round; all vote change is because of undecided voters". All I need to show why we can't *solely* count the probability for this scenario is to show you the support (in probability theory terms) for other scenarios is non-zero. That's it. I understand your broader point that *if* we want to think about whether the elections are legitimate or not from first principles, we need to apply Bayesian priors and updating. This unfortunately requires an immense of data to carry out in practice; none of the posts you're citing are doing real Bayesian analysis either. We're instead just doing a very traditional (frequentist) method of scientific debate where X says "This test rejects the null hypothesis!" and someone else says "Actually, one of your key assumptions is flawed, therefore the null hypothesis is not rejected by your work." On the projection thing: from the video alone I can't tell what method he's actually using for his simulations (which he doesn't even show). But his claim that all simulations show "multiple alternations" (and, according to his hand gesture, they can happen in the latter stages) sets off a very strong bullshit alarm. The probability of alternations in the sum of a random walk with drift actually decreases by power law in the short run and exponentially in the long run: If the drift is 2%, P(alternation at vote n) ∼= e^(-0.0002n) The last alternation of the 2026 Brazilian election was around n= 60 million where Lula overtook Bolsonaro. At n= 60 million the probability is e^(-12000) which is 10^(-5212) The "professor" is insisting that we should have many more alternations *after* this point. I don't know where he fucked up, but he fucked up something. (now of course in a more realistic model we'd have batches of votes coming in at a time, with some geographic correlation. But the professor's model isn't doing that either; I'm just reconstructing his model and showing why it can't be consistent with his claims. In any case, both of these assumptions actually decrease the expected number of alterations, although they do shift some of the support from early stages to mid stages) A final point on being a good Bayesian: there are 60 million very frustrated voters. We're seeing "tonnes of arguments" on social media (none that I've seen yet are good) from a large base of people who have been frantically looking for any corroborating evidence they can find. It's very important to remember what an insanely large number of potential accusers that is, so we don't over-update our priors in one direction. Our minds are naturally adapted for small-society lifestyles where "if 2 people make a strong accusation, we should probably believe them". That we're dealing with 60 million potential accusers here just isn't something our mind adapted for. Hence, we have to set up the counterfactual carefully: if election fraud *did* occur on a systematic level, it should leave a certain amount of evidence. What is the probability that *all* such evidence in this massive country can be hidden, that not an explicit or secret Bolsonaro sympathizer existed in the large group of people responsible for vote tabulation from these suspicious regions, which could have provided us first-hand evidence of cheating? Instead, the best evidence that 60 million people can present to us are, so far, several bad statistical models. That no better evidence exists should make us update strongly to the prior that the elections did not have systematic fraud. p.s.: Poliwave, which is from a team that I trust to be statistically competent, had a stable and accurate projection 30 minutes into the vote count.