Tip of the hat to the peerless Andrew Sullivan for posting this spectacular graph.
What you are looking at is overwhelming statistical evidence of election fraud. In any population sampling, the variance in the estimates should be greatest early on. In other words, at the beginning, the precision for the ratio of Ahmadinejad/Mousavi votes should be at its worst. As the votes are counted, the precision increases. Therefore, in any non-fraudulent election, a graph like this should show some spread either side of the line at the left side which eventually funnels down as more and more votes are counted. By contrast, here we are seeing an implausible correlation coefficient of 0.99 which could only have happened if the ratio of Ahmadinejad/Mousavi votes was fixed. Indeed, we would expect to see even more variability early on than what is just attributable to random effects. For example, rural and urban areas reported votes at different times which should, if anything, exacerbate the lack of precision early on in the count.