The SLLN says that the average of a sample of a population becomes an increasingly good predictor of the true average of the population as the sample size increases, eventually reaching the true mean.

If we assume that the voting population of NH Dems is homogenous, regardless of voting method then the proportion of Obama voters should stay the same across hand counted and machine counted ballots, or that is what you would expect in a fair contest.

So the fact that the proportions exactly flip flopped is suspicious.

From the programming perspective it should be noted that flipping the candidate names is the easiest and least traceable method of stealing the election in the machine. It could be as easy doing a table or alteration.