[xnull1guest]

First, note how the only numbers addressed are the '2,776 instances publicly known'. So from the start we're working with unreasonable numbers. But let's run with it.

Hmm. If the average number of mistakes an analyst makes is 0.1 per year and there were 700 mistakes only, this means 7000 analysts (or do we need to model this as a poisson distribution?). Is 7,000 analysts reasonable? Anyone have more details on this?

The NSA has said that it performs about 20 million queries a month, or 240 million queries a year. If these are done by analysts that's 16 manual queries an hour or 130 a day assuming a standard work week. That seems reasonable. Or at least reasonable"ish". [240,000,000 / 7,000 / (5/7 * 8 * 365)]

But it would also imply an error rate of 700/24,000,000 = 0.000002917 (which is absurd, if the error are presumably due to 'typos').