I read that page and I think he might assume that the distribution is approximately normal around the average. It's a fair approximation if the average is far from 1 or 0 but probably not so good for 1/16 (assumed T1 HQ rate) unless you have a lot of data.
I can tell you that, assuming 1/16 on T1, the odds of going 4/73 are (1/16)^4*(15/16)^69*73!/(4!*69!), or about 19.33%. A 95% confidence interval would range from 1.51% to 13.44% with the highest peak at 5.48% and a 47.5% probability of the true value occurring in each of the tails (each tail is even.)