Good article.

Of course, the point you omitted is how the actual probability of any given failure alters the calculations.

With the lottery example, because each of the numbers is (supposedly) equally likely to be chosen, Pascal's triangle is a perfect model, and shows the true (im)probability of winning.

With the car example, QA testing ensures that parts generally have a much higher probability of working than failing, thus although there's only one way for the whole system to work perfectly, it's more likely than the millions of ways it can fail.