TL;DR : stub out your random source to avoid sporadic failures
I wrote a spec that ensured that a given input was randomized. It passed the vast majority of the time, but failed occasionally. This bothered me greatly after a day of having it randomly pop up, so I spent some time trying to fix it.
I tried increasing the amount of input, thinking that the odds that a combination of 25 items would be the same are very slim; it would still fail sporadically, but not nearly so often. I would get irritated less.
Of course, increasing the amount of input made no difference (other than maybe making the spec slower.)
Solution: stub out the random source to always return a given output. Now I can get the test to fail or pass consistently, as I like.
And yes, this sort of a “well, duh.”