It only fails in cases where you wouldn't notice if somebody else had noticed. In a school full of terrified children, each of whom incurs a huge risk in speaking up unilaterally / going to the media about the evil headmistress, it's easy to believe that no one would have said anything. If it happened today, in the real world, I'd check www.ratemyteachers.com, where the incentives to rat on the headmistress are totally different.

The dominating principle (pun totally intended) is:

P(you heard about someone noticing|it's true) = P(you would have heard someone noticed|someone noticed) * P(someone noticed|it's true)

From there you can subtract from one to find the probability that you haven't heard about anyone noticing given that it's true, and then use Bayes' Rule to find the chance that it's true, given that you haven't heard about anyone noticing...

...I think; I don't trust my brain with any math problem longer than two steps, and I probably wrote several of those probabilities wrong. But the point is, you can do math to it, and the higher the probability that someone would have noticed if it wasn't true, and the higher the probability that you would have heard about it if someone noticed, the higher the probability that, given you haven't heard of anyone noticing it's true, it's not true.

For you to justify the rule in this post, you'd have to prove that people either systematically overestimate the chance that they'd hear of it if someone noticed, or the probability that someone would notice it if it were true.