Basically a "you failed science 101" level of embarrassment mistake in my estimation. Testing two brand new signs with the different fonts should be a fairly obvious first idea. The hard/interesting part is taking into account learning effects for the previous font. There's a pretty interesting research question in how you'd age signs in there as well (as you want to know if the gains are positive over the entire lifetime of the sign) and some interesting ethics questions about field testing.
Meh. The American way would be "Build a street for font, and see which one people are willing to pay more for"
A/B testing is notoriously hard to do well, especially if you're comparing something the user already knows well with something different. Controlling for familiarity is pretty much impossible.
Only when someone points out to me that the signs they are replacing are probably not new would it become "obvious". (In fact, while I read my first article on this, my internal thought was 'so why isn't this better if it tested so well?', then they mentioned the old signs, and my thought was 'oh, right, that makes sense'.
So embarrassing mistake? No doubt. Unbelievably embarrassing mistake? No way, not with the sadly believable mistakes I've made in the past.