The problem is that they did understand the state of affairs (or at least thought they did). It's very easy for us to look back at this with hindsight and see that they didn't understand the problem, but given the knowledge at the time, they understood the situation pretty well. The biological function of vitamin C that we now understand would have been an unknown unknown to them - they hadn't even discovered vitamins in the first place! What they did understand was that fruit keeps scurvy at bay, and citrus fruit is especially good at it. So exchanging one citrus fruit for another made sense according to Chesterton's fence - they understood why they were eating lemons in the first place and understandably assumed that limes would work in the same way.

As an aside, this also highlights one of the key flaws with Chesterton's fence, or at least its overuse: there is almost always _something_ that we don't understand about our changes. In the case of the limes, they understood that citrus fruit helped, but they didn't have the correct biological mechanisms by which it functioned. But this is true today - there are plenty of subjects where we understand lots of things about how something functions, but don't necessarily understand why it behaves that way. And that's just the cases where we know we don't understand something. There are surely many completely unexpected scientific discoveries to come. If we were to apply Chesterton's fence too liberally, we'd get very little done.

The same applies to more mundane cases. In terms of code, I'm generally a big fan of understanding why some feature was implemented before I start making changes to it, but I've had cases where I've tried to understand the original reasoning, but that reasoning later turned out to be wrong. Sometimes the best thing to do is simply to throw out all the old and embrace the unknown, accepting that it will probably go wrong in some ways, but also hoping that a new approach can be found.

So while I'm generally a fan of Chesterton's fence, I think it's important to recognise that it's just one approach in a toolbox of many valid but often contradictory approaches.