I thought about remarking on documentation, but I wanted to keep the comment brief.
Even where we have documentation, what experience buys you — what Alice in the example has — is that she's going to the documentation to spur her memory, as a quick double check, and just to give her a good guide rail to make sure she hasn't missed anything. But because she knows the area, she can still be quick about it. Someone new to the system are stuck using the docs not for reference, but for tutorial and basic understanding. They're going to need to read every word, and worse, comprehend every word, and that will take far, far more effort than the "I'm a bit foggy on detail D, what did the docs say? Oh right." that Alice would be doing.
But let's accept your "Maybe that the system is badly implentmented" — quite possibly! But it's the experience of having worked with it, learned how it modeled its problem, and learned how its model falls down and then, finally, that insight into "this would be a better way". That takes time and understanding, if you want to build a next system, particularly one that avoids a second-system effect.
Oftentimes, the questions or problems are of the form "we have tool X, and it does job J, but we have this other job, J', and it's … sort of like J? Can tool X do it?" Now tool X has never encountered this particular problem — by definition it cannot be documented — and it might be we need to change tool X. Again, experience makes this take a lot less time. Even for a well designed system, it still requires someone new to it to dig in and figure out first how it ticks before they can even approach such a question.
The reality is, of course, far less rosy, and there are many places where the docs could be better, often a lot better. But we've lost the most qualified person to improve them!