I think one of the thing R demonstrates is how important the distinction between
"we want to make something good by making it structurally impossible to do something we think is bad"
and
"we want to make something good, so we'll tell you doing that is probably not a good idea but it's still possible. Whatever happens when you do that is sort of on you tho."
Allowing people to do (good) things that you didn't anticipate is literally the best possible outcome of making anything, and the balance between unstructured, groping in the dark chaos and the world of grammarlike intuition is the first, only, and total challenge. The evaluation model of R is a product of serendipity and reverse engineering that is objectively horrible to imagine, but has resulted in some of the most downright intuitive tools I have ever had the pleasure of using.