> But I’m not convinced why these graphs have to be generated by a single rule.

Wolfram claims that they don't need to be; on the contrary. Take another look at the last section of the article.