In general, once you understand some set of rules well enough to know why they exist and what benefit each yields, you know when and how to break them to get a better result than the rules would give.
A small example, perhaps relevant to your point about the individualised iOS choices (about which I know nothing) made by Google et al.:
after Microsoft's round-vs-square convention for exclusive vs inclusive choices became the default standard, Borland deviated by making their "round" ones lozenge (playing card "diamond") shaped, among other differences that I don't now remember (maybe they indicated latching by indenting or similar).
They could do that because their HF people understood that roundness wasn't important, what was important was an immediately-obvious difference to the widget for making inclusive choices, simplicity and regularity (for easy learning), and visible latching. Those were the factors that mattered, not the exact shape or behavior.