The example is fine, you are calling out confounding variables. Just call it confounders, instead of Simpson paradox.

FWIW, the arithmetic in that example also has a glitch. For the "Mobile" "Control" case, 100/3000 is about 3% rather than 10%.