Surprisingly I think it’s what you’d get if you trained a (simple) classical ML model on a world map coloured using the usual western countries bs the rest colouring. It chose a pretty simple boundary.