Which means you want to make do with as few shots as possible, while still providing a fair estimate … which again means you want a maximally representative sample from the underlying probability distribution. This, is why you want a sunflower – it has the most uniform distribution of points around a centre of symmetry possible, and you can scale the radial increments of each new point to match your desired aim accuracy distribution perfectly 
I’m shamelessly (self-)referring to this post, in a somewhat horrid thread.
2 Likes