As said in the README:

(...) Each distinct point exists in a unique leaf node; coincident points are represented by a linked list. (...)

So this is a quadtree that ALWAYS subdivide when adding a new point, except if this is a point(datum) exactly the same as another that already exists?

I'm very curious about this design choice of not allowing to have custom criteria on when should subdivide leaf nodes (ex: above 10 points), it's just that I have this intuition that it would alleviate so much memory and time entering so many nodes