Click or drag on the canvas to add points; move the pointer to aim the query box.
The science
A point-region quadtree recursively splits a square into 4 quadrants whenever it would hold more than c points. For n points in a reasonably uniform distribution the depth is about log4(n/c), insertion costs O(log n), and a range query returning k points costs about O(log n + k) square tests, against O(n) for a linear scan.
The page counts real work: tests = point-in-box checks + square-overlap checks done by the tree, versus n checks for brute force. The independent check compares the set returned by the tree with the set from a brute-force scan on every query.
Honest limits: heavily clustered or duplicate points make the tree deep (depth is capped at 12 here so identical points cannot recurse forever); for the benchmark the speedup falls as the query box covers more of the plane; tests are counted, not timed, so it is an algorithmic not a wall-clock measurement.