index = pyflatbush.Flatbush(arr.shape[0], 100)
out = index.add_vectorized(arr[:, 0], arr[:, 1], arr[:, 2], arr[:, 3])
index.finish()

Then index._indexes stores the sorted order of all input shapes. So arr[index._indexes] would return all input boxes sorted by their hilbert value.

index._levelBounds gives the starting (?) index in index._boxes of each level of the tree. So when nodeSize is smaller, levelBounds is a larger array, and vice versa. The beginning of levelBounds is the smallest boxes; the end of levelBounds are the largest boxes.

So for example looking at the boxes from 3 to 4:

np.array(index_cy._boxes[index_cy._levelBounds[3]:index_cy._levelBounds[4]]).reshape(-1 , 4)

Looking at the boxes from 4 to 5:

np.array(index_cy._boxes[index_cy._levelBounds[4]:index_cy._levelBounds[5]]).reshape(-1 , 4)

https://cython.readthedocs.io/en/latest/src/userguide/fusedtypes.html