You correctly understand that the AVL tree is determined by almost the same height of its edge nodes, but your confusion seems to be related to the difference between the node position and the edge weight.
That is: In the AVL tree, the depth of the edge nodes will be the same +/- (but not both!). This makes no claims regarding the cost associated with the edge between nodes. For an AVL tree with a root root and two children, the left path can be twice as expensive to go along the right path. This would make the tree weight-unbalanced, but still retain the definition of the AVL tree.
This page contains additional information: weight balanced tree - wikipedia
From Wikipedia:
The binary tree is called μ-balanced with
if for each node N the following inequality holds: 
and μ is minimal with this property. | N | is the number of nodes under the tree with N as root (including the root), and Nl is the left subtree of N.
Essentially, this means that children in the AVL tree are not necessarily evenly distributed across the lowest level of the tree. Taking N as an indication of the root of the node tree, you can build a valid AVL tree that has more children to the left of the root than to the right of it. With a very deep tree, there can be many nodes at this lower level.
Defining an AVL tree will require that they all be at one of the deepest points, but makes no guarantee that they are children of node N.
Fiarr source share