An optimized compacted sparse merkle tree without pre-calculated hashes

Your scheme doesn’t seem to offer meaningful changes over the one discussed here.