pull is faster because fewer refs means the discovery process is smaller