Lfs is frustrating. The clean/smudge model means files could be in the state of pointer or actual data. Lost count of how many times I didn't have everything smudged, or accidentally commited a 500mb model, or gotten into some wacked-out state where I couldn't clean, smudge, push, or pull without errors and had to nuke and start fresh from remote.

Data Version Control (dvc.org) is a far better model.