Comment by orph

Why not apply changes to the underlying model so that you crush every available eval?