"Full" is a pretty coarse grained scheme in which no execution reads or writes to data in scope of a vertex v when the vertex program for v is running. This seems to give us the highest probability of correctness but could introduce false sharing. The other schemes are a bit finer grained.

What would be defined as the scope of v if we've specified any node attached to it by an edge in a separate section? Would "Full" just correspond to the entire graph in this case?