I think he simply means that, taking his first grid example to understand which 4 points would normally be neighbors on that grid, yet in some other arrangement, verify that there are no other points infringing on the boundary of every possible 4-neighboring-point shape (already explained well as the n, n+1, n+10, n+11 set). I do not think the shapes need to be convex.
For example, take points:
1,2,11,12 and verify that none of the remaining 96 points fall on the area defined by those 4 points.
2,3,12,13 and verify that none of the remaining 96 points fall on the area defined by those 4 points.
3,4,13,14 and verify that none of the remaining 96 points fall on the area defined by those 4 points.
etc...
Exclusions:
Points 10,11 are not neighbors.
Points 20,21 are not neighbors.
Points 30,31 are not neighbors.
etc...