Select $t$ vertices from each block in index order. Integer Markov bounds control all pairs involving earlier samples and later blocks. Delete from each sample the vertices whose degree to another sample exceeds $t/(2E)$. The hypothesis $P\geq 128k^3E$ ensures that at least half of every sample remains, and the surviving noncomplete pairs are $1/E$-sparse in both directions.