When performing constrained clustering, we enforced a cannot-link constraint between those two records so they would not end up in the same chunklet.