Build several identical clocks and watch the ensemble drift relative to each other.

The lattice clock in this story has numerous clusters of atoms. They could fill only a few clusters and measure the performance relative to a good clock, then statistically derive how much it would improve by using the full set of clusters.