Dispersion is a measure of the variation between replicates is an MSA ensemble.
On benchmark tests, dispersion correlates strongly with error (fraction of incorrect columns).
The distance between two replicates is calculated as the fraction of columns which differ, and the dispersion (D) of an ensemble is the mean distance. When sequences have few differences, their MSA is robust against parameter variations and replicates are identical (D = 0). In more challenging cases, replicates vary (D > 0) due to alignment errors.
