Halliday (2008: 121):
Agnation means that the outputs of different passes through the same network will be related; and the distance between them can be measured, by the number and location of the features that they share — how much their “selection expressions” have in common.