If there are measurements whose outcome is not used afterwards (typically, when transpiling a circuit with mid-circuit measurements but no feed-forward), then you can ignore them when considering the causal flow that gives the optimal measurement order.
The idea is that you can still get the optimal space when circuits have measurements.