I think the problem was using filters and ranking on the graph. Those are influencing the outcomes.
Using a filter on an aggregated measure which uses _ForAll_ is not returning outcomes as I expect. I expect it to behave like the 'HAVING' clause in SQL but that is not the case because the _ForAll_ value is changed by applying the filter. The _ForAll_ value is based on the result set that is displayed in the graph.
With ranking it's very important to display 'All Other' when using an aggregated measure which uses _ForAll_. That way the numbers are correct when ranking the graph.