I just posted this bug on connect.microsoft.com – there is an issue when you use unary operator in a parent-child dimension combined with a many-to-many relationship.

The problem is that the ‘+’ operator on non-leaf nodes seems to not consider the children value and bypass the intermediate measure group. Everything works well with the ‘-‘ operator. It sounds like an optimization that has a side-effect…

I found the bug thanks to a post on www.sqlbi.eu forum, that is starting to get feedback from people who are using the models described in the Many-to-Many Revolution paper.