Hi,
I just saw that ambiguous relationships are possible in Power BI. In the diagram below I would have expected that I get at least one inactive relationship or an error. But it actuallay works. Table 1 filters Table 3 on 2 paths. The resulting filter context for table 3 is the intersection of the filters that come through both relationsships. Is this new behaviour in Power BI or just new to me?
The most relevant insight for me was that if you turn one relationship into an inactive relationship and then you turn that on with USERELATIONSHIP, in this setup it does not turn off the ambiguous relationship so that you might miss some data in the reuslt. If you want to make sure that only one relationship applies at a time you need to turn both relationships going to table 3 into inactive relationships.
Yes, they are definitely possible, and the really dangerous thing is that they WON’T throw an error, but will return results. Sometimes those results will look ridiculous and you’ll know something is wrong, but often the results that are returned will look reasonable, despite being completely wrong. This is why (other than in the very limited, controlled case of the period slicer) I never use bi-directional relationships and always try to adhere to a star schema.
This is a terrific video explaining the ambiguity issue, followed by a recently updated version of their seminal 2018 article on this topic: