Power BI would refuse to treat the non-unique column as a "one" side — or, if forced
through a workaround, offer many-to-many, the multiplying trap. The key check predicts the
dialog's behavior before you meet it: cardinality is a fact about your data, and you
can know it first.
The relationship is doing the merge's job — at query time, per visual,
without materializing a wider table. Lesson 3's merge baked Category into rows once;
the model looks it up live, every time, for every visual, and stays current when products
change. Same concept, better machinery — which is why modeling replaced merging the moment
we had more than one fact.