This is a combination of one or more tuple data elements that uniquely identify an entity. It is the relationship of non-Candidate Key elements to the Candidate Key(s) that form the basis of the first four Normal Forms.
Before 2NF, in particular, there will generally be multiple Candidate key(s), and they won’t identify the same entity. That’s what Normalisation will sort out. (Hopefully it’s quickly obvious that this can become a big, and interesting, part of the discussion of Normalisation and its application.)
Candidate Key(s) become the basis for selecting a Primary Key.