The Subject of some coherent data set is the (single) entity to which that data set relates/pertains. Examples of such sets include attributes, Claims/Assertions, files/dossiers, (verifiable) credentials, (partial) identities, etc.
Note that without this data set, the use of this term is meaningless. Therefore, subject must be seen as a relation (between a data set and an entity), rather than as this entity itself (as is commonly done).
Sometimes, the term 'subject' is also used in the meaning of 'reference to the (single) entity [etc.]' or 'identifier for the (single) entity [etc.]' rather than the (single) entity itself. We discourage such practices, and suggest to use the term 'subject identifier' in such cases.
The term subject provides a generic way to link entities that exist in the real (physical) world with knowledge about them (e.g. characteristics, partial identities and such) that exists in the minds of various %parties%.