But both objects have the same set of properties, which is why it’s not two “different” objects, it’s just that one document represents only a subset of the properties that one object has, which is completely valid.
e.g. the algorithm for processing an Update activity from a linked data perspective could be something like “iterate through the triples defined against the object in the document you’ve been sent, and then update those triples to point to their new value, deleting them if their value is null”. This doesn’t break linked data integrity in any way.