The example from the documentation does not work as expected. It creates wrong results:

query emb = [1 1 1 1 0]
match emb = [1.  1.2 1.  1.  0. ] score = 1.2806248474865694
match emb = [1.  2.2 2.  1.  0. ] score = 0.19999999999999787
match emb = [1.  0.1 0.  0.  0. ] score = 2.9342801502242417

The reason is, that if only_id is set to False (which is the default), it uses the same in-memory objects if a Document happens to be a match for two queries. Then scores are overwritten.