I would say: composability. As you can see in the examples the datasources are very different, and for some one off observation this can be totally handy (if you don't need to massage the data too much, probably).