In reviewing the email thread about URIs and URLs, I think I dropped off because 1) the conversation veered into the topic of whether Fedora should be an LDP at all and 2) some of the tech speak was over my head and off putting (but it was a tech forum).  So, if we make this a discussion open to developers and metadata practitioners at the same time, could we try to 1) keep it well scoped and 2) try to be mindful of language barriers.  I went to a meeting of developers and metadata librarians once where from the start both sides agreed that anyone could raise their hand and say 'language' which would indicate we were using terms of art that others did not understand.  The meeting leaders made it less intimidating by raising their hands and yelling 'Language!" frequently at the start of the meeting.  It actually worked well to break down barriers and get people to think about how they were expressing their thoughts to a mixed audience.  Not saying that would work on a call, but you get the idea.

That aside, I stand by the accuracy of my original post.  With apologies to Doron, I would make a change to his introduction –

<Parent URL> <has child> <Child URL> should be expressed as <Parent URI> <has child> <Child URI>, followed by <Child URI> <has web location> <URL>. 

And further explanation (with apologies to all if I'm stating the obvious) -

URIs in RDF should be machine dereferenceable; they are intended for machine use, not human readability although they may redirect to a human readable web page.  An example is the Wikidata Concept URI for NLM https://www.wikidata.org/entity/Q611833, which redirects in a browser to a human readable web page (URL) about NLM https://www.wikidata.org/wiki/Q611833.  A machine only accesses the Concept URI to read and analyze Wikidata's RDF graph about NLM and to get the URL to display the web page and to gather the data displayed in the web page. In Fedora4, we were seeing developers equating the web page with the Thing, and there was also an issue about being able to use and maintain established RDF URIs from external RDF sources, like MeSH RDF or id.loc.gov or local triple stores.

Using the example above to drive home the URL/URI difference -

The Child Thing is the RDF Subject (center) of the Child graph.  The Child URL is the RDF Object of <has web location>, i.e., it is part of the Child Thing graph, but not the Child Thing itself.  The Child URL is accessible by referencing or querying the Child Thing graph.  There could conceivably be a graph around the Child URL as well - that graph would be about the web page itself (when it was created, size, when it was updated, encoding, etc.) and that would be a separate but related graph to the Child Thing graph.  Does that help with understanding?

RDF is a very precise and granular expression format intended for relating and re-using data, so making correct and precise RDF statements is important to creating accurate data and graphs for that purpose.  If you want to say something about a Thing, you must be able to unequivocally identify that Thing with a machine dereferenceable URI in order to create an accurate, usable, and re-usable graph about It.  The Thing URI facilitates the ability to use and re-use data in RDF graphs for the purpose of realizing granular relationships among different Things (possibility resulting in serendipitous/unforeseen findings), which is core to LD and the Semantic Web.  An LDP should support that.