Thomas T., I think you are raising exactly the right questions. Yes, LPGs appear (but I think only superficially so) ahead of RDF when it comes to programmatic interfaces and OGM, but that is a big part of what we are trying to do with FDR - fill that void. But also, backend APIs are still a perfectly valid and often desirable architectural choice. FDR + SPARQL endpoint is not the only scenario. In fact, the project using FDR I alluded to above does NOT have a SPARQL endpoint, but an API at the back.
However, I just want to make it clear that we don't have a complete OGM yet. Some ideas, but not ready to commit to an approach. You are right that there are some interesting attempts using SHACL. I have never build serious apps directly over SPARQL endpoints, only quick POCs. Mid-tier APIs are ok, they have a role to play, I just don't think they have to fully "own" the business model so to speak. I also don't think what one does to map RDF<->runtime object model has to follow the traditional ORM approach. As you realized, ontologies change and that's one of the advantages over RDBMs (the no schema, or flexible schema aspect), so we have to content with that. In general, ORMs allow a programmer to easily move from a relational representation (and mental model) to an object representation (and mental model). We need to find a way to do something analogous for RDF: move between the two representations & mental models. But the analogy stops there, the solution may not look exactly the same as an ORM.
