I imagine it could be done similarly to how we do federation. Though I’m not clear on how you determine the subtype, is it encoded in the ID? The spec seems somewhat vague on the point so I’m guessing we can’t assume much about the ID structure, which makes the design somewhat tricky.

Though I’m not clear on how you determine the subtype, is it encoded in the ID?

Ah, yes. That's an important detail I forgot to mention! In our code, we have typed all of our IDs. In the schema, they are simply ID, but in our backend code they become e.g. ID[Foo] where we have a structure/mechanism for expressing this typed ID.

How do you determine the type of the id though? Is the type prefixed to the id Foo_<id>? or is it like a packed byte representing the type? In federation we always get something in an object format {__typename: "Foo", id: ""} so we can do a lookup for the resolver of that type.

Took a stab at an implementation here if you could take a look. I've never used the spec myself so I'm not sure what the best DX would be here, so I just followed what we did for federation.

Sorry for the late reply!

How do you determine the type of the id though?

Sorry, I should have written about that as well. We've made a structure for this that constructs ID's like this: subgraphname:entitynr:primarykeys where our service defines the subgraphname and we manually assign entity numbers for each entity and the primarykeys are a json representation of the primary key in our database.

Is the type prefixed to the id Foo_<id>? or is it like a packed byte representing the type?

It's prefixed.

Took a stab at an implementation here if you could take a look. I've never used the spec myself so I'm not sure what the best DX would be here, so I just followed what we did for federation.

Thanks! I'll take a look and give you some feedback as soon as possible!