> we're currently investigating what a longer-lived cache might look like, but it comes with its own concerns
FWIW a while ago I sent some papers about caching in sparql, which is pretty much the exact same problem (how to cache the result of a graph walk in a triple store and invalidate it when the graph changes)