Had some more time to drill into this and we've ended up with a very similar approach to a lot of the metadata definition and resolution - I'd love to chat sometime about how you've solved some of the common problems (table selection w/ multiple sources, the constraint vs output projection, aggregation level, etc, etc)