Substrate as a common language for query plans appears in a few query engines. It may be pragmatic to provide support for substrate plans, I think this would be close to the boundary between the python-heavy planning phase and the cython-heavy execution phase, although I expect some optimization of the substrate plan would still be needed (e.g. based on cost).