This caches various queries to ensure that multiple reference extraction
runs re-use any objects queried in previous runs.