The problem that prompted me to make this extension to SQLite was a web-page request, so it needs to be processed with low latency. I initially tried loading the graph into memory and processing it that way. But there are approx 100K nodes in the graph, only a few dozen of which are relevant to the answer. It took a lot of time to pull in 100K nodes from disk. Running the query entirely in SQL is not only much less code to write, debug, and maintain, it is also much faster.