Recent regressions have highlighted the importance of being able to programmatically test the expected output of our DB queries. To do that, we could use something like tmp-postgres to spin up a sandboxed Postgres instance during tests, load there the schema.sql, populate the DB with some mock corpus and test that our queries return what expected.

In the same vein, we could later use something like tasty-bench to add benchmarks, paving the way for fine turning of these requests.