We made single-table search fast. But a search join (one query across tables, ranked by combined relevance) still meant denormalizing, or letting Postgres join and score millions of rows to return 20.
Now a 3-table search join runs 1000x faster, and still uses the query you'd naturally write.
Two weeks ago, we made Top K faster. This week, we made COUNT faster, and explored how real search workloads combine text with filters and sort orders.