Search Relevance: Why We Weight Titles Above Body Text

Barkpark's search runs on a Postgres tsvector index, and the single most impactful tuning we shipped was weighting a document's title above its body in the search_vector. A query for 'headless CMS' should surface the post literally titled that before it surfaces a paragraph that mentions it in passing. This post is a tour of our relevance pipeline: how we build the index, how A/B/C/D weight classes map onto title, excerpt, and body, why we added a 'did you mean' spelling suggestion, and how click feedback quietly teaches the engine that 'hadless' means 'headless'. It pairs with the search doc and the indx-engine retriever seam for anyone building search on top of structured content.