FTS v2 is turbopuffer's latest text search engine, boasting speeds up to 20 times faster than its predecessor, FTS v1. The key improvement comes from an upgraded search algorithm, particularly effective for the long queries often used by agents. In tests using a 5 million document dataset from Wikipedia, FTS v2 consistently outperformed FTS v1, showing significant latency reductions even for complex queries that include many terms or only stopwords. The article breaks down two algorithms—MAXSCORE and WAND—designed to optimize query evaluation. MAXSCORE sorts query terms by their potential score contributions and uses essential terms to filter candidate documents. If a document can’t meet the current minimum score, it gets skipped, improving speed. WAND, on the other hand, focuses on document-centric evaluations. It looks for the next document that could qualify based on the scores of all terms, potentially skipping more documents but at the cost of lower throughput due to increased computation per document. Both algorithms are enhanced by block-max variants, which allow them to skip entire blocks of documents based on local scores rather than evaluating each document individually. WAND is generally more effective in skipping irrelevant documents, but it can lag in performance due to its complexity. Apache Lucene’s experience highlighted these trade-offs, prompting a switch to MAXSCORE for better throughput on certain queries. The article emphasizes the evolution of these algorithms in the context of search engine efficiency and their practical implications.