Adopt the same top-k collection algorithms as Lucene?

[jpountz]

Lucene recently worked on vectorizing its algorithms for evaluation of top-k hits, and it was quite successful, especially on hard queries such as queries that contain stop words or queries with many terms.

You can find a description of the algorithm for conjunctive queries at https://jpountz.github.io/2025/08/28/compiled-vs-vectorized-search-engine-edition.html and for disjunctive queries at https://jpountz.github.io/2025/10/11/vectorized-evaluation-of-disjunctive-queries.html.