Heads up on a new paper.

A. Use BM-25 to gather candidates. B. Use direct interaction (grep etc) on that smaller set.

This is deceptively simple, and the intuition is better aligned when it comes to recall then precision. With a little query expansion on the front end, there may be no need for embeddings.

Dr-DCI: Scaling Direct Corpus Interaction via Dynamic Workspace Expansion https://arxiv.org/abs/2606.14885

Thanks to JKB for bringing it to my attention.