Search, Inspect, Fetch: Exploiting Boolean Retrieval for Deep-Research Agents
Sieve cuts deep-research context by fetching selected webpage sections instead of whole pages.
The paper introduces a search-inspect-fetch workflow using a Boolean Query Language to filter webpage fields such as titles, headings, sections, and metadata. Its result cards expose page structure before the agent chooses what to fetch. Across three QA collections, Sieve beat the strongest Search-Visit setup while using 20.7% to 50.6% fewer tokens. The authors say Boolean filtering improved every tested ranker, with the gains holding across retrievers and agent backbones. HF Daily Papers' note
The paper introduces a search-inspect-fetch workflow using a Boolean Query Language to filter webpage fields such as titles, headings, sections, and metadata. Its result cards expose page structure before the agent chooses what to fetch. Across three QA collections, Sieve beat the strongest Search-Visit setup while using 20.7% to 50.6% fewer tokens. The authors say Boolean filtering improved every tested ranker, with the gains holding across retrievers and agent backbones. HF Daily Papers' note
score 5