Repository navigation
Optimize flat scan by pre-filtering low-similarity results #1001
Description
Some clients currently filter out low-similarity results during post-processing.
We should evaluate whether applying this filter earlier—before inserting candidates into the priority queue—provides measurable performance benefits.
If it does, we should add support for this optimization to the flat scan algorithm.
Metadata
Metadata
Assignees
Labels
Type
Projects
- Status
Activity
harsha-simhadri commentedon May 1, 2026
We do have range search in the graph index. Could you be more specific about the request
[-]Support for range search (search within a distance radius)[/-][+]Flat scan should support filtering out low-similarity results[/+][-]Flat scan should support filtering out low-similarity results[/-][+]Optimize flat scan by pre-filtering low-similarity results[/+]arrayka commentedon May 5, 2026
Done. I've added details to the description.
harsha-simhadri commentedon May 5, 2026
Alex, this is still confusing. Initially this issue was about range search. Now range search does not figure in the title/description.
Further, are you asking about fallback to point looks for very selective filters or that we change the graph walk to aid in better accuracy. They are different and complementary.