I'm the project lead for VectorChord. I have tested ScaNN on AlloyDB Omni but have struggled to achieve reasonable recall on the GIST 1M dataset, with results p

I'm the project lead for VectorChord. I have tested ScaNN on AlloyDB Omni but have struggled to achieve reasonable recall on the GIST 1M dataset, with results peaking at only around 0.8. The limited documentation makes it challenging to understand the underlying causes of this performance. Additionally, I couldn’t find any performance benchmarks for ScaNN integrated with PostgreSQL, particularly in comparison to pgvector or its standalone. The publicly available metrics focus exclusively on query-only indexing outside of the database. On our side, we’ve implemented the fastscan kernel for bit vector scanning, which is considered as one of ScaNN’s key advantages.