Been digging through nostrdb source code and finally grok why it's so fast:

No JSON parsing at query time. Events are stored in a binary format (like flatbuffers), memory-mapped via LMDB, and accessed with pointer arithmetic.

Most databases: parse JSON → allocate strings → copy data → query

nostrdb: mmap → pointer math → done

Zero-copy access. The event is just *there* in memory already.

Same trick strfry uses. Local-first Nostr apps benefit massively from this — Dave in notedeck queries local nostrdb instead of hitting relays, which is why it feels instant.