Bug: RRF weights in hybridQuery prioritize `lex expansions` over `original vector` in common path

[tada-zako]

Summary

hybridQuery currently assigns 2x RRF weight by list position (i < 2) rather than by query intent/type.
In the common path (when expansion includes lex), this gives 2x to:

• original FTS
• first lex FTS

instead of the intended:

• original FTS
• original vec

Description

Current Behavior

In src/store.ts (hybridQuery, Step 4):

const weights = rankedLists.map((_, i) => i < 2 ? 2.0 : 1.0);

Push order before Step 4 is typically:

1. initialFts (original FTS)
2. lex FTS results (from expansion, Step 3a)
3. original vec results (Step 3b)
4. expanded vec / hyde

If any lex expansion exists, it occupies index 1, receiving the 2.0x weight, while the original vector query (the primary semantic signal) is pushed to index 2+ and receives only 1.0x.

Expected Behavior

2x weight should apply to lists whose queryType === "original" (i.e., original FTS + original vec), and expansion-derived lists should stay at 1x.

Proposed Fix

Use rankedListMeta (already built in parallel with rankedLists) to assign semantic weights:

const weights = rankedListMeta.map(meta =>
  meta.queryType === "original" ? 2.0 : 1.0
);

Questions

1. Does this alignment with the original design intent make sense to the maintainers?
2. I have a local fix ready. Should I target the main branch or a specific development branch for the PR?