Hybrid Search Relevance: Lexical Field Boosts Zeroed in Vector Mode

In vector hybrid mode, lexical field weight boosts are now automatically zeroed out, so semantic similarity drives the relevance ranking rather than competing with it.

01 Why the boosts had to go

Vector similarity scores sit in a 0 to 1 range. Lexical field boosts do not: even a small boost could inflate a document's score well above that range, so a keyword match on a boosted field outweighed everything the vector leg had to say. Semantically relevant results were pushed down the page by documents that merely contained the right word in the right field.

02 What still uses your field weights

Field weights are not discarded. When search falls back to pure lexical mode, they apply exactly as before, so your tuning still governs keyword ranking wherever vectors are not in play.