Resolves the cosine-similarity floor the semantic arm applies before it hands candidates to fusion. A unit scoring below the floor against the query embedding is never a semantic candidate, whatever the recall budget.

The floor is the per-bank config field recall_semantic_floor (default 0.3). It is model-dependent: BAAI/bge-small-en-v1.5 scores unrelated text at roughly 0.33–0.36 cosine and related text at 0.55–0.78, so on that model the default admits nearly the whole bank and leaves the reranker to do the work, while ~0.45 already sheds off-topic candidates. Measure on your own corpus before raising it: a floor above the model's related-text band empties the semantic arm.