When a prompt enters the search command center, it decomposes into parallel subqueries to retrieve candidate chunks across indexes. Next, the reranker and hallucination gates score each context chunk against your citation strictness threshold, discarding ungrounded claims. Adjusting fanout and strictness updates latency and grounding live, verifying citation sources before synthesis to protect responses from hallucination risk.