We added reranker selection work so result ordering can be handled after initial retrieval. The first retrieval pass finds candidates; reranking decides what deserves attention.
Keeping those steps separate is important. It lets the system inspect retrieval quality and evidence ordering independently before synthesis.
Keyword retrieval work gave the RAG layer a lexical path for exact terms, identifiers, and phrases.
Retrieval stitching work was added to join evidence from multiple retrieval paths without losing source boundaries.
We clarified the boundary between conversation memory, retrieved evidence, and answer synthesis in the Agentic RAG runtime.