The Living Index is the lifecycle that keeps a retrieval system tracking the reality it is supposed to represent, rather than drifting away from it. The framing matters: retrieval-augmented generation exists because the index is mutable, so a retrieval system is a living system, not a launch — and one nobody tends rots silently, with no error in the logs (stale index, embedding drift, permission drift, silent eval regression). The vector store and embedding model are the most commoditised, swappable and least durable parts; durability lives in the lifecycle, not the store.
It has three disciplines: Freshness (replace-on-change indexing, explicit deletion, change feeds, recency-aware ranking), Permissions (resolve identity → permitted documents → query-time filter, re-synced on change — but security trimming is not authorization, so back it with least privilege), and Evaluation (a standing IR + RAG regression gate — recall@k, nDCG, context precision/recall, faithfulness — that makes decay visible). Together they produce the outcome everyone wants: durable retrieval. The index is a mirror — when retrieval degrades it usually reflects a deeper problem upstream — which is why the Living Index is the retrieval link of the context-supply-chain and the see-side of the agent-gateway. The depth treatment is Retrieval Architecture That Doesn’t Rot.