Everything in this lesson chains from one fact: a model's context window is finite and its training data is frozen, so anything specific and current has to be handed to it fresh, every call, and only the relevant slice of it. Vector search is the industry-default mechanism for finding that slice. Aperture proves the mechanism is swappable — an LLM-as-router and a structured graph do the identical job with zero vectors involved. Know both, and you'll recognize "RAG" as a shape, not a specific stack, the next time someone hands you a requirement instead of an implementation.