The second post in the Beyond Retrieval series. Introduces the knowledge base manifest — a machine-readable specification that describes what a knowledge base is for, what inputs it requires, how to query it, and where its boundaries are. Shows how a meta-knowledge layer, paired with a router or planner, addresses the applicability failure modes identified in Part 1: ambiguity, implicit conditions, compositional errors, authority conflicts, granularity mismatch, and path convergence.