Audit existing simulation substrates before creating a parallel vertical slice

A pure resolver can be correctly isolated yet still produce a poor integration architecture if the repository already has physical transport, observer-local learning, bounded controllers, reservations, and event provenance. Before adding new durable tables, inventory each existing subsystem by guarantees rather than names: conservation and custody, event timing, evidence locality, uncertainty representation, candidate selection, shared allocation, and level of detail. Keep the new pure kernel if it adds genuinely new simultaneous physics, but connect it with thin adapters. Arrival events should feed the new coordinate ledger exactly once; local measurements should feed the existing distribution learner; the existing controller should produce resolver requests through a spatial snapshot adapter. Separate prototype ledgers are useful as executable specifications, but they should not become second live writers. Documentation should call missing work adapters when the underlying capability already exists, not claim the whole capability is absent.