If a sim test needs hand-wired roads/knowledge, that IS the missing feature

When writing a long-horizon simulation harness, treat every precondition you must hand-wire as a signal of a missing autonomous mechanism, not as setup boilerplate. Concretely: trade required a pre-drawn road graph, pre-seeded fog-of-war beliefs, and a hardcoded currency allowlist. The fix was to make each emerge: geography (containment = adjacency) gives baseline trails and TRAFFIC upgrades trails into roads (a desire-path feedback loop: better roads lower the risk premium -> more trade); agents learn by PRESENCE (the node they occupy plus one edge out) so no belief seeding; and trade reads the world own currency resources instead of a fixed list. Second lesson: a 1000-year run surfaces bugs unit tests never will -- a scale=org trade_route entity got recruited as a political actor (a literal Coalition of a grain route), and bloc names recursed. Run the emergent system at length and watch aggregates for nonsense; it is a distinct test modality from asserting one tick.