Model emergent politics as a Rule kernel, not a coded subsystem

For a simulation with LLM agents, do not build bespoke diplomacy/war/loyalty/contract systems. Trade is the template: nobody enforces trade, it emerges from ledger + prices + processes + agents chasing gradients. Generalize — nearly every mechanic is one shape, Rule=(scope, condition->effect, trigger), run by two engines: a deterministic tick that FIRES satisfied conditions, and an agency loop that DRIVES actors toward unmet ones (a generalization of an auto-arbitrage router). Politics then emerges: an alliance is a cluster of mutual commitments that forms when cooperation scores positive under local resource/threat FIELDS (so a resource-rich region breeds denser politics than a poor one). Loyalty is not an attribute but a recomputed inequality trust > defection_temptation, so betrayal is pressure resolving rather than caprice; empires form by AGGREGATION of allegiance (same LOD roll-up as population/power). Crucial: encode this discipline in the world-builder AI prompt itself so it authors fields+drives+commitments and never decrees an alliance.