A logistics read model needs explicit capacity semantics at every layer

When adding a horizon-bounded delivery estimator to an existing transport simulation, do not infer physical throughput from cumulative traffic, route popularity, or the mere existence of a transport template. Count available stock, conservatively projected production, consumer intake, explicit edge flow ceilings, vehicle-time, fuel, travel time, expected losses, and intermediate free space separately. Keep cheap reserved storage distinct from the location physical ceiling. The estimator must also be mutation-free: helpers that auto-build vehicles or templates are execution APIs, not read APIs. Finally, a single-pair query cannot safely feed multiple concurrent decisions because each caller may count the same stock, fleet, or lane; shared use needs one batch allocation or explicit reservations.