144,867 checkpoint scans collapsed to 26,003 real actual-vs-planned transit legs.
SimPy queue model of outbound dispatch, vehicle pool as a shared resource.
Swept fleet size 2–24 vehicles — on-time rate plateaus early.
Swept the SLA buffer itself — found the route-type gap.
Across all 1,487 source hubs nationally, real delivery time runs well past the routing engine's naive plan — and FTL (long-haul) legs are, on average, slightly more predictable than Carting (local) legs.
Gurgaon hub, Medium-distance tier (77–299km, where both route types are genuinely used). SLA promise fixed at 2× the planned time. Fleet size swept from 2 to 24 outbound vehicles.
By ~8 vehicles, average dispatch wait time is already near zero — yet on-time rate caps at just 42–50% even at 24 vehicles. The real constraint is transit-time variance, not fleet capacity. Adding trucks would not fix this.
Fleet size fixed at 12 vehicles (congestion-free, per Test 1). SLA buffer swept instead, to find the promise window each policy can hit 90% of the time.
For Medium-tier corridors out of Gurgaon, an all-FTL policy reaches a 90% on-time rate at a 3.0×-of-plan SLA buffer — tighter than Carting or the historical 55/45 mix, both of which need 3.5×. Quoting FTL for this tier lets the hub promise a shorter delivery window at the same reliability target.