How this is computed: there's no dedicated per-package status-history table in the backend, so cycle time is package_order.createdAt (when assigned to the supplier) to package_order.modifiedAt (verified to match delivery time within ~1 second for the package that actually fulfilled the order). Reassignments between suppliers are sequential, not concurrent, so this is cleanly attributed per supplier. A ship-vs-deliver split was tried using checkout_order.inTransitAt and dropped — that field turned out to have bulk-backfilled timestamps clustered together across unrelated orders, not real shipment events, so it can't be trusted yet. Cancel % and reassigned-away are based on orders assigned in the window; cycle time is based on orders delivered in the window — these are different cohorts, so the counts won't match exactly.