An airport authority must keep baggage X-ray assist inference available during ISP outages to the public cloud. Which infrastructure trade-off should lead the decision?

Preferring on-premises inference capacity for baggage X-ray assist when ISP reachability to public cloud is at risk correctly leads with availability. The trade-off works because operational assist at the airport lane must remain usable during cloud reachability loss; local inference capacity removes dependence on the ISP path for every assist decision. Cloud-only assumptions that "the public cloud never becomes unreachable from the airport" ignore the actual failure mode stated in the stem.

Preferring cloud-only inference because public cloud never experiences reachability loss fails because airports can lose ISP paths even when the cloud region itself is healthy—the failure is reachability, not solely cloud uptime. Disabling all AI assist and relying solely on paper logs during any network event fails conceptually as a primary design choice: it abandons the assist capability the authority wants to keep available rather than placing inference where outages do not kill it. Moving sealed customs corpora exclusively to unmanaged public object storage fails because it worsens control/residency and does not solve on-lane availability during ISP outages.

Exam caveat: on-prem assist still needs capacity planning for peak lanes and secure model updates; hybrid designs may keep cloud for non-critical analytics. Operational check: run ISP-failure drills proving X-ray assist continues on local inference, document RTO for assist, and reject cloud-only or "paper only" as the leading answer when continuous assist availability is required.