Start here. This is the direct spoken answer to practice first.
Overview
Cross-Region inference can improve throughput, but every possible destination must satisfy policy and be allowed by AWS controls.
I choose a Bedrock inference profile whose geography and destination Regions satisfy the workload's data policy. The runtime role is allowed to invoke that profile and the required model resources, while organization policies permit every destination the profile may use. I verify the cross-Region token quotas and record the inference profile and request identifiers on each trace.