County IT wants to prototype a new EVPN tenant design for Spectrum-X without risking live switches. Where should they identify issues first?

County IT prototyping a new EVPN tenant design for Spectrum-X should find issues before any live switch is at risk. They should identify issues first in an NVIDIA Air simulation of the BGP-EVPN design before production cutover. Air acts as the pre-production digital twin where tenant VRFs, route targets, multihoming, and underlay connectivity can be exercised without impacting county production AI Ethernet. Catching RT typos, missing VTEP config, or broken isolation in Air is far cheaper than discovering them on live leafs. With NVIDIA NetQ telemetry review of the live fabric following tenant activation fails by concept: it waits until after the tenant is already live to spot problems, exposing production AI Ethernet to avoidable issues. In a Cumulus VX virtual lab using generic VXLAN configs for Spectrum-X fails because generic VXLAN configs do not emulate the Spectrum-X underlay or NVIDIA-specific BGP-EVPN behavior, so tenant-specific issues can still reach production. Through an NVUE config audit against live switch state during the cutover window fails both as a safety practice and as a test method—it audits the fabric only while changes are being applied, creating real disruption risk for issues that should have been caught pre-cutover. Exam caveat: Air reduces cutover risk but does not remove the need for peer review of RT policy and a monitored production rollout checklist. Operational check: model the new county EVPN tenant in NVIDIA Air, validate isolation and dual-home behavior there, export or mirror the proven config into the change plan, and only then apply to Spectrum-X with NetQ baselines captured before and after cutover.