Boundaries
Management, trusted clients, services, IoT, sandbox systems, and guests occupy separate VLANs, with inter-VLAN traffic denied by default. Wireless networks preserve those boundaries rather than creating a parallel trust model.
Each boundary creates a place to express an operating decision—and contain its consequences. I routinely test systems at their limits, including intrusion, exfiltration, failure, and misconfiguration. Segmentation limits the blast radius when those tests succeed in ways I did not intend. Reachability is granted for a purpose rather than inherited from proximity.
Resilience
The network is designed to remain useful when the systems attached to it are not. Network equipment and server infrastructure have separate power paths, allowing switching, routing, and wireless connectivity to survive a server-side outage even when the services normally reached through them are unavailable.
That principle extends to dependencies. A power failure exposed a recovery path that depended on server-hosted DNS. Independent recovery paths followed. Redundancy within one failure domain is not resilience; the objective is to preserve enough network capability to diagnose and recover the rest.
Operability
Physical and logical organization deliberately correspond. Addressing conventions map infrastructure to VM identities; patch cables identify endpoint purpose; VLANs describe functional boundaries; and hardware remains modular rather than concentrating unrelated services behind common equipment.
Those choices reduce ambiguity during failure. A server switch should not take out Wi-Fi. A cable should communicate what it serves before it is traced. An address should provide context before a console is opened. The network is organized for diagnosis as deliberately as it is for normal operation.
What it means
The architecture assumes that components will fail, experiments will escape their intended boundaries, and dependencies will occasionally prove wrong. Its job is to contain those events, preserve enough independent capability to recover, and make the resulting problem understandable without first reconstructing how the network was assembled.