What must you verify before claiming that a local LLM deployment is offline?

What must you verify before claiming that a local LLM deployment is offline?

1. Inference still starts and completes with network access blocked, and logs, telemetry, backups, and cloud features have explicit policies.
2. The model weights fit entirely in accelerator memory.
3. The HTTP endpoint uses localhost in one development configuration.