Truth first

A dashboard that looks healthy while data stopped arriving is worse than no dashboard. GRIDCITY's first rule: every light maps to exactly one documented data field, and unknown is never shown as zero or healthy.

The rules

1. Unknown is unknown. Every metric is nullable. null is drawn as "no light" and shown as a dash, never as 0. Rates (CPU, network, queries per second) stay unknown until two polls exist.
2. Offline is dark. A server whose connection failed has no light and no motion. Its last values are not shown as current.
3. Stale freezes. Data older than three poll intervals freezes and turns grey with "last seen N ago", in the city and in the panels, recomputed every second in the browser.
4. Demo is labelled. The simulated city always shows DEMO DATA, and demo and real data are never mixed in one session.
5. The past is labelled. Time travel shows REPLAY, and anything not recorded at that moment is unknown.
6. Status rules are one function. Server and browser compute status the same way (from packages/shared), and the demo runs its fake samples through the real rules.
7. Could not tell ≠ none. Service discovery says when a tool is missing, rather than "no containers".
8. Proven or not. The Doctor marks a fix proven only when a real connection with it succeeded.

Status

Status	Server	Resource
unknown	never polled successfully	never polled successfully
online	connected, within all thresholds	reachable, within all thresholds
degraded	CPU > 90 % for 3 polls, memory > 90 %, or any disk > 90 %	connections, lag, slow queries, cache hit, storage, latency, certificate, targets (see Resource kinds)
offline	SSH connect failed or heartbeat lost	the poll failed

All thresholds are configurable (Settings, per-server in Edit, resources.yaml).