[autoantohaki]

Requested by mau

Before: the clients from #1 could poll everything, but relay-dash itself still showed an empty frame.

After: relay-dash opens a live dashboard. The top half lists every impossibleG service with a ready / not ready / unreachable mark, the reason it is not ready, queue depth (highlighted when anything is waiting), active requests, and loaded models. The bottom half shows pack-sentinel's totals, which collectors are running, active incidents sorted by severity (with a "N not shown" note when they don't fit), and the latest signals. It polls every intervalSeconds, r polls now, q quits. --snapshot polls once and prints one frame.

▌ Relay Dash · updated 0s ago
SERVICES 1/2 ready
  name       state               queue  active models
● inference  ready               3      2      qwen3-8b
× ocr        connection refused  –      –

PACK-SENTINEL
● 1284 signals · 17 incidents · collecting ssh-guard

Incidents 4 active
CRIT open den       SSH brute force from 203.0.113.9                    2m
HIGH open den       Secret committed to relay-chat: AWS key            14m
MED  ack  den       Honeyfile opened: payroll.xlsx                      3h

Incident titles and signal text come from logs, so everything drawn from a watched system goes through clean(), which strips escape sequences and control characters (SECURITY.md now says so).

How: usePoll runs pollOnce without overlapping a slow poll and aborts requests on unmount. DashboardView is a pure view of one poll's results, which --snapshot and the view tests render directly. The frame is still the local stand-in; it moves to Relay Shell once that is published. This reopens #2, which was merged into feat/sources instead of main; the change is identical.