press_settings.presses and settings.downtime_reasons are re-read on every 12s
overview poll; route them through the kernel cached_select() TTL cache so they
stop consuming a shared-pool checkout each poll. Part of the gateway
connection-pool-saturation fix (see fastAPI/agent_docs/db-pool-relief.md).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On a failed fetch the store now keeps the last real payload (stale > fake)
and flags `disconnected` instead of overwriting live data with bundled demo
data. Sample fallback is gated behind import.meta.env.DEV, so `vite build`
tree-shakes it out entirely (verified: no sample strings in dist).
- cold start / no data: centered reconnect spinner that keeps retrying
- data present + drop: stale data stays, header shows amber "reconnecting"
- polling retries faster (4s) while disconnected, recovers automatically
- sample.ts data moved into sampleOverview() so the module fully tree-shakes
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
