/admin/api-keys is usually ~260ms (Faro), but queryKeyUsage() (lib/prometheus.ts) is uncached (cache: 'no-store') and fires its two PromQL queries sequentially, each with a 3s timeout — so when Prometheus is loaded/restarting the page blocks up to 6s before failing soft. Spiky, so it doesn't show in the few admin Faro samples.
Fixes:
Promise.all) — worst case 6s → 3s.unstable_cache (return serializable entries, rebuild the Map in the page — a raw Map doesn't survive the data cache) — takes the Prom round-trips off the hot path; a slow Prometheus no longer blocks every load.(Perceived-nav snappiness for this page was already handled by PAR-133's root loading.tsx + admin nav prefetch.)