What's currently true.
The exact build this response was rendered from, real-user performance measured from actual visits, and uptime checked by an independent monitor. Updated continuously, not a dated snapshot.
This deployment, exactly.
No deploy-timestamp field: Vercel doesn't inject one, and a cold-start clock read isn't reliably the real deploy time on a warm instance; that would be a claim without a mechanism behind it, so it's left out.
Real visits, not lab runs.
Core Web Vitals collected from actual visitor page loads (only when analytics consent is granted), aggregated into a rolling 30-day p75. Below 20 samples, a percentile is noise, not a signal. Those metrics show a collecting state instead of a number.
A p75 below 20 samples is noise, not a signal.
Checked independently.
100.00% uptime over the last 112 checks · checks since Jul 10, 2026. Checked every 10 minutes by an independent GitHub Actions workflow, not self-reported by this app, since a server can't honestly measure its own downtime while it's down.
GitHub Actions cron is best-effort with documented delay under load, and disables entirely after 60 days of repository inactivity. This is not a dedicated-SLA monitor.