HBMHBM Rocket
ALL DOCS
DOCUMENTATION

Audits and alerts

6 MIN READ

HBM Rocket runs Lighthouse against your live URL every night, on both mobile and desktop. Scores are stored, charted, and compared to your target. Regressions fire alerts. The self heal pipeline reacts.

Audit cadence

  • Initial audit (mobile and desktop) within minutes of site creation
  • Nightly at 02:00 UTC for every active site
  • Manual on demand from the site detail Reoptimize button
  • Automatic re run when self heal completes a regen

What we audit

Lighthouse with these categories: performance, accessibility, best-practices, seo. Mobile uses the standard 4G simulated throttle and 4 times CPU slowdown. Desktop uses the standard preset.

Metrics tracked

  • Performance score 0 to 100
  • LCP (Largest Contentful Paint)
  • FCP (First Contentful Paint)
  • TBT (Total Blocking Time, proxy for INP)
  • CLS (Cumulative Layout Shift)
  • Speed Index, TTI, TTFB

CrUX real user data

When the Google CrUX API is enabled (set GOOGLE_PSI_API_KEY on the control plane), we also pull rolling 28 day real user metrics. Use the dashboard toggle on the site detail page to switch between Lab and Field views.

Alerts

When they fire

  • Performance score drops more than 10 points below target on two consecutive audits
  • Site has not phoned home in more than 30 minutes
  • Two consecutive audit failures (network, timeout, 500 from origin)
  • Worker exception during a queued job

Where they go

Email by default. Slack and Discord channels when those integrations are connected. Alerts auto resolve when the underlying condition clears (score recovers, heartbeat returns, audit succeeds).

Self heal

An hourly cron checks every site. If the latest audit is below target by more than the threshold (5 points by default), it queues:

  • Critical CSS regeneration
  • Local cache purge
  • Cache warmup from sitemap
  • A fresh audit

If the second audit is also below target, the alert opens. If it recovered, no alert is opened and an info event is logged.

Self heal is conservative by design. We would rather quietly fix a transient regression than cry wolf. The job timeline on the site detail page shows everything self heal did, even when it succeeds silently.

Audit retention

All audit JSON is kept for 90 days. Score and metric summary rows are kept indefinitely. You can export the full history from the API or the dashboard at any time.

Audits and alerts | HBM Rocket