DEV Community

Nandawula Kabali-Kagwa
Nandawula Kabali-Kagwa

Posted on

Hardcoding Reality: Replacing Platform Vanity Metrics with Ground-Truth CI/CD

We need to talk about dashboard theatre.

For the past year across our 8 production apps, including VarsityOS keeping students afloat through NSFAS budgeting cycles and AdminOS powering SME back offices, I’ve noticed a dangerous developer trap: we build our mental models around what our hosting providers want us to see, rather than what our systems are actually doing.

This week in JarvisOS, we broke the illusion.

The Day We Fired Vercel From Our Cockpit

Our engineering dashboard used to look pristine. Vercel’s API fed our deployment status, painting a picture of unbroken global health. But here in East London, South Africa, running code against intermittent power grids and unpredictable local networking, a glowing deployment badge across the Atlantic doesn't prevent a silent Supabase timeout.

So, we performed open-heart surgery on our tooling:

  1. Shifted the Source of Truth: Our deploy-impact reader was rewritten to read GitHub Actions directly instead of trusting external API wrappers.
  2. Purged Environment Ghost Towns: We replaced legacy Vercel environment advice with native Cloudflare Worker secrets across our messaging layers.
  3. Enforced Strict Postmortems: Postmortems are now strictly grounded in measured facts. If there is no outage evidence in the logs, there is no postmortem written. No drama, just telemetry.

The Code Behind the Curtain

We rolled out Phase 1 of our probe state machine, complete with automated tests and explicit cost modeling on Cloudflare. We also locked down our security perimeter with path normalisation running before the auth gate, an SSRF guard on our email image proxy, and targeted patches for two XSS sinks.

// Enforcing strict status rules in our health worker
export function validateHealthPulse(metrics: ServiceMetric[]): HealthVerdict {
  const failing = metrics.filter(m => m.hasError);

  if (failing.length > 0) {
    return {
      status: 'degraded',
      failingChecks: failing.map(f => f.checkName),
      timestamp: Date.now()
    };
  }

  return { status: 'healthy', failingChecks: [], timestamp: Date.now() };
}
Enter fullscreen mode Exit fullscreen mode

Per-app Supabase health alerts now name the exact failing checks, fire precisely once a day, and refuse to point backward at obsolete Vercel routes. Our CI quality net now runs on the runner itself: tsc --noEmit, gitleaks, and npm audit act as unyielding sentries before any container (v97 and counting) gets stamped for production.

The Takeaway for Full-Stack Builders

Stop letting your deployment provider define your system's health. Build your probes to scream when they are disconnected. Make your error logs specific. Ground your metrics in the soil where your users actually live.

The code doesn't care about your feelings, and your infrastructure shouldn't either. Make it tell the truth.

Top comments (0)