benchmark

Navigate to each target URL in a real browser and record LCP, CLS, INP, FCP, and TTFB against standard targets like LCP under 2.5s and CLS under 0.1. Also track page weight, JS bundle size, and render-blocking resources.

Save a baseline of current metrics, apply your changes, then run a comparison that outputs a delta table marking each metric as better or worse. Integrating the comparison into CI catches regressions on every pull request.

Track p50, p95, and p99 latency across repeated requests, plus response sizes and status codes. Testing with concurrent requests, such as 10 parallel calls, reveals behavior under load against SLA targets.

Yes, baselines are stored as JSON files in a `.ecc/benchmarks/` directory that is git-tracked. Committing these files lets every team member and CI pipeline compare against the same reference numbers.

Synthetic benchmarks miss real-user variability like device diversity and network conditions. Pair them with post-deploy monitoring such as canary watching to catch issues that only appear in production traffic.