Synthetic tests

Synthetic tests run scripted checks from controlled environments—Lighthouse CI, Playwright timings, k6 smoke, or uptime pings—to catch regressions before users do. They are **repeatable** but may not match real devices and networks.

| Use case | Tooling examples | |----------|------------------| | Web vitals budget | Lighthouse CI, SpeedCurve | | API latency smoke | k6, Artillery, cron + curl | | Critical journey | Playwright with performance marks | | Deploy gate | Fail CI if LCP budget exceeded |

					performance.mark('start');
await page.goto('/checkout');
performance.measure('checkout-load', 'start');
				

Run synthetics on **staging that mirrors production** config (CDN, compression, SSR). Store history to diff builds. Complement—not replace—field data.

On interviews: lab vs field; flaky synthetic causes; what to gate in CI vs monitor only; synthetic geography choices.

Common pitfalls: testing localhost without CDN; single-run Lighthouse scores; ignoring variance between runs.

The trade-off is early regression detection versus maintenance of realistic test environments.

Checklist:

• Budget key routes in CI.
• Run multiple samples; use medians.
• Match production caching and CDN.
• Pair with RUM for validation.