Know what users actually experience: capture and group errors with context, collect real-user performance data, sample and batch telemetry, connect frontend traces to the backend, and alert on what matters.
Your backend runs on a few hundred machines you control. Your frontend runs on millions of devices you don't: different browsers, extensions, CPUs, networks, and screen sizes, running whatever version of your app they loaded three days ago.
Backend teams wouldn't run without logs, metrics, and traces. Yet many frontends ship with only lab Lighthouse scores and user complaints. Frontend observability has four pillars:
Lab data vs. field data: Lighthouse runs on one emulated device under one network profile. It's great for catching regressions in CI, but it can't tell you that 20% of users in India on low-end Android get a 6-second LCP. Only field data can. Make decisions from field data (p75 by segment), and use lab data to debug.