A short take on what tends to be worth watching day to day, not the full list every monitoring tool can spit out. A common failure mode is tracking everything that blinks green. Alerts get noisy, people mute the channels, and the slowdown that matters still shows up as a support ticket or client email. Teams that keep the list small usually stick to metrics someone will actually look at on a Monday.
Server and origin metrics worth paging on
Availability and TLS: can the server (or the CDN edge that matters) be reached over HTTPS, and is the certificate still valid with enough warning?
Response time at the edge: time to first byte, or a simple timed GET on a URL that should stay healthy. Not the same as full page timing, but it shows the box or the CDN path is sick before anyone digs into front-end metrics.
Error rate: 5xx ratio on the main host or load balancer. A page that still “loads” while 5xx is climbing is still an incident.
CPU, memory, disk, and queue depth on the hosts that serve traffic. Alert when a limit stays broken across a few checks, not on a single spike.
Most teams do not page on every CPU blip. They page when the site is down, errors climb, or latency stays high long enough that users will feel it. One-off spikes usually belong in a weekly review, not a 3 a.m. call.
Page metrics on the important URLs
A healthy backend is not the same as a fast page. An API that looks fine with a heavy third-party script can still make mobile checkout feel broken. That is why page metrics are usually tracked separately from infrastructure.
LCP, INP, CLS on the pages that matter (home, pricing, signup, checkout, main templates). Field data from CrUX or Search Console when the URL has enough traffic; scheduled lab runs so deploys get caught before the field data moves.
First Contentful Paint and Total Blocking Time when diagnosing, not as the only numbers shown to other people.
Transfer size/request count on those same pages. Useful when a marketing launch quietly adds three tags.
Uptime of the public URL as users see it (not only the origin health check). CDN or DNS failures show up here first.
Lab and field answer different questions. A green Lighthouse run on a quiet Tuesday does not cancel an amber CrUX status on the same page. When they disagree, different people often own the follow-up.
Stuff that usually stays off the daily list
Dozens of Lighthouse audits as standing alerts. Apdex scores nobody can explain in a standup. Tracing every endpoint on every marketing page. Fine for a deep look later, not for the daily list.
Simple rule from how this usually plays out: if nobody owns the threshold and nobody does anything when it breaks, it should not be on the list.