A blog's speed is three numbers, not an adjective

A blog is two workloads wearing one theme. Logged-out readers hit finished HTML that a cache can answer without waking PHP; you hit the dashboard, the media library and the editor, none of which can be cached at all. Almost every complaint that a blog feels slow is one of those two, and they have different fixes.

So start with figures rather than adjectives. Five runs of `curl -s -o /dev/null -w '%{time_starttransfer}'` against a cached post give you a median TTFB. One Lighthouse run in Chrome DevTools, Mobile, throttled to Slow 4G, repeated three times with the median taken, gives you LCP and INP. Save all three with today's date.

The failure this page is written against is the cargo-cult fix: a second caching plugin installed on top of a server cache, an image plugin fighting built-in WebP conversion, and a fortnight spent on a number nobody measured before or after.

A published post is identical for every logged-out reader, so LiteSpeed hands over stored HTML and neither PHP nor MySQL wakes up. Nothing else on the site enjoys that. Category page twelve, an internal `?s=` search, the editor autosaving, wp-cron firing on somebody else's page view — all of those run the full stack every time, and they are where a slow blog actually hurts.

That is the layer an object cache repairs. Redis stores the result of individual queries rather than whole pages, which is why it speeds up wp-admin when a page cache cannot. Measure it honestly: append a unique query string to bust the cache, time that URL, and compare it with the cached figure. The gap between the two is your real server speed.

A post that gets picked up somewhere large is the easiest traffic a blog will ever serve, because every one of those readers is logged out and asking for the same cached bytes. What breaks that arrangement is a template full of third-party scripts and an uncompressed hero image; WebP conversion is built into the platform, so the image half of that costs nothing but ten minutes.

After any change, re-run exactly the same tests at the same time of day: five curl runs, one throttled Lighthouse run, medians written next to the old ones. If TTFB did not move, hosting was not your bottleneck and you have learned something more useful than a faster number would have told you.