speed.press
Lab data: Lighthouse 13.4.1 mobile run, graded above · Reproduce this run on PageSpeed Insights ↗
Why is this different from the score in my browser?
This view grades the mobile run. Lighthouse 13.4.1 runs at Google through the PageSpeed Insights API on an emulated mid-range Android with a 4× CPU slowdown and a throttled connection. Chrome DevTools on your own machine, on the Desktop preset, applies almost none of that, and the same page routinely scores 30–50 points higher there. Neither number is wrong — they are different tests.
Mobile is what Google ranks on, so this is the number that affects your
search performance. This report predates the desktop-graded headline: only the mobile run was graded. To reproduce it, open the same run on PageSpeed Insights,
or pass strategy=mobile to the PageSpeed API — it defaults to desktop.
Lighthouse also varies a few points run to run on the same URL. Treat a trend across several scans as the signal, never a single digit.
Lab data: Google PageSpeed Insights (Lighthouse, mobile). Server probes (TTFB, compression, cache, headers) are measured separately from Vilnius, Lithuania (EU).
Astro what actually works on this platform
Astro ships no JavaScript unless you ask for it, so a slow Astro site is almost always one of two things: islands hydrating when they do not need to, or the HTML being built per request when it could have been built once.
- Nothing indicated the HTML came from a cache on this request, which is worth confirming: a static Astro build should be served straight off a CDN.
- A crawler reads about 5,703 characters of text from your HTML — healthy, which is what Astro is good at.
- No CDN detected in front of the site — for a static Astro build that is the cheapest win available.
- Check your client: directives before anything else. client:load hydrates an island immediately and costs the visitor JavaScript on first paint. client:visible defers it until the component scrolls into view and client:idle until the browser is free; most islands want one of those. A page full of client:load is the usual reason an Astro site feels heavy.
- Know whether each route is static or server-rendered. Astro can output a static file, render per request, or mix the two per route. A page with no per-visitor content that is rendered per request pays for it on every visit. Prerender the ones that can be, and let the CDN serve them.
- Use astro:assets for images. It emits modern formats at the sizes actually used and reserves the space so the layout does not shift. Raw <img> tags are usually the largest remaining payload on an otherwise lean Astro site.
- Watch what a shared layout pulls in. An import in a layout is charged to every page that uses it. One heavy component or icon set there outweighs anything you will save on an individual route.
The full instruction set, and a machine-readable copy of this report, live at speed.press.md — paste that URL into your AI builder's chat and it can work straight from these measurements.
Top fixes ranked by points recovered
This is an Astro site — act on the report with your AI assistant
- ✓Platform detectedAstro — every check below links the official guideline for the fix in this stack.
- →Act on it with your AIThe machine-readable twin of this report gives an agent every failing check with its evidence, remediation and guideline link.
- →Check the originHosting sets the floor on TTFB. xCloud hosts any stack — agent/MCP access built in, bring-your-own-server pricing.
xSpeed is a WordPress plugin, so it can't automate Astro fixes — and that's fine. Every check below links its official guideline, and the machine-readable twin of this report gives an AI everything it needs to plan the work (5 checks name a concrete remediation):
Read the speed report at https://xspeedcache.com/scan/r/speed-press-34bb062720.md — it lists every failing check with evidence and the official guideline for the fix. Produce a prioritized fix plan for my Astro site, apply what you can, then re-test by POSTing {"url":"https://speed.press/"} to https://xspeedcache.com/api/scan and compare the scores.Run this scan without the form
Plug xSpeed Scan into your agent stack, or test any site from a chat.
Connect this scanner to your AI assistant
Free. No API key, no sign-up. Scan any public site and get the same graded report as this one, with the fix for every failing check.
Endpoint
https://xspeedcache.com/scan/mcpClaude Code — one command
claude mcp add --transport http xspeed-scan https://xspeedcache.com/scan/mcpClaude Desktop, Cursor, or any MCP client — add to your config
{
"mcpServers": {
"xspeed-scan": {
"type": "http",
"url": "https://xspeedcache.com/scan/mcp"
}
}
}run_speed_scanScan any public URL and return this whole graded report.get_speed_scanRead a scan already run, by its id.
Core Web Vitals & lab metrics
5/5 checks passed · 40/40 pts earnedServer response & caching
2/6 checks passed · 9.5/27 pts earnedAI prompt
Using my site's xSpeed MCP connection: enable page caching and warm this page with the preloader, then re-check TTFB.AI prompt
Using my site's xSpeed MCP connection: enable page caching, start the preloader so pages are pre-warmed, and confirm get_cache_status reports hits.AI prompt
Using my site's xSpeed MCP connection: enable browser-cache headers for static assets, then re-run the audit.AI prompt
Using my site's xSpeed MCP connection: connect Cloudflare (or configure the CDN rewriter) and purge, then re-scan.Asset optimization
4/5 checks passed · 16/16 pts earnedAI prompt
Using my site's xSpeed MCP connection: preload the LCP image for this page and exclude it from lazy loading, then re-test LCP.Past scans of speed.press newest first · every report keeps its link
| Scanned | Score | Grade | Graded on | TTFB | Lighthouse | |
|---|---|---|---|---|---|---|
| 2026-09-20 10:22 UTC | 97 | A+ | desktop | 130ms | 100 | View report → |
| 2026-09-15 10:25 UTC this report | 79 | C | mobile | 686ms | 100 | |
| 2026-09-15 09:37 UTC | 75 | C | mobile | 504ms | 100 | View report → |
Reports graded on different runs are not a trend: mobile-graded rows are the older rubric's headline, desktop-graded rows the current one. Compare like with like.