Four domains. Five pages. Twenty 100s.

The claim, stated carefully

On July 3–4, 2026, every page listed below scored 100 Performance, 100 Accessibility, 100 Best Practices, and 100 SEO on mobile PageSpeed Insights — Google's own lab test, which anyone can re-run at pagespeed.web.dev in about thirty seconds.

Two honest qualifiers before the table. First, PageSpeed lab scores fluctuate run to run — network conditions, test-server load, and Lighthouse version changes can move a score a point or two in either direction. These are the scores on the verification dates shown, not a promise that every future run returns the same number. Second, this covers the pages we operate or actively maintain — five pages across four domains — not some larger portfolio we're quietly excluding.

Why does this matter beyond bragging rights? Performance and accessibility are ranking and conversion inputs, not vanity metrics. One of the client sites above outranks a national insurance brand's local agent listing on mobile search. The other loads instantly on the phones its customers actually use. Perfect scores are the visible symptom; the cause is a site with nothing broken in it.

What a 100 actually costs

None of these pages got there with a plugin or a one-time "optimization pass." They follow a written engineering standard — the same numbered ruleset, applied to every build and retrofitted into older ones. The rules that do the heaviest lifting:

• Analytics never blocks the page.

  Where a site uses analytics at all, the tag is injected 1.5 seconds after the window load event — the visitor's first paint owes nothing to a tracking script. This one rule is routinely worth 2–3 Performance points.

• Zero font downloads.

  Every site runs on system font stacks — Georgia for display, the visitor's native sans for body text. No Google Fonts request, no render-blocking stylesheet, no invisible-text flash. Today's retrofit removed three external font requests from a single client page.

• Video never autoplays, never preloads.

  Hero video loads only after the page is fully interactive, and visitors who prefer reduced motion never download a single frame of it.

• Contrast is a token, not a hope.

  Every color pair is checked against WCAG AA. When a shade fails, the design token itself is corrected — once — and the fix propagates to every page that uses it.

• Touch targets have a minimum size.

  Every tappable element is at least 24 CSS pixels tall with real spacing — a rule we learned the hard way this week (see below).

• One self-contained file per page.

  No framework payload, no build pipeline, no folder of dependencies that can silently break. Every page is a single HTML file that can be audited by reading it top to bottom.

• Structured data tells the truth.

  Schema markup carries verified facts only — real service areas, honest price ranges, languages actually spoken. Search engines reward markup they can trust.

The one we broke today

A perfect record nobody ever broke isn't a record — it's a claim. So here is today's entry from our public incident log, unedited in substance:

The same discipline cuts the other way, too. This week we A/B tested a flashier version of our own homepage — animated score rings, more motion, more spectacle. It scored 98/100/96/100. The plainer master scored perfect. The flashy version lost, so we don't run it. The rings you see at the top of this post are its salvaged parts, demoted to a blog post where they can't cost anyone a point.

What this means if you own a website

Run your own site through pagespeed.web.dev on the Mobile tab — that's the test Google actually experiences your site through. Most small-business sites land in the 40–70 Performance range, usually because of the exact problems listed above: blocking analytics, font downloads, autoplaying video, oversized images.

Every number in this post is reproducible while it stands, and every mistake behind it is published. That's the whole pitch.