The three numbers to ask for before you sign

A studio can show you a green circle and still ship a page that feels broken on a phone. Lighthouse runs in a lab. Google grades Core Web Vitals from the Chrome User Experience Report, at the 75th percentile of real visits. Three quarters of your visitors have to clear the bar. A fast average does not hide a slow tail.

The bars, from web.dev, are still the ones published for 2026. Largest Contentful Paint at or under 2.5 seconds. Interaction to Next Paint at or under 200 milliseconds. Cumulative Layout Shift at or under 0.1. Poor starts at 4 seconds, 500 milliseconds and 0.25. INP replaced First Input Delay on 12 March 2024. A proposal that still talks about FID has not been updated.

The three fail for different reasons, which is why a CDN is not a complete answer. LCP is usually one image, a font, or a slow first byte. INP is JavaScript on the main thread after the bytes have arrived. CLS is an image with no width and height, a banner injected above the text, or a font that swaps and shoves the line. web.dev’s Renault write-up, from more than 10 million visits, tied a one-second LCP improvement to a 14-point drop in bounce rate and a 13% conversion lift. That is one case, not a promise for yours.

Ask for the field section in PageSpeed Insights and the Core Web Vitals report in Search Console, mobile and desktop separately. The HTTP Archive’s 2025 Web Almanac, published January 2026, put the mobile pass rate for all three at 48%. Desktop can look fine while the phone does not. If the site is a static brochure with no client script, INP often has nothing to measure. Say that. Do not invent a score.

We treat those three as acceptance checks on a custom website, not as a ranking guarantee. Relevance still outranks a fast page about the wrong thing. Speed is the part of the build we can actually hold.