Core Web Vitals are assessed at the 75th percentile of real user sessions. A Lighthouse score is a rehearsal: one synthetic load on a clean machine, with none of your actual visitors in it. Measured works to the field number, and every engagement reports before-and-after field data instead of a promised outcome.
Andrei Alex — Senior frontend engineer with over ten years on high-traffic commerce: React, headless Shopify, and the render-path work that decides Core Web Vitals.
One engineer, not an account manager. You talk to the person writing the code, and the invoice comes from a registered company.
There is nobody to hand your ticket to. That is the offer rather than a limitation of it: the person who reads your field data is the person who opens the pull request, so nothing is lost between the diagnosis and the fix, and there is no layer between you and the reasoning. It also means capacity is finite, and that I will say so when a job is not a fit.
Google assesses Core Web Vitals at the 75th percentile of real sessions, which means a quarter of your visitors are having a worse time than the number you are judged on. Lighthouse runs one synthetic load on a clean machine and never sees those visitors. The two can disagree for a long time without anyone noticing — the case study is an engagement where they disagreed for exactly that reason, and where the lab number was the reassuring one.
So the reporting is the same shape every time: field data before, field data after, and a plain statement of what moved and what did not.
Worth knowing that the last category exists as a business model: Shopify publishes a guide to detecting fake performance apps — services that sniff for Chrome-Lighthouse in the user agent and serve a manipulated page to the test tool, leaving real visitors exactly where they were. Shopify says it has removed most such apps from its App Store.
Measured publishes its own measurement work rather than reheating other people's. The current study scanned 2,123 Shopify storefronts and found that 78.5% still carry script tags, which stop loading on 1 March 2027. The sample frame, the method and the known biases are published alongside the result on the research page, including the parts that weaken it.
The same standard applies to client work: if I cannot show you a number, I do not claim it.
Send the storefront and whatever you already know about its numbers. If an engagement is not the right first step, I will say so.