The site already in place, brought back under budget and kept there.

Usually not, and finding that out is part of the diagnosis. A build creates a new site from scratch; a remediation pass fixes the site already in place. Most slow sites are not fundamentally broken, they carry a handful of specific, correctable faults. Measurement comes first, and the read is straight: if a pass will clear the budget, a pass is scoped. If the site is structurally beyond repair, that gets said plainly rather than selling a fix that cannot hold. The diagnosis takes priority over an upsell into a rebuild that is not needed.

Because the lab score and the grade Google actually uses are two different readings. A lab test runs one simulated visit on one machine. Google grades the page on field data, the real experience of real visitors, gathered in the Chrome User Experience Report at the 75th percentile. A page can score well in the lab and still fail in the field, most often on Interaction to Next Paint, which only shows up when real people tap real things. Diagnosis runs against the field data, which is the reading that moves ranking (web.dev Core Web Vitals; corewebvitals.io, 2026).

The work and the measurement are guaranteed, not a number no firm controls. The commitment, in writing, is to bring the pages under the agreed budget and to show the readings before and after. Field Core Web Vitals are drawn from a rolling window of real-user data, so the public grade updates over weeks as that data refreshes; the exact day it flips or a specific ranking movement cannot be promised, because those depend on Google and on traffic no firm controls. A promise of a guaranteed pass by a fixed date is a promise no one can deliver. The work measures, fixes to budget, and reports the readings.

It is real engineering, directed by a person. A plugin papers over a symptom and leaves the cause; a lab screenshot flatters a page real users still find slow. Each fault is traced to its root in the live page and corrected there: the loading path, the JavaScript cost, the layout stability. Every engagement is directed by a specialist and reviewed before delivery. Better tooling makes that expert work faster and more precise. It does not replace the diagnosis or the judgment, and a generator's output is never handed over and called a fix.

A two-fault fix on one template and a deep pass across a heavy, script-laden site are not the same job, and a single published number for both would be a fiction. The sequence that holds up is to measure the site first, see how far it sits from budget and why, then scope the exact work and confirm the figure in writing before any commitment. The deliverables and the standard are published so the substance is visible first, and the figure follows only once the site has been read.

It removes a penalty and a leak, which is different from promising a spike. Core Web Vitals are a confirmed ranking signal, and a slow page loses visitors before they reach the offer, so a faster page competes on fairer terms and stops shedding traffic it already earns. What the fix cannot do is invent demand or move a ranking on its own; those depend on the content, the market, and engine behavior no firm controls. The technical cause is fixed and the readings are reported. No revenue figure gets attached that cannot be stood behind.

In most cases, yes. Part of the diagnosis is confirming what the platform allows to be changed and how deep the access goes. Many of the highest-impact fixes, image handling, script loading, layout stability and font strategy, are reachable on common platforms and page builders. Where a platform hard-caps what can be corrected in place, the exact faults that can be cleared and the ones that are structural get stated plainly, with a straight recommendation rather than a fix that cannot hold.

The fixes themselves are permanent, but a live site changes, and every new script, image or feature can spend the speed back. That is why the guardrail tier exists: a performance check wired into the deploy process that catches a budget-breaking change before it ships, instead of months later in the field data. Without it, periodic re-measurement is the recommendation. The risk that applies to a given site is stated plainly, so the level chosen fits how often it changes.

Yes, outright. The changes are made to the site in place, on its own platform, under your control, and what was changed and why is documented in plain language at handover. There is no lock-in and nothing to rent. You are free to run it independently, keep the firm on for further visibility work, or hand the documentation to another team. The next engagement is meant to be earned, not locked in.