When speed comes up, everyone looks at the number in the coloured circle first. It is a useful number, but the way it is useful gets misunderstood. A Lighthouse score is several metrics weighted into one summary, and a summary is not a task. “Please take us from 68 to 90” tells nobody what to do.
What the score actually measures
Lighthouse loads the page once under a simulated set of conditions: a throttled connection and a device with limited processing power. Under those conditions it records when the screen paints, how long the page stays unresponsive, and how much the layout shifts. Two consequences follow.
First, the score moves between runs. Network conditions and third-party response times are never identical. So you do not measure once and report it — run the same URL three times, take the median, and use three runs on each side of any comparison.
Second, this is not real-user data. The devices and connections your customers actually use may differ from the simulation. Which one to trust when they disagree is the next part’s subject; the short answer is that both are correct and they answer different questions.
Read below the score
Scroll past the score and you reach the opportunities and diagnostics. That is the actual work order. Each entry names a target and a cause: an image served far larger than it is displayed, a script occupying the main thread for a long stretch, text invisible while a font is still downloading.
A marketer’s job here is not to fix these but to translate them. Open the list and a good share of it traces back to things marketing added — the hero image, the tracking tags, the chat widget, the extra font weight. In other words, the list mixes items for a developer with items that are a marketing decision, and sorting them is your part.
What to send along with the report
Throwing a report link over the fence starts a long conversation. Four details end it quickly: the URL you measured (an actual entrance page, not the home page), the device profile, when you ran it, and two or three items you have prioritised.
It also helps to agree that 100 is not the goal. The goal is a customer who is not waiting, and the score is a proxy for that state. Stripping out a feature the business needs to chase the last few points swaps the end for the means. If you want something to check with yourself, there is a site diagnostic plugin among our free tools, and the route that hands both diagnosis and repair over as one job is the optimization program.
Next part
Sooner or later the tool’s score and your real-user data will disagree. The next part is about that gap — and the conclusion is that neither of them is wrong.