Define a precise question
Start with a hypothesis: a better-sized image should reduce transferred data; a deferred script should reduce initial work. Note the change and expected metrics. If you change the theme, scripts and server at once, you may see an overall change, but it becomes harder to identify which intervention caused it.
Keep conditions consistent
Compare the same final URL with the same profile and engine. Local mode depends on your computer and connection, while hosted audits run elsewhere. Note any changes to the Lighthouse version. Close unnecessary applications to reduce competition for resources, and check that displayed content is comparable between tests.
Repeat measurements
A single before and after test is not enough to establish a reliable trend. Run multiple tests to observe the spread. Very small changes may come from the network or served content. Pingsivo lets you compare two reports; that view is not a statistical analysis of all visits and does not calculate certainty about your fix’s effect.
Interpret the differences
Lower LCP, TBT or CLS generally points in the desired direction, but also check quality and functionality. A score may move differently from one metric because it combines several measurements. If a value is missing, its difference must not be calculated. Keep reports and descriptions of changes to make later review easier.
Your next action
Choose one page, note what you want to check and save a report. After making a change, use comparison to examine the differences. Review the context before concluding that there has been an improvement.
Open the optimization checklist ↗Sources and scope
This guide suggests a working method. No numerical improvement is guaranteed. To check definitions and explore the audits: official documentation.