A speed score is a measurement, not a permanent label. The tested page, upstream services and lab environment can all vary between runs.
Common sources of variation
Server load, cache state, ad or analytics requests, changing content, third-party response times and the test environment can move individual measurements.
Look for patterns, not one point
Run comparable tests, keep the device strategy consistent and pay attention to repeated opportunities. A one-point movement is usually less important than an LCP problem that appears in every mobile test.
A better comparison method
- Record the first result.
- Make one meaningful change.
- Run the same device test again.
- Compare measurements and opportunities, not only the overall score.
- Confirm important improvements with real-user data when available.