Fix the input
Use one hostname, a known object size and a stated protocol. Changing multiple inputs makes comparison unreliable.
Method
Make the result repeatable.
A useful check records enough context for the same request to be repeated from another device or network.
Separate stages
Record resolution, connection, TLS negotiation, first byte and transfer time as distinct observations.
Use a short series
Run enough samples to expose variation. Keep the median, the range and any failures rather than selecting the fastest result.
Change one condition
When comparing networks or configurations, change one condition at a time and keep the endpoint and test object fixed.
What the numbers do not prove
A successful request proves that one path worked at one moment. It does not establish global reachability, an uptime guarantee or the cause of a later failure.
Likewise, packet loss from diagnostic traffic does not automatically imply the same loss for application data. Some networks treat diagnostic protocols differently.
Results on this site are explanatory examples. They are not a substitute for measurements from the affected client network.