Start with the Result Type
Read what a tool claims to do before choosing inputs. A calculator predicts from a model; a browser test reports accessible signals; a guided check produces next steps from selected observations; a lookup returns an external record. These cannot substitute for one another. A connection estimate is not an ISP benchmark, and an IP registration country is not GPS location.
Units and Independent Examples
Decimal file units use powers of 1,000 and binary units use powers of 1,024. A byte contains eight bits. One decimal GB transferred at 100 Mbps takes 80 seconds at a constant ideal rate. One GiB at that same rate takes about 85.899 seconds. The tool keeps those inputs separate instead of treating GB and GiB or Mbps and MB/s as equivalent.
Models use separate pure functions so their results can be checked without rendering a page. Tests compare known cases and independently derived values, not just a copy of the displayed formula. Examples on the page identify their assumptions and do not claim to be a measurement of the visitor’s device.
Input and Boundary Validation
Checks cover empty and malformed numbers, unsupported units, negative sizes, zero rates, fractions where a whole count is required, date boundaries and numerical overflow. Local decimal conventions are accepted where implemented; ambiguous thousands separators are avoided. A rejected value stays an error rather than becoming a plausible-looking zero.
Displayed precision is selected for reading. Transfer duration may be rounded upward to a complete second while a raw model uses fractional seconds. Compare at the stated precision and keep the exact units when reporting a discrepancy. Estimates with changed efficiency or reserve values represent changed assumptions, not measured performance.
Browser and Network Evidence
Keyboard and pointer tools can report events delivered to the page. They cannot certify switch electronics or raw polling rates. Media tools need browser permission and do not measure acoustic sound pressure or optical quality. Guided display and connectivity checks retain unknown answers instead of discovering settings they cannot access.
Live workflows validate protocol, response size and supported providers. Controlled fixtures exercise missing data, invalid records, timeouts and private-target rejection. Production-loopback HTTP tests confirm real status codes and endpoint bounds, but they do not prove that every external provider will work from a future hosting plan.
Translation and Technical Sources
The five editions use the same calculations. Reviews compare field meaning, units and equivalent numerical examples, including French Mo and Go where appropriate. en-GB uses its own route and the preferred English canonical for identical content. No native-language sign-off is claimed.
Primary specifications and provider documentation support technical rules. Competitor inspection informs coverage gaps without supplying copied prose or a ranking guarantee. Dates on pages indicate a method or content review, not a certification or a promise that firmware, provider limits and platform rules never change.
What Remains a Human Check
The available automated suite covers calculation cases, types, production builds, visible article content, links, metadata and loopback HTTP behavior. Real browser hydration, device permissions, physical printing, assistive technology and production hosting require separate review. Security patches are tracked with their own dated report.
When a result matters for a purchase, configuration change or operational decision, compare the model with the current device specification and a relevant real observation. The correction workflow asks for evidence that lets a result be reproduced, so a confirmed problem can be fixed in the model, content or both.