Processor and graphics card specifications are live

·Blog

5 min read

What a measurement revision is

SystemCheck Updated 29 October 2026

A measurement revision names exactly how a run was measured: what each stage ran, for how long, what it sampled and how each figure was derived.

  • It is written as year, month and a sequence, such as 2026.09.18, and every report carries it.
  • Any change to what a stage measures moves the revision in the same change, and the methodology page records why.
  • Some revisions change a figure's scale, and reports across them are not compared. Others only add information, and figures still compare.

Why a score needs a version

A score is only meaningful inside the method that produced it. Change a stage's length, a sample gate or a weighting, and every earlier number shifts underneath you, even though both still print as a plain figure. So every SystemCheck report is stamped with the measurement revision it was made under, and the methodology page records the changes under its corrections policy.

Two kinds of change

Some revisions change what a figure means. Revision 2026.09.17 replaced the CPU stage's JavaScript kernel with WebAssembly, because the old kernel's speed was largely the JavaScript engine's. On the validation laptop the new kernel read about 5.4 times the old one, so CPU figures are not compared across that revision.

Others only add information. Revision 2026.09.16 attached an interval to each headline figure without changing any figure, so results compare across it. The revision still moves, because a result now carries something it did not before.

Changes that were not to the stage itself

A revision can also record a change to what a stage competed with. In revision 2026.09.11 the site's animated floor was made to hold still while a stage measures: interleaved on the validation laptop, path tracing read 21 samples a second with the floor moving against 32 with it still. The stages' work had not changed, but the figures had, so the revision moved and reports across it are not compared.

How to use it

When you compare two of your own reports, check that they share a revision, or that the methodology page says figures compare across the revisions between them. If they do not, run again rather than comparing. The revision is the reason SystemCheck can say what changed and when, instead of quietly moving the numbers.

Terms used here