What is ours
Logging friction, paywall behaviour, cancellation flow, whether a feature survives eight weeks of real use, and whether the arithmetic an app shows you is internally consistent. These are experiential and procedural, and we can observe them honestly.
We buy every subscription at retail. We have taken no review units, no briefings and no affiliate arrangements, and there are no affiliate links on this site. If that ever changes it will say so on the page it changes on, in the same size type as everything else.
What is not ours
Accuracy. We do not run a weighed-meal protocol, because we do not have the apparatus for it, and a training desk claiming a laboratory result it cannot produce would be committing exactly the error this site exists to point at.
Where an accuracy figure appears here it is an outside measurement, named on the page with a link. We also distinguish three states, always:
- Independently measured and reproduced. Two unrelated groups, two different test sets, the same result. This is rare and it is the only thing we treat as settled.
- Independently measured once. Better than nothing, and not conclusive — a single outside test cannot rule out its own design doing the work.
- The maker's own figure. We will print these, labelled. A vendor number is not a lie; it is just not a measurement in the sense that should move your decision.
Why the distinction matters here specifically
Test-set composition alone moves food-estimation results by several percentage points. An estimator that handles flat plated food well and composite dishes badly scores very differently depending on what the tester cooked. When the tester and the vendor are the same organisation, nobody outside can tell which you are looking at — and the vendor may not be trying to mislead anyone.
That is why we weight replication above sample size. A larger single study narrows the error bar around a result that may still be an artifact of one test's design. Only a second party building a different test rules that out.
Corrections
We change numbers when they turn out to be wrong and we say so on the page rather than editing silently. If you can show a figure here is wrong, that is a favour and we will treat it as one.