How this desk measures anything, and where it stops
Most sites in this category publish numbers without saying where they came from. This page exists so that every figure elsewhere on the site has somewhere to point. If a number on this desk cannot be traced to something on this page, it should not be trusted.
Four kinds of figure, and only four
Everything we publish falls into one of four buckets, and each bucket is labelled on the page it appears on.
| Kind | Where it comes from | What it can support | What it cannot |
|---|---|---|---|
| Published rule | Public regime texts such as customs thresholds and tax collection mechanisms. | A statement about how a regime is structured. | A prediction about your specific parcel, because classification is item-specific. |
| Carrier document | Service-level documentation: divisors, restricted lists, charging units. | A statement about how a line bills or what it will refuse. | A rate, because rate cards change without any change in structure. |
| Community-reported range | Buyer threads and self-reported windows, collected and counted rather than averaged. | A range within which most reports fall, with its width stated. | An average, because our sample is not constructed for one. |
| Editor measurement | Counting, resolving and comparing things by hand on a stated date. | A statement about a specific set at a specific time. | Anything about the population those items came from. |
Why ranges rather than averages
A range is not vagueness. It is the honest output of inputs that are themselves ranges: a service fee that moves with promotions, a payment channel priced as a band, a carrier that applies a peak surcharge by calendar window, and a tax that depends on the classification of an item rather than on its price alone.
Given those inputs, a single total would require choosing a value for each variable and hiding the choice. That is what most estimators do, and it is why estimates tend to be wrong in one direction — the chosen values are the favourable ones. Our tools return bands so the choice is visible and therefore arguable.
Where an average would be defensible, we still do not publish one on transit, because our transit sample is a convenience sample of self-reported windows. A convenience sample over-represents whatever was easiest to find, and the resulting average would look authoritative while being biased in a direction we cannot measure.
What we refuse to publish
Four categories are excluded by policy rather than by omission.
- Coupon codes. We tested the genre and retired it. In our Q1 2026 sample of twenty publicly listed codes, nineteen followed a mechanical naming pattern and all twenty carried the same no-public-expiry note. None resolved against a checkout total. A code list has a shelf life measured in days, and a page that is wrong most of the time trains readers to distrust the whole site.
- Prices and availability. This desk is not a shop. We do not display product prices, we do not show stock state, and we do not emit structured data that describes an offer. Buying happens elsewhere, and the index here describes shapes rather than stock.
- Product photographs. We do not mirror another site's imagery, and we do not use placeholder images to make a page look fuller. The visual system here is typographic and diagrammatic by design, which also happens to be why the pages load quickly.
- Figures we can no longer derive. When a figure cannot be re-derived from its source, it is retired and the retirement is recorded rather than carried forward. Our Q2 2026 snapshot records two such retirements.
The four ways a figure here can be wrong
We would rather name these than pretend they do not exist, because a reader who knows the failure modes can compensate for them.
- Stale by structure. A published rule can change while our page still reflects the previous structure. The review date on each page is the defence, and the quarterly cycle is why it is quarterly rather than annual.
- Right figure, wrong scope. A divisor recorded per carrier when it is actually set per service is an example we have made ourselves: it was corrected in Q1 2026 and the correction is in the snapshot series. Scope errors are more common than value errors.
- Sample bias. Hand-collected figures describe what we could find. Where we collected eight lists, the conclusion is about eight lists.
- Classification uncertainty. Tax ranges depend on how an item is classified, and classification depends on material and use. Our ranges account for the rate band, not for a misclassification by whoever files the entry.
The review cycle
Every page carries a review date, and that date changes only when the content changed. We do not refresh dates to look current, and we do not touch a modified timestamp to encourage a crawl. Quarterly reviews cover the six destination pages, the eight tool pages and the twelve provider pages; write-ups are reviewed on the cycle recorded on each one.
When a review changes a figure's basis rather than its value, the change is recorded in the snapshot series and summarised in the changelog.
Corrections
If a figure here is wrong, we want the page address and what is incorrect. You do not need to provide a source, though it helps and we will say so if we cannot verify the correction.
Write to [email protected] with the page path in the subject line. Corrections that change content are recorded in the changelog with the date they were applied.