How the numbers on this site are produced
Three source labels appear across this site. Every figure carries one of them, and the label answers a specific question: what would have to be true for this number to be wrong?
The three labels
- Measured
- Recorded from our own sample of 96 orders and parcels, collected on 2026-10-05. Stage durations, weight observations and the billed-weight comparison carry this label. The sample is small enough that we publish intervals rather than points, and every measured figure states its sample size.
- Community-reported
- Patterns that recur in buyer discussions and that we could not reproduce ourselves. Forecasts carry this label too, and are marked as forecasts on the page. It is the weakest of the three and we say so rather than dressing it up.
- Published
- Read from carrier tariffs, agent fee pages and destination threshold documents. Item values in the census are published figures. Published terms change without notice, which is why every one carries a collection date.
Where none of the three applies, we write Not measured. That is a deliberate choice: an empty cell tells you where the uncertainty sits, and a plausible guess does not.
Collection, in six steps
| Step | What happens | Failure handling |
|---|---|---|
| 1. Category enumeration | Each catalogue category page is read and its item count recorded. | A category that returns an error is dropped and the total is reduced rather than padded. |
| 2. Entry selection | Entries are taken in page order up to each category’s quota. | If a thumbnail is unavailable the entry is dropped and the next one is taken. |
| 3. Field extraction | Name, source price and image variants are read from the listing itself. | Missing price means the entry is dropped; we do not estimate a price. |
| 4. Band assignment | A weight band is assigned from the category baseline and title keywords. | The assignment rule is published, because it is the largest error source on the site. |
| 5. Term collection | Agent fee, storage and consolidation terms are read from published pages. | Unpublished terms are recorded as null, never estimated. |
| 6. Freeze and date | The set is frozen and dated. Nothing is backfilled afterwards. | Corrections go to the errata log with the date and the original value. |
Nothing in this process writes back to the source catalogue or to any agent system.
Four things this site does not do
| Refusal | Reason |
|---|---|
| Predict an individual order’s final bill | The census supports ranges, not predictions. Anything narrower would be false precision. |
| Claim an agent is the cheapest | Fee models cross over at different order sizes; “cheapest” without an order size is not a claim we can support. |
| Replace a carrier’s terms | We paraphrase and date; the operator’s own wording always governs. |
| Give tax advice | Duty figures are modelled from published thresholds. Thresholds and rates change, and the rules depend on your circumstances. |
Each refusal has a practical consequence: where we decline to advise, the tool says so rather than producing a number.
Review cycle
Census figures are re-checked monthly against the source catalogue. Agent terms are re-read quarterly. Thresholds are re-read when a destination publishes a change. Corrections are logged on the errata page with the original value, the corrected value and the date — and update timestamps move only when the content actually changes, never to prompt a re-crawl.