Method and limits: how these numbers were made
Every figure on this site has a sample, a window and a formula behind it. This page states all three, and states what we refuse to publish.
joyagoo.mobi is an independent record. Links to Kakobuy carry a source tag so the click can be counted (marked joyagoo.mobi) — it does not change what we list. How links work.
1. Sample and window — Every timing figure on this site comes from the line-day log, which holds 24 UK line records collected between 2026-08-01 and 2026-09-30. A record is opened when a parcel leaves the warehouse and closed when it is delivered, and warehouse days are counted separately rather than folded into the total. The window is stated wherever the numbers appear, because a 24-parcel sample describes one season and one destination country, not a service level.
2. Rate re-checks — Rates were re-checked 12 times across the same period, and each re-check is stored with its date. A band that has not been re-checked since the last sweep is displayed as Not verified rather than carried forward, and no band is ever interpolated between two neighbouring tiers. Where a published table changes mid-window, both readings are kept, because the change itself is worth recording.
3. QC readings — The QC protocol rests on 85 readings taken from the 85 recorded entries, which produced five positions and three outcomes per position. The positions are the main image, the wash or size label, stitching density at load points, hardware travel and finish, and the sole or size tag on footwear. The outcomes are pass, ask for a further photograph, or send back, and the outcome turns on where a fault sits rather than on how large it looks.
4. Weight tiers — The tier table covers 6 lines across 9 tiers spanning 0.5 kg to 10 kg, which is 54 recorded cells. A cell is filled only from a published table read on a dated re-check; empty cells stay empty. Because the table is read as a grid, a reader can see immediately whether two parcels a few grams apart fall into the same tier or straddle a boundary.
5. Volumetric weight — Volumetric weight in kilograms is length in centimetres multiplied by width in centimetres multiplied by height in centimetres, divided by the divisor for that line. With dimensions in inches, convert first: 1 in = 2.54 cm. A carton of 40 x 30 x 20 cm measures 24,000 cubic centimetres, so a divisor of 6,000 gives 4.0 kg, and a divisor of 5,000 gives 4.8 kg for the same box.
6. Chargeable weight — Chargeable weight is the greater of the scale reading and the volumetric figure, rounded up to the next 0.5 kg step. The rounding happens after the comparison, never before it, because rounding first would change which of the two figures wins on a borderline parcel. A scale reading of 3.24 kg against a volumetric figure of 3.10 kg becomes 3.5 kg, and the 0.26 kg difference is the whole reason the step is worth understanding.
7. Unit conversions — Imperial figures are converted rather than approximated, so that a reader working from a US listing can reproduce the result exactly. One inch is 2.54 cm, and one pound is 0.45359237 kg. A 2 lb garment is therefore 0.90718474 kg before rounding, which then becomes 1.0 kg of chargeable weight once the 0.5 kg step is applied. Conversions are shown on the page so that the input can be checked line by line.
8. Landed cost assembly — The landed cost total is built from four recorded lines: the item price, the domestic leg from seller to warehouse, the international line, and the add-ons chosen at packing. Credits and any confirmed coupon value are subtracted last, because a deduction applied earlier would change the tier a parcel sits in. We record the parts rather than a single merged total, so a reader can see which line moved when a figure changes.
9. Warehouse days — Warehouse days are counted from the date an item is booked in to the date the parcel leaves, and the free window inside that span is shown as a count of days rather than as a promise. A parcel held over a holiday week is recorded as held, with the holiday named, because that distinction is what makes the timing log useful in the following year.
10. Known biases — Four biases are worth naming. The sample is small at 24 parcels and self-selected, since only parcels we sent are counted. One destination country means the UK findings do not transfer to other markets. The window contains a national holiday, which inflates the slowest runs. And warehouse-day conventions differ between agents, so a like-for-like comparison across agents carries more uncertainty than a comparison across lines.
11. Known biases in the source data — A separate set of limits belongs to the upstream listings rather than to our own log. Published weights are sometimes absent, sometimes stated for one size only, and occasionally differ between the listing text and the packed reality. Style counts and size ladders move without notice as a seller edits a listing, so a recorded span is a snapshot with a date attached, not a permanent property of the item.
12. What we do not do — We do not publish promotion codes, because a code value would be stale before the page was indexed. We do not score, rate or review any agent, because a score would be an opinion rather than a measurement. We do not track availability or listing status, so an item that has disappeared from the source is simply absent from the record. We do not estimate a fee we have not re-checked, and we do not carry a figure forward on the assumption that nothing changed.
13. Review cycle — The rate table is re-checked weekly on Thursdays, the transit log is appended after each delivery rather than on a schedule, the QC protocol is reviewed once a quarter against new readings, and the seller records are re-checked quarterly for whether each object page still exists at the guide target. A page that shows a reviewed date older than a quarter is due for a sweep, and the [changelog](/changelog/) records when each sweep happened.
14. Reproducing a figure — Any published figure can be rebuilt from the page that carries it. Take the recorded inputs, apply the formula in step 5 or step 6 exactly as written, convert units with the constants in step 7, and compare the result with the number on the page. If the two disagree, the page is wrong and we want to hear about it at corrections@joyagoo.mobi, with the entry and the arithmetic you used.
Index scope: the record publishes 85 entries and the index carries 30 of them. Selection is mechanical rather than editorial — at least three entries per category, then the longest write-ups, with ties broken by the most recent check date. The remaining 55 are marked noindex, follow: a crawler may still read and follow them, they simply do not compete for an index slot. The rule is applied by script, not by hand, so it can be re-run and audited.
Why the other 55 stay published at all: hiding a record because it is short would make the list look tidier than the underlying data. Keeping them visible, with the word count shown on each row, is the honest version. If a short entry later earns a longer write-up, it enters the index on the next build without anyone deciding anything.