Build1 publisher3 min readPublished
Delivery dates only work calculated backwards, with a six-week Chinese New Year shutdown as one factor among several
A wholesaler shipping 385,315 listings from 5,233 Chinese factories to 62 countries has published how it computes a promised date. Transport mode, stock depth and the factory calendar each move the answer by weeks.
The Engineer · Build desk
What happened
- A Chinese wholesaler runs 385,315 listings from 5,233 factories and ships to 62 countries, and every product page has to answer when the goods will reach the buyer's shelf.
- Chinese New Year 2027 falls on 6 February, with an official holiday of about a week and, by the wholesaler's account, a real shutdown of roughly six weeks around it.
- The post computes dates backwards from the shelf date to the order date, counting each block separately, and says only that direction works.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint There is nothing in an empty rail cell to backfill or average into a regional mean, and the post's warning is that a UI treating it as unknown will eventually offer a train to Brazil.
- capability Because the calendar can exhaust a season instead of sliding a date, the system can distinguish the two answers a buyer needs: a later date, or none this season.
- decision The wholesaler argues the cheapest improvement available to a small importer is to order six weeks earlier, ahead of negotiating unit price, and that it costs nothing.
- cost A week lost to Golden Week, 1 to 7 October, cannot be recovered, because it falls when every hour of fourth-quarter capacity is already booked.
The calendar function walks the date forward one day at a time and decrements the remaining day count only when that day falls outside a blackout window [15]. A shutdown therefore does not push a promise back by its own length. It can exhaust the season, and the wholesaler's stated reason for writing it that way is that a customer needs one of two different answers: later, or not at all this season [15].
Add the backward blocks at both ends. Marketplace check-in runs up to 14 days, freight 2 to 45, consolidation and QC 3 to 5, production 0 to 45, and calendar blackout 0 to 42 [16]. Best case totals 5 days and worst case 151, roughly a factor of 30 [2]. The example the post gives of the model it replaced is a lookup returning the string "18-40" [20].
Sixty-two destinations by three modes gives 186 cells, and only 153 of them can ever hold a figure: air and sea for all 62 destinations, rail for 29 [1]. The other 33 are empty because rail needs continuous track from China [3]. Every European destination has a rail number, Iran is the only one in the Middle East, and the Americas, Africa and the island parts of Asia-Pacific have none [3]. The availability list is built to match, with a spread that appends the third mode only when the lane has one: `const dostupno = ['air', 'sea', ...(lane.rail ? ['rail'] : [])];` [5].
Across modes to the same country, the post puts the gap at a factor of five [7]. Within a single mode it can be wider: sea into Asia-Pacific runs from 2 days to 30, a factor of 15, though that spread is across destinations and not a comparison at one of them [8][5]. The largest term in the equation, by the wholesaler's account, is whether the quantity is in stock, since a stocked carton ships with zero production days and anything past available stock becomes a 15 to 30 day order, stretching to 45 in peak season [11][12].
These are one company's figures, taken from a 6,000-item sample of its own live catalogue: a median price of 11.04 yuan, 47 percent of items under 10 yuan, and a median MOQ of 72 pieces with common cartons at 24, 36, 48, 72, 96 and 120 [18]. Three conditions have to hold before the model transfers. Cartons are never broken, so a quantity is a multiple of the MOQ or the order is refused [10]. About half the lines are stock and half are made to order [9]. And production sits behind the Chinese factory calendar [13].
The six-week shutdown is the wholesaler's own estimate, and it gives a cause: workers leave two to three weeks before the 6 February 2027 holiday, and because a meaningful share do not return to the same employer, full output takes three to four weeks to come back [13]. Six weeks is 42 days, the upper bound of the blackout block in the backwards count [4]. The sample under all of it is 6,000 listings out of 385,315, about 1.6 percent [3].
What to watch
- Whether the wholesaler publishes the per-lane table itself, so the factor-of-five mode gap can be checked destination by destination.
- Whether the 2027 wind-down starts when the model assumes it does, two to three weeks before 6 February.
- Whether anyone outside this catalogue measures post-holiday worker return rates, since the three-to-four-week recovery term rests on that figure.