The shape of a wrapped link
Group A holds thirty-one wrapped addresses copied from publicly visible sheet pages, and every one of them shares a single shape: a host, a path, then a short run of named fields after a question mark. Nothing in that collection varied in structure. The variation sat entirely inside the fields, in the numbers and codes they carry. That is why a rebuild is a matter of reading position rather than hunting for a replacement page.
A wrapper is not a redirect dressed up as one. It is a standalone address that a platform issued so its own pages stayed in the middle of a purchase, and when the issuing platform goes dark the address stops resolving while the product it names has not moved an inch. That split is the whole subject here. It is also why a sheet full of dead entries can still be worth keeping.
Thirty-one is a small number and small numbers deserve caution. Those addresses came from pages that were already public and already indexed, which tilts the sample toward sheets that survived long enough to be copied by somebody else. A private sheet, a sheet behind a login, or a sheet its owner deleted after the shutdown leaves no trace in the 2026-W39 window at all.
Two things are worth doing before anything else. Copy the wrapped addresses into a plain text file, one per line, and leave the sheet untouched until the rebuild has been read through. Marking up the original burns the only clean copy you have. The legacy link resolver takes a pasted list and shows what each entry rebuilds into, which beats editing cells one at a time and squinting at the result.
What each segment carries
The first field after the host holds the product number, and in the observed format it is the same number that sits at the end of the source address. Sample 2026W39-014 carries 7577045115 in that position; the rebuilt source address carries the same digits, in the same order, with nothing added. That one-to-one match is what makes a rebuild possible without visiting the platform that issued the wrapper in the first place.
The second field names the source market. Across all thirty-one Group A records the value was WD, and WD points at Weidian, which is the entire basis for the single market shown in this period. Other values may well exist and almost certainly do. This site has no observed sample of them, so nothing written here assumes where a different marker would lead.
A third field trails the first two and holds a short code. None of the thirty-one rebuilds needed it, and this site does not publish a meaning for it, so treat it as opaque. Leave it out when you assemble a source address. The address that answered in this window was built from the product number and the market marker alone; the extra part was never part of the tested result, in either direction.
Field order matters more than field values. Swap the product number and the market marker and you get an address that looks entirely plausible and answers nothing at all. Thirty-one records is enough to confirm the order for this one format and nowhere near enough to confirm it for formats this site has never seen. Read the order off a working example before you trust a pattern you noticed.
Turning the wrapper back into a source address
The rebuild is three moves and no lookup. Read the product number out of the wrapper, read the market marker sitting next to it, then write both into the source address that the marker names. Keep the number exactly as it appears, including any long run of leading digits that looks like it could be trimmed. Nothing about this needs an account, a session, or a visit to the platform that issued the wrapper.
Doing it by hand holds up for five or six entries and falls apart at fifty. The failure is not arithmetic, it is attention: digits get transposed, a marker gets dropped from one row, and the address that comes out still looks correct to anybody scanning it. Paste the whole column into the legacy link resolver and check the rebuilt list against the sheet row by row before you rely on any of it.
Keep the rebuilt address beside its row rather than in place of it. The wrapper records what you were actually handed by the platform; the rebuilt address is a working substitute that depends on the source listing staying reachable. Sheets that come through a shutdown usually come through because somebody kept both columns and could still tell, months later, which of the two had answered on which day.
Batch the work in three passes. One pass extracts product numbers, one pairs each number with its market marker, one assembles and checks. Mixing the passes is where mistakes hide, because assembly feels like progress while checking feels like delay. Sixty-two directory addresses were read in the same window as Group A, under the same ordering discipline, and they were pasted and checked in exactly the same way.
Confirming the rebuild landed
An address that opens is not automatically the address you wanted, so run three checks. The product number in the source address has to match the number inside the wrapper. The page that loads has to show the listing you were expecting. The host that answered has to be the one the market marker named. If any of the three disagrees, mark the row unconfirmed and move on.
Thirty-one rebuilt addresses returned a normal response in the 2026-W39 window and all thirty-one landed on the host their marker named. That is a clean result and a narrow one at the same time: it says the rebuild was right for these thirty-one rows, on that date, under a plain check that used no credentials and stopped waiting after a fixed interval. It says nothing about next month.
Write the date next to every confirmed row, because a confirmation is a snapshot and not a state. The weekly snapshots page keeps the same discipline for the public sample: each record names the day it was read, and two weeks can only be compared when both carry a day. Adding dates afterwards, from memory, is guesswork wearing the clothes of data.
One more habit saves arguments later. Write down what answered, not what you concluded from it. "Answered on the source host" is a fact that will still hold up in six months. "Item is fine" is a claim about a seller, a warehouse and a listing all at once, and not one of those three was measured. The checking method page draws the same boundary around the public sample.
When the rebuild cannot be done
Some wrappers carry no market marker this site has ever seen, and those are not rebuildable here. Leave them alone instead of reaching for the one marker that happens to be documented. A guessed rebuild produces a working address for a different listing, which is worse than no address at all, because it looks like a result and gets copied into the next version of the sheet.
A wrapper can also rebuild correctly and still fail. The source listing may be gone, the host may refuse the request, or a check may return a wall instead of a page. Those outcomes mean different things and should not be flattened into one failure column. A refusal during collection is not evidence that a listing disappeared; one hundred and thirteen directory addresses were left out of the table in this period for exactly that reason.
Then there is the wrapper whose platform no longer exists at all. ACBuy stopped taking orders inside the window described on the Kakobuy page in the platform ledger, and a wrapper issued by a platform in that position has no live service behind it. The wrapper does not come back. What can come back is the source address inside it, on the day you check, and only for as long as the source host keeps answering.
Where nothing can be confirmed, say so on the row itself. A blank cell gets filled in by whoever reads it next with a guess, and the guess outlives the doubt. Writing not rebuildable here, checked on the date, costs one line and stops the same row from being reopened every time somebody new picks up the sheet. A note that names the day and the reason is worth more than a colour.