The sampling frame
The frame came from two places at once. Sheet pages anyone can open gave up wrapped ACBuy addresses; a maintained directory gave up plain marketplace addresses. A collection pool of 194 candidate addresses fed that frame, and only the rows that completed a reading were published.
Rows that never completed were left out rather than written down as failures. That choice keeps the table honest about what was actually seen, and it keeps a rebuild rate from being dragged down by work nobody finished. An unfinished check is not a broken link, and the two should not share a column.
Requests ran one at a time with a gap between them, under an 8-second timeout, with no cookies and no account. The checking method page states that setting in full. A reading you reproduce yourself should look the same as ours, which is the point of writing it down.
What the sample contained
The published sample split into two groups with different histories. Wrapped addresses pulled off sheet pages had never been cleaned by anyone. Directory addresses came from a list that prunes what it can no longer reach. Same listings underneath, different maintenance above.
Sample 2026W39-031 carries item 7579044010. The row holds the item ID in one cell, a wrapped address with a source marker beside it, and the plain marketplace address behind both. Three pieces of one listing, written down in the order they were found.
Every record that survived the frame traced to Weidian, which is a property of the addresses collected rather than a claim about the market. The wrapped addresses available to collect all carried the same source marker, so the frame can describe only what that marker points at.
The rebuild rate we measured
Thirty-one wrapped addresses went in. Thirty-one working source addresses came out. Same identifiers, same listings, one layer removed. A rate that clean on a base that size is worth stating exactly, and worth refusing to round up into a promise.
That rate belongs to one marker. Every wrapped address in the period carried the same one, so the measurement covers a single route between a wrapper and a marketplace. Other markers exist; this site holds no measured example of any of them, so no rate is published for them.
The directory group needed no rebuild at all. Sixty-two marketplace addresses, already plain, already answering, with no wrapper in front of them. Side by side, the split says something simple: rebuilding is a repair for wrapped rows, not a step that every row needs.
What that rate does not prove
A perfect rate is a statement about 31 addresses on one day. It does not promise that a wrapped link you hold right now will rebuild, and it says nothing about whether the listing behind it is still trading, still stocked, or fairly priced.
Nothing in the rebuild step inspects a seller, a photograph or a size chart. A rebuilt address that answers is a page that answered, and reading quality into that response asks the check to do work it was never built for.
Rebuild the rows you care about in small batches, and write the date beside each result. The legacy link resolver handles the address side of the job, and the checking method page explains what each answer means. Then run the batch again later: a reading has a shelf life, and the address is the part that keeps.