Store signals you can check before the link is paid

Three link shapes, and how to tell which one you hold

A row in an agent sheet looks uniform: a picture, a name, a marketplace tag and a row number. Under that surface there are three different situations, and each one asks for a different check before money moves. We logged 192 rows across eight pages of the Joyagoo channel list on 2026-10-06, and all 192 resolved to the same marketplace, Weidian. What changes from row to row is whether the title names a brand, and whether a store page behind the link still opens.

Sort the row into one of three shapes before you read anything else. Shape one: the title names a brand and the store behind the link resolves. Shape two: the title names a brand and the store page does not resolve. Shape three: the title names no brand at all. The sort is not a verdict on quality, and it settles one thing only: which piece of evidence you can still test before payment, and which one you will have to accept unverified.

Take the sample as a rough prior on how the shapes are distributed. A brand string sits inside the first 24 characters of 136 titles, which is 70.8%, and 56 titles, 29.2%, name no brand anywhere. Short titles are the norm rather than the exception: the median sampled title runs to 24 characters and four words. A bare-looking row is therefore in a large minority, not an anomaly.

One limit applies to all three branches. No sampled row carries a store field. The ten parsed fields are slug, id, name, price, imagePath, rank, source, category, brand and variants, and none of them names a seller. The sheet cannot tell you whether two rows came from the same store, so the store test belongs on the item page and never on the list.

Branch A: the title names a brand and the store resolves

Branch A hands you two independent pieces of text: a brand token the title claims, and a store page that still opens. Check them separately, because they answer different questions. The title answers what the seller says the item is. The store page answers whether anyone is still trading under that link. Neither answer covers the other.

Ask the item page for three things and write down what you see. First, the store name exactly as the page prints it. Second, the product noun next to the brand token, so you can confirm that the title and the page describe the same object. Third, the date you looked, because a page that opens today may not open next week and your note is the only record that it once did.

Then test the brand string itself. Search the brand with the product noun and see whether the same row can be found outside this sheet. A brand token that appears only inside one sheet is a keyword rather than a product line. None of the 192 sampled titles uses the words shop, store, seller, factory, official or original, so the store is never introduced by the title; that information comes from the page or from nowhere. If that search returns nothing at all, the row behaves like branch B in practice, because the claim can no longer be tested outside the sheet.

When all three checks pass, keep the evidence beside the row. A dated screenshot, the store name in plain text and the ten-digit row id are enough to rebuild what you were looking at if the link dies after payment. Keep those three even when the order goes smoothly, because the moment they are needed is the moment the page is gone.

Branch B: the title names a brand and the store is gone

Branch B is the uncomfortable one. The title makes a brand claim and the page that would support it no longer opens. Read the missing page as a missing record rather than as proof that the seller vanished. The source site is a directory that refers visitors onward, its footer states that it does not sell and does not take payment, and two of its own paths, the spreadsheet path and the finds path, return 404. Dead paths on a directory say nothing about a store.

Treat the brand string as unverified and try to replace the missing evidence with something you can still open. Search the brand with the product noun on the marketplace itself. If a store page appears with the same product, you hold a candidate rather than a confirmation, and the candidate has to pass the specification check before it earns the brand claim in your notes. Write the candidate id and the date beside the original row, so the two records can be compared later without repeating the search.

Decide in advance what you will do when nothing replaces the page: pay without the evidence, or drop the row. Both are legitimate outcomes, and only one of them leaves you able to answer a later question about what you bought. If you pay, write the brand as claimed in the title, mark it unverified, and keep the phrase next to the row until the parcel is in hand.

Ask the operator one question in writing before paying: which store will receive this order, and can that store page be opened from the account the order will be placed on. A written answer is checkable later. A chat line that scrolls out of view is not, and it is the version you will be trying to remember in three weeks.

Branch C: the title names no brand

Branch C covers 56 of the 192 sampled rows, close to three rows in ten. The row gives you a product noun and very little else, and the absence of a brand token is itself information: nothing in the title claims a trademark, so there is no brand claim to test. What you lose is search leverage, because a generic noun returns a wide field of candidates.

Move the anchor from the brand to the row id. Every sampled slug ends with the same shape, a hyphen, a ten-digit id and the suffix kox, and all 192 ids are unique, so the id is the one part of the row that points at a single listing. Save the id with the title, because three titles in the sample appear twice and a title-only search returns more than one candidate for those.

Read the item page for the three facts the row never carries: what the item is made of, what it measures, and how many pieces sit in the parcel. Across the sample, 181 of 192 titles name neither a material nor a measurement, so those facts will not be found in the sheet, whatever branch you are in.

A generic title is also the cheapest place to catch a mismatch. When the store page prints a different noun from the row, the row is a copy of a copy and the page is the better source. Write the page wording down word for word, because a paraphrase drops the exact term you would need if the order is later disputed. The same habit protects a later declaration, where the noun printed on the page is the noun you will be asked to describe.

What a wrong branch costs

Running branch A checks on a branch C row wastes the check and produces a false negative. You search for a brand that was never in the row, find nothing, and conclude that the listing is gone while it is open one page away. Running branch C checks on a branch A row does the opposite: you treat a testable brand claim as background noise and pay without testing it.

The expensive mistake is paying on a branch B row with branch A confidence. Once the store page is gone and the payment is made, the brand claim rests on a title you cannot reopen. Nothing in our sample measures how often that happens, and the sample does show something adjacent and useful: because no row carries a store name, the loss cannot be repaired from the sheet afterwards.

Judge the branch before the payment rather than after it. Write the branch letter next to the row, date it, and keep the file until the parcel is in hand. The protocol page sets out how a row is read and where our reading stops, and the channel index lists the eleven channels the source site publishes, five of which carry no self-description at all. Two of the three branches are settled by evidence you can collect in five minutes; the third is settled by whether you are willing to pay without it.