How many titles actually name a brand
Three things this piece settles
- 136 of 192 sampled titles yield a brand read verbatim, and 56 yield none.
- Every brand read starts within the first 12 characters, so a 24-character window never binds.
- Per-page counts swing from 14 to 21 of 24, so treat 70.8 percent as a range.
Which stop this touches
8 stops · open a stop to read its windows · “n open” = actions still available
- Stop 1: Pick the channel — 3 actions still open. A channel is a filtered view of one agent, not a shop. Learn what its row count proves and what it cannot.
- Stop 2: Hand the link over — 3 actions still open. A link carries an identity, not a product. Keep the identity intact and the order stays traceable.
- Stop 3: Pay for the item — 2 actions still open. Item payment and freight payment are two separate moments. Most closed doors close at the first one.
- Stop 4: The warehouse clock — 3 actions still open. Arrival starts a clock. Verify the window on your own account before you plan around it.
- Stop 5: QC photos — 3 actions still open. A photo set is evidence of coverage, not a verdict on quality. Ask for angles, not for opinions.
- Stop 6: Merge and pack — 3 actions still open. Merging trades a second freight charge for a bigger volumetric weight. Run both sides.
- Stop 7: Freight and customs — 2 actions still open. Freight is charged on chargeable weight. Customs is decided on declaration fields you fill in.
- Stop 8: Checking the parcel — 2 actions still open. Delivery starts the claim clock. Photograph the unopened parcel before you open it.
How the sample was taken
Log the list rather than screenshotting it, because a screenshot cannot be counted. We opened the joyagoo channel list and registered every row of eight consecutive pages by hand. Each row went into one fixed field set rather than a free note. The entire capture was done in a single sitting on one day. Anything not visible on the list page was left out of the sample entirely.
The fields were slug, numeric id, title as written, list price as an internal sort key, thumbnail path, position on the page, source marketplace, our own category group, the brand read from the title, and a flag for titles declaring more than one style. That is ten fields per row and 1,920 recorded values across the sample. Keeping the set fixed is what makes the count reproducible at all. Adding a field halfway through would have invalidated every earlier row and forced a full recapture. The field list is part of the method rather than an implementation detail to be tidied away later.
Two of those fields are judgements rather than observations, and both are marked as ours. Category is assigned from keywords in the title text. Brand is recorded only when the title spells it out plainly. Anyone repeating the count should expect those two fields to differ from ours. The other eight fields should match exactly if the capture is repeated honestly.
Nothing was inferred at capture time, not even in obvious cases. Where a title names a line or a nickname rather than a house, the brand field stays empty. That rule holds even when a reader would recognise the product instantly. The parser records and does not guess at intent. A guess would have raised the headline figure and lowered its value.
All eight pages were captured in one sitting on 2026-10-06. A later page renumbering could not silently shift which rows sit inside the sample. The snapshot is dated for that reason rather than to look current. Re-running the count against a fresh capture will produce a different list. The date is part of the method rather than decoration on the top of the page.
The sample: 192 rows across 8 pages
Eight pages at 24 rows each gave 192 rows and 192 distinct numeric ids. Three titles appear twice inside the sample, so it holds only 189 unique title strings. The duplicates are Godspeed Tee, Dior B22 Sneakers and Nike Air Jordan 4 Sneakers. Each duplicate turned out to be a separate listing rather than a repeated row. Separating the two cases is why the numeric id was recorded alongside the title.
Every row came from Weidian without a single exception. No other marketplace appears anywhere in the sample, not even once. The source was determined from the image naming prefix attached to each row. That makes the sample single-market rather than mixed across marketplaces. Every rate computed from it inherits that single-market limit whether or not the rate is quoted.
A brand was read from 136 of the 192 titles, which is 70.8 percent. The remaining 56 titles, 29.2 percent, produced no brand at all. Read that as a statement about what the titles contain rather than what the products are. Plenty of the 56 name a recognisable product without naming a house. The title and the product are two different objects.
Those 136 reads resolve to 51 distinct brands across the sample. Thirty-two of the brands repeat somewhere in the sample and nineteen appear exactly once. Concentration is mild rather than extreme in either direction, which keeps the average meaningful. The head of the distribution is not dominated by any one label, which keeps the count stable. That makes the sample more varied than a typical resale list of the same length, which helps the count travel a little further.
Nike is the most frequent single brand at 14 rows. That is 7.3 percent of the sample and 10.3 percent of the branded rows. Everything after it falls away fairly quickly. The tenth most frequent brand sits at four rows only. A long tail rather than a monopoly, which is good news for the count.
Three known biases
The first bias is position within the channel list. Eight pages out of 351 is 2.28 percent of the channel. The sample is therefore the head of the list and nothing below it at all. Any claim about the catalogue inherits that restriction whether or not it is stated. State the limit and the claim stays honest; leave it out and the claim changes meaning.
The second bias is source. All 192 rows are Weidian, so what is measured here is one marketplace naming habit. A different marketplace would very likely produce a different rate on the same day. The figure travels no further than the marketplace it came from. It should never be quoted as a platform-wide rate or as a rate for another channel.
The third bias is the parser, and it is the one most likely to be missed in a re-run. A brand counts only when the title spells it out explicitly. House lines and nicknames therefore land in the unbranded 56 by construction. That is a rule effect rather than an observation about the list. The 29.2 percent is partly a fact about naming and partly a fact about us.
Two smaller defects belong with the third one. Eight titles carry an HTML entity where an ampersand belongs, which breaks a literal string comparison. Two rows flagged as multi-style use bracket text that a styles-only pattern misses. Both defects are small, and both are recorded here rather than smoothed over in the write-up. Recorded defects can be corrected later and unrecorded ones cannot be corrected at all.
The 24-character window is not a bias, because it never binds. Every one of the 136 brand reads begins at character 0 to 12 of the title. A hundred and twenty-two of them begin at character 0. Widening the window from 13 characters to 24 would add exactly zero rows. Narrowing it would also change nothing at all until the window fell below 13 characters.
Where these numbers do and do not apply
Use 70.8 percent as a property of the head of one channel. Do not use it as a rate for the catalogue sitting behind that channel. The catalogue is roughly 8,424 rows and we read 192 of them. A rate taken from 2.3 percent of a population describes that 2.3 percent. The rest is unmeasured rather than equal, and that difference matters when the figure is reused.
Per-page counts ran 21, 15, 17, 19, 14, 16, 17 and 17 out of 24. The spread is 58.3 percent to 87.5 percent across pages captured minutes apart. Page-to-page variation is therefore wider than the headline figure suggests. A single blended number hides that variation entirely. Publishing the range alongside it costs nothing at all.
Category moves the rate a long way in both directions. Outerwear reads 19 of 20 and shoes read 9 of 10. Pants-shorts reads 16 of 27 and headwear reads 6 of 10. Comparing two categories on one blended rate would be a mistake. Compare within a category or not at all, and always say which category you compared.
Quote the method alongside the number, every single time. A figure of 70.8 percent without the eight-page limit is a different claim. Without the single-marketplace limit it is a different claim again. The limits are part of the number rather than a footnote to it. Move them and you have made a new claim without saying that you did.
Re-run the count before reusing any of these figures. The figure is dated 2026-10-06 and the list does not stand still. A re-run takes one sitting and produces a fresh number with its own date attached. Reusing an old figure without its date is the failure mode. The method is the durable part rather than the number that came out of it.
The data point behind this piece
136 of the 192 sampled titles, 70.8 percent, yield a brand read verbatim, and 56 titles, 29.2 percent, yield none. The 136 reads cover 51 distinct brands, of which 32 repeat and 19 appear once. Nike is the most frequent at 14 rows, or 7.3 percent of the sample.
Basis: Node parse of src/data/finds.ts on 2026-10-06; 192 rows registered from 8 list pages of the joyagoo channel; count of rows where brand !== null, then a distinct count of the brand strings.
All 136 brand reads begin at character 0 to 12 of the title and 122 begin at character 0, so a 24-character window excludes none of them; widening the window from 13 characters to 24 adds zero rows to the count.
Basis: Same 192-row parse; index of the recorded brand string inside the lowercased title, matched case-insensitively, maximum observed index 12.
Per-page brand counts in the sample run 21, 15, 17, 19, 14, 16, 17 and 17 out of 24, a spread of 58.3 percent to 87.5 percent across pages captured minutes apart.
Basis: Same parse, rows grouped by list rank into pages of 24 (ranks 1-24 through 169-192); count of rows where brand !== null in each group.
Twenty-six of the 192 rows, 13.5 percent, are flagged as multi-style; 24 of them carry a literal bracket of the form number plus styles and two use other bracket text. Brackets of some kind appear in 44 rows, or 22.9 percent.
Basis: Same parse; count of rows where variants === true, cross-checked against a regular expression for bracket text in the title.
This piece answers searches about “joyagoo spreadsheet data”. It is part of the rules and scope cluster.
Where to go next
Links to Kakox are tracked so the destination can attribute the visit. Nothing on this site is sold by us.
Reviewed 2026-10-06 · Weekly review · next review weekly review