The case that decides the threshold: two listings close to the line. Red bits are the ones that differ.
Measured 2026-10-23
Every number, with its method.
A figure you cannot check is a claim. Everything here is read from the live index or the held-out test, and says how it was produced, including what could not be produced at all.
Live from the indexBuilt 2026-10-23
- 376Products indexed387 listings, merged by photo growing
- 387Listings readSeller pages, fingerprinted growing
- 74,691Photo pairs comparedEvery photo against every other growing
- 9Sold by more than one sellerOf 376 products growing
- 1Catalogue connectedOf the 6 networks on the board growing
01 · The index
What is in here.
A product is one physical thing; a listing is one seller’s page for it. 387 listings collapse into 376 products. That is the whole idea in one sentence.
376ProductsListings whose photos are within the threshold are merged into one product.
11Listings collapsedFound to be another listing of something already indexed.
9Sold by more than one sellerProducts with two or more listings after merging.
105CategoriesAs the supplier assigns them. Coarse, and not something we rank on.
02 · The measurement
Every pair, by distance.
Every product photo compared with every other, 74,691 pairs, by how many of 64 bits differ. Unrelated photos pile up near 32; the same photo sold twice sits near zero. Drag the line.
10 bits
Matches0 bits apart163248
14Pairs merged as one product
12 bitsClosest pair we keep apart
10 bitsWhere we draw the line
Safe. Every pair merged here is the same photograph.
74,691 pairs, every one in the index, on a log scale. Tested on 60 photos we held back: 98% of true matches found, and no false matches at any line up to 14 bits.
7Pairs between 11 and 15 bitsWhere the method works hardest. None were merged.
74,670Pairs judged differentEverything over the line. 96% of all pairs sit between 24 and 40 bits.
03 · Accuracy
98% found. Zero false matches.
Measured on 60 real product photos, separately from the index. Same photo after the resizing, recompression and padding a marketplace applies, against different products from the same search, the hardest fair test.
Calibration run60 photos · 1,762 negative pairs
98%of true matches found
Threshold
False matches
0
The line we use. Clean, with room to spare before the first false match.
0 bits apart8162432
The 2% that are missed are heavy edits: tiny thumbnails and thick padding. They are missed rather than mismatched, which is the safe direction. Why the line is at 10 →
04 · Ship-from
Where things ship from.
Declared by each seller, per listing. A product appears under every origin any of its sellers declares, so these add up to more than the index.
- China → the US315 · median $9.97
- China230 · median $6.13
- the US109 · median $58.90
- the UK57 · median $54.29
- Germany6 · median $24.64
- Canada1 · median $51.99
05 · Coverage
Coverage by category.
The 24 largest of 105 categories, as the supplier assigns them. Shown as data: on Vesim you ask for things rather than browse a category tree.
- Home Office Storage54
- Furniture52
- Body Care19
- Silicone Cases12
- Men's Backpacks11
- Totes11
- Facial Care11
- Christmas Decoration Supplies9
- Pet Nests9
- Fitness & Bodybuilding8
- Garden Tools7
- Suits & Sets7
- Action & Toy Figures6
- Belts & Cummerbunds6
- Sports Bags5
- Night Lights5
- Skiing & Snowboarding4
- Camping & Hiking4
- Kitchen Storage4
- Other Replacement Parts4
- Smart Watches4
- Home Improvement Materials4
- Storage Bags & Cases & Boxes4
- Fashion Backpacks3
06 · Limits
What cannot be measured yet.
Each of these is reported as not measured, everywhere on the site, rather than estimated.
- Not measured
Breadth
1 store connected (cj). Until a second connects, the spreads shown are within one seller’s own catalogue, which is why they are small.
- Not measured
Delivery dates
No source returns them, so arrival is reported as not measured rather than estimated.
- Not measured
Price history
The index is one snapshot. History cannot be backfilled; it starts when daily snapshots do.
- Not measured
Discount codes
Not available from any source yet.
Now try it
Now try it on anything.
Describe it. We’ll show every price, which one fits, and tell you when nothing does.