Open data · snapshot 2026-10-26

The whole index, as two files

Every number this site prints about prices comes from one place: 307 rows read from the catalogue at build time. This page hands that file over, so the claims can be checked instead of trusted — and so anyone building their own sheet starts from the same rows we do.

Two downloads, both stamped 2026-10-26: a JSON file with a typed header and a CSV for spreadsheets. 307 rows, 12 columns, 133 KB and 68 KB respectively. Licence CC BY 4.0. No sign-up, no email capture, no rate limit worth worrying about.

307rows in the snapshot

12columns, all documented below

133 KBJSON · 68 KB CSV

CC BY 4.0reuse, including commercially

Download

Same rows, two shapes. The JSON carries the column dictionary and the method inside the file; the CSV is the bare table, UTF-8, comma separated, quoting only where it has to.

kakobuy-rows.json · 133 KB kakobuy-rows.csv · 68 KB

What is in the file

Twelve columns, all taken verbatim from the catalogue except for two derived fields: price_usd is rounded from the raw float, and image_url is assembled from the platform, the item id and the image index. Nothing is normalised, translated or cleaned.

Column	What it actually is
`spu_no` · string	Catalogue row id. Stable within a snapshot, can change between them.
`name` · string	Listing title as the agent shows it, unedited.
`category` · string	Shelf from the source catalogue. Two known mis-files, listed on this page.
`brand` · string	Brand field, often empty. Empty is not "unbranded".
`platform` · string	Source marketplace. Every row in this snapshot is WEIDIAN.
`item_id` · string	Marketplace item id — the part of the Weidian URL that matters.
`price_usd` · number	Agent price in USD, unrounded, as served on the snapshot date.
`price_cny` · number	Same row in CNY. The ratio implies the FX rate of the day, not a fee.
`created_at` · string	Listing date. No timezone, no update timestamp.
`popular` · boolean	Popularity flag exactly as the catalogue serves it — a boolean, not a count. Do not sum it.
`image_url` · string	Absolute URL to the agent-hosted WebP. Hotlinking is not encouraged.
`original_url` · string	Source listing URL. 404s when a seller delists.

How it was built

The public catalogue endpoint is read once per build; no scraping loop, no paginated crawl.
Rows are keyed by spuNo. Duplicates collapse onto the newest read; nothing is merged by title.
Prices are written as served, in both currencies. The USD/CNY ratio in the file is therefore the day's rate, not a fee schedule.
Image URLs are assembled deterministically from platform, item id and image index — which is why they can be rebuilt, and why hotlinking them will break without warning.
No manual edits. The two known category mis-files stay in, and are named on this page.

What is not in the file

Five things people assume a shopping-agent dataset contains, and this one does not: weights, sizes, stock, QC photos and agent fees. That absence is the honest state of the source, not an oversight.

Weights

No weight field exists in the catalogue feed, so freight per row is an estimate no matter who prints it. The freight floor on the budget page uses a stated average, not a per-row weight.

Sizes

Also absent. A size chart built from this file would be invented arithmetic; the QC workflow explains how to measure a garment against a real chart instead.

Stock and live price

In stock when read, maybe not when you click. Sellers delist and reprice constantly; original_url 404s are normal, not a data error.

QC photos

Those exist only after a parcel reaches the warehouse, for that parcel. The image_url field is the seller's listing photo and never a QC set.

Three rows the catalogue files wrong

A $0.18 football tattoo sticker is filed under Shoes, a $0.48 bracelet under Hoodies & sweaters, and a $1.02 placeholder under Jackets. They stay in the file, unchanged, because a dataset that quietly re-sorts itself is not evidence of anything.

Practical consequence: any per-shelf median you compute here needs a row count next to it. The price index prints those counts for exactly this reason, and flags the shelves where a single row moves the median.

Licence and how to cite it

CC BY 4.0. Copy it, chart it, put it behind a paywall — just keep the attribution. The individual prices remain the sellers' and the agent's; the compilation is ours.

Kakobuy Spreadsheet 2027 — Independent Index (2026-10-26).
Kakobuy row dataset, snapshot 2026-10-26, 307 rows.
https://kakobuyspreadsheet2027.lat/kakobuy-dataset/

Why publish at all

Because every competing sheet asks you to trust a number it will not show you. Publishing the file turns “median row $35.27” from a claim into arithmetic anyone can redo in ten seconds — and it makes the two mis-filed rows findable by the people who would otherwise discover them and conclude the whole index was padded.

It cuts both ways: if the rows turn out to be worse than advertised, the file says so before a reader does.

Questions about the files

Can I use these rows in my own spreadsheet or site?

Yes, under CC BY 4.0 — keep the attribution line printed on this page somewhere visible. The prices belong to the sellers and the agent, not to this site; the compilation does.

Why is there no weight or size column?

Because the source does not publish them. Weights are what determine freight, and the catalogue has never carried them, so any site quoting a per-row shipping cost is estimating. This file would rather have twelve honest columns than thirteen with one invented.

How often is the snapshot replaced?

Whenever the index is rebuilt. Each file is stamped with its snapshot date (currently 2026-10-26) in the JSON header, so two downloads can be told apart without diffing every row.

The price in the file is not what I see on the agent today. Why?

Sellers move prices without notice and the agent mirrors them. Everything here is the price on the snapshot date. Use the file for structure and distribution, not as a live quote.

Are the two wrong categories going to be fixed?

No — they are kept as served. Editing them would make the file disagree with the index it is supposed to verify, and the disagreement is the interesting part.