How do you get Guazi (瓜子二手车) used-car listings as JSON, in yuan and in dollars?

The city filter does not mean what it looks like. Guazi sells nationwide into each city market, so a page scoped to Beijing legitimately lists cars sitting in Shenyang, and one scoped to Shanghai returned cars from nine separate cities. Scrape it believing you asked for a location and you will build a geographic analysis out of rows that are not geographic, with nothing in the markup to warn you.

The detail page disagrees with itself about the city. The structured data in the page head gave Shenzhen for a car that is in Suzhou — and guazi's own plain-text mirror of the same page says Suzhou. One field is the market you browsed in, the other is where the car is. A scraper that reads the obvious one stores the wrong city on every record.

The unit is a ten-thousand-fold trap. China quotes used cars in 万元 and distance in 万公里, so a listing reading 14.43 and 10.94 means 144,300 yuan and 109,400 kilometres. Take the numbers as they appear and drop them next to a dollar auction feed and everything is wrong by four orders of magnitude. Guazi's own price bands are published the same way: the band labelled 13 to 18 means 130,000 to 180,000 yuan.

There is no VIN on the domestic market, and that is not a wall. The seventeen-character number appears nowhere on a domestic listing under any label, Chinese or English. Guazi's export marketplace does publish one, but masked to the first three and last four characters, with the middle absent from the page entirely. If your pipeline joins vehicles by VIN, the domestic Chinese market cannot be joined that way at all.

The paging stops long before the count does. A city page reports forty-three thousand cars and serves forty a page to page twenty, then nothing — eight hundred rows out of forty-three thousand. The pager does not admit this, and the 'showing 1 to 40 of N' line on the page never changes as you move through the pages, so a scraper that reads it concludes it is still on page one forever.

On the export marketplace the price field means two things. We measured one row publishing thirty dollars for a 2023 Hyundai Palisade with 39,900 km, while the cheapest car on every other facet that minute was between three and thirty-four thousand. It was not a parse error and it was not a thirty-dollar car: the row was a live auction and the field carried the current bid. One field, two meanings, and nothing in its name to tell them apart.

Both marketplaces are one engine, with ten actions. Domestically: search, detail, brands, series, cities, price_bands and listing_ids. On the export side: export_search, export_detail and export_brands. The domestic calls are cheap and the export calls cost more because the source is heavier there; the split is deliberate so that a domestic price sweep does not pay export rates.

The city question is answered rather than hidden. The response flags that the city you passed is a market, and every row carries its own city — the place that car actually is. On the detail record the car's real location comes back as city and the browsing market as market_city, so the contradiction in guazi's own page cannot land in your database as a single wrong value.

Every price and distance is absolute. price_cny is yuan, mileage_km is kilometres, new_car_price_cny is the original list price where guazi shows one, and the price_bands action returns each of the sixteen bands with its real yuan floor and ceiling rather than the 万 figure. Nothing in the payload requires you to know what 万 means.

The condition disclosures are the point. A domestic detail record carries insurance_claim_count, transfer_count, a condition grade, an appearance score out of a hundred and whether an official two-hundred-point inspection report exists. An export record states no-accident, no-flood and no-fire as three separate flags next to an A-to-D grade. Both came back on every car we sampled, and a field guazi does not publish stays null rather than defaulting to something reassuring.

vin is null on the domestic record with a flag saying the source does not publish it. On the export record vin_masked carries exactly what guazi shows — three characters, a mask, four characters — and nothing is reverse-engineered into a full-looking VIN. What you can join on domestically is the listing id, which round-trips unchanged between search and detail, as a string, with its c prefix intact.

The export price is split by meaning. price_usd is an FOB asking price, current_bid_usd is a live auction bid, and price_kind names which one the row carries — so a repricing model cannot read an opening bid as a market value. The export row also carries guazi's own yuan-per-dollar rate, which is what lets you reconcile the two marketplaces instead of guessing at a conversion.

The paging ceiling is published, not discovered. The response reports the page you got, the real maximum and the source's own headline count, and a request for page ninety-nine comes back marked as clamped rather than silently serving page twenty. On the export side the response states outright that no second page is available, because the source's own filter and page links do not work without running its JavaScript — so the engine exposes none of those parameters rather than accepting one it knows will be dropped.

There is a bulk id feed. Guazi publishes a crawler index of every live domestic listing — thirteen pages of roughly ten thousand ids with a last-updated date on each, about 127,000 cars. listing_ids returns a page of that, priced by the rows it returns, which makes a nightly diff affordable: compare dates, then call detail only on what moved. No guazi account, login or token is involved anywhere in the engine.