How do you get che168 (二手车之家) used-car listings as JSON without getting blocked?

The unit is the trap, and it is a ten-thousand-fold one. China quotes used cars in 万元 — ten thousand yuan — and distance in 万公里. A car whose page says 151.8 and 1.3 costs 1,518,000 yuan and has done 13,000 km. Scrape the numbers off the page as they appear and drop them into a column beside a dollar auction feed and every figure you have is wrong by four orders of magnitude, with nothing in the data to flag it.

There is no VIN. This is the single biggest difference from the Korean and American markets, and it is not a scraping failure: che168 does not publish the seventeen-character number at all. What its page carries is a 32-character opaque vehicle token plus a 48-hex signature, used only to build its own claim-history and service-record links, and both of those sit behind a login. If your pipeline joins vehicles by VIN, this market cannot be joined that way.

A mistyped city does not fail. We asked for two cities that do not exist and both returned HTTP 200 with a full page of cars from all over the country — so a typo in a city slug hands you national data labelled as one city's, with no error anywhere. Separately, a city whose filtered result set is thin gets quietly padded with cars from neighbouring markets, which looks identical unless you check every row's own city id.

There is no result count to page against. We searched a complete half-megabyte result page for one. Every number-of-cars string on it turned out to belong to a script template, a sidebar shop's own inventory, or a tag tooltip. And paging is finite in a way the pager does not admit: pages 1, 2, 3, 20 and 99 of one city shared zero ids, but page 100, page 101 and page 150 returned the identical rows.

Every price and distance arrives in both units. price_wan_cny carries the source's own figure and price_cny carries the multiplication done; mileage_wan_km sits beside mileage_km. We publish both rather than choosing for you, because the 万 figure is what a Chinese dealer will quote back at you and the absolute one is what your database wants. new_car_price_cny carries the struck-through original list price where the listing shows one.

vin comes back as null with vin_published false, and the two opaque identifiers travel under names that say what they are: vehicle_token and vin_lookup_signature. Nothing is reverse-engineered into a plausible-looking VIN. What you can join on instead is brand_id, series_id, spec_id, first_registration and mileage_km — and spec_id is strong, because it names the exact factory trim rather than the model.

The trim sheet is the thing this market has that the others do not. Pass a search row's spec_id to model_specs and you get 71 named parameters out of Autohome's model database: trim name, manufacturer, the price when new, body structure, class, engine, gearbox, fuel, seat count and about fifteen grouped sections. A trim never changes, so it is cached for a week and costs one credit. car_options is the companion call for what that individual car was fitted with.

The city filter is checked for you. The engine compares the slug you sent against the market the source says it applied, and refuses the call with INVALID_PARAM when they differ — so the silent fallback to national results cannot reach your data. Where a valid city is padded with neighbouring stock, the response reports city_padding true and lists every city id present rather than hiding it.

total_reported is null with a note saying the source publishes no count, and pagination.pages_available carries the one real signal it does publish: the highest page its own pager offers. has_more is false at page 100, the measured ceiling. Nothing multiplies 56 by 100 into a total that would look authoritative and be invented.

Eleven actions share the engine: search, dealer_cars, car_detail, car_summary, car_photos, car_options, condition_report, model_specs, suggest, cities and brands. The cost shape matters here: search reads a half-megabyte page because the source renders its list server-side, while car_summary is one kilobyte, car_photos 2.9 KB and car_options 2.4 KB. Search once, then enrich with the cheap per-car calls — car_summary answered in under a second on every call we made.

brands is where an integration starts, because the search brand filter wants the source's own pinyin slug: 618 brands with their Chinese names, and 94 series for BMW when you pass the brand back. suggest does a different job — it turns a keyword into the source's model series with the live number of cars on sale and the real minimum and maximum asking price, which is a market-size answer in a single call. No che168 account, login or token is involved anywhere.