OpenRTB 2.x did not nest the Native request as JSON objects under imp.native. It stuffed a whole Native markup request, itself JSON, into a string field named request. A decoder that maps that field to an object will fail closed or, worse, fail open and bid without assets.
Two JSON documents, one HTTP body
The bid request is JSON. The Native request is also JSON. The second document is quoted inside the first. That is IAB Native 1.1/1.2 as carried by OpenRTB 2.x, not a vendor quirk. imp.native.request is a string. imp.native.ver names the Native spec version. Assets, eventtrackers, and context live inside the string.
{
"id": "req-1",
"imp": [{
"id": "1",
"native": {
"request": "{\"ver\":\"1.2\",\"assets\":[{\"id\":1,\"required\":1,\"title\":{\"len\":140}}]}",
"ver": "1.2"
}
}]
}
If your type for request is an object, most languages will error on unmarshal. If your type is a string and you never parse it, you will bid on a native impression whose title length, image sizes, and required assets you never read. That looks like fill. It is inventory you cannot render.
Where it breaks
Exchanges sometimes send the Native object unquoted because an internal model already had structs. Bidders that only accept a string then drop the imp. Bidders that only accept an object drop the spec-shaped requests. Both payloads exist.
OpenRTB 3.0 / AdCOM moved Native into the shared object model so this double encoding could die. Plenty of 2.6 traffic has not moved. If you still speak 2.x, you still have a string.
What to check
- Decode
imp.native.requestas a string, then JSON-parse that string into the Native request. - Read
veron both the OpenRTB native object and inside the request document. - Reject required assets you cannot fill.
- Paste a real native request into the tester. Field list: Native object reference.
This is a parse contract, not a fraud check. RTBlint is independent of IAB Tech Lab.
Original: imp.native.request is a JSON string
