This legitimately happened to me a few months ago. A vendor API was returning HTTP 200 with the error details embedded in the JSON response. It was a pain in the ass to troubleshoot.
I guess I might be evil but when I made APIs for my projects I do this, since I blindly accept the response then look at the JSON to see if it was accepted or not
Something like
if (body_has(JSON)) do_stuff_with(JSON) // including error handling if the response has an error else error_no_json()
I do this since I feel like JSON errors should be separate from HTTP errors
The problem I ran into was the response returned a JSON body, but then had an "error" attribute that was returned in it that had the error details. So we were parsing the JSON and loading elements into our database. We were hitting the API passing in a datetime of when the last success job was run, so basically saying "give me everything that's changed since I last called you."
So yeah, eventually we noticed we were missing small chunks of data. It turned out that every time the API errored out, we'd get a valid JSON response that contained the error message, but it didn't have the attributes we were looking for. So didn't load anything, but updated our timestamp to say when our last successful call was.
Huge pain in the ass to troubleshoot, because the missing data was scattered with no distinguiable pattern.
That would have been fine for me too. I don't own the API, so I can only speak from a consumer perspective in saying: I don't want a HTTP 200 if my request didn't succeed.