A request id is the identifier stamped on one API call. It comes back in the X-Request-Id response header and is repeated inside the body — in meta.request_id on a success and in error.request_id on a failure — so the same string exists on both sides of the call.
What it is for
Quoting one to support turns "a call failed yesterday" into a row somebody can read. Every request an account makes is recorded with its id, the endpoint, the status, the typed error code, the credits it cost and how long it took; the account's own list is at /requests, and the support form offers the newest ids so nobody has to go and find one.
The header is exposed to browser JavaScript on purpose, because the documentation tells callers to read it and a browser cannot read a header that is not listed in the CORS policy.
What the record does not hold
Not the value the call was about. The route is stored as its template, never as its path, so a request log row says that a phone lookup happened and never which number. Rows are deleted after 30 days.