I’m preparing a Batch API job with many requests and need to match each result back to the record that created it. Are results written to the output file in the same order as the input JSONL, or should I rely on the custom_id field to match them?

If the order isn’t guaranteed, I’d also be interested in how people handle missing or failed records when processing the output. I’m mainly trying to make sure my import step doesn’t silently attach an answer to the wrong request.