A hash proves identity, not truth.

A cryptographic hash can tell us that the bytes we received match the bytes that were committed. A signature can connect those bytes to a signing key. Neither operation decides whether the source was complete, current, correctly interpreted or true about the world.

Keeping these questions separate prevents a technically strong integrity mechanism from being used as a shortcut for factual authority.

A record may represent a report, an observation, a claim, a derived statistic or a missing value. The same field name can carry different meaning across sources, and a stable identifier does not remove that difference.

Good data systems preserve source identity, reference period, transformations and missingness instead of flattening everything into one confident score.

Content-addressed snapshots and computation receipts are useful because they make a change inspectable. They help distinguish a changed input from a changed parser, model, query or execution environment.

That is a foundation for reproducibility, not a substitute for validation. A reproducible mistake is still a mistake; the value is that it can be found and corrected without guessing which bytes changed.

The final product should explain what a source supports, what it does not support and what a user should verify independently. Abstention is often more useful than a precise-looking answer with an invisible assumption.

This is how provenance becomes practical: not by making every result certain, but by making the next responsible action easier to choose.