How to publish facts you did not personally verify

If you publish data at any scale you will eventually publish something you did not personally confirm. Pretending otherwise is how sites end up quietly wrong for years.

The honest structure is three parts. A citation on every data point, so a reader can check it without asking you. A confidence grade on every cell, so “verified against the state website last month” and “found in a forum post” are not displayed identically. And uncertainty shown rather than hidden — a visible “we are not sure about this one” beats a confident number that is wrong.

Then a scheduled job that re-checks the whole set on a cycle, because the failure mode is not usually a bad number at publish time. It is a good number that went stale and nobody looked again.