The number changed. Did the story?

A revision is information

The September employment release illustrates a recurring research problem: a new observation arrives with changes to earlier observations. A chart updated in place can make the earlier interpretation look inexplicable, even when it accurately reflected the information available then. U.S. Bureau of Labor Statistics ↗

Keep two dates

An evidence ledger should store both the period observed and the release date. It should also identify the value’s status: initial, revised or final where the source uses those labels. Those distinctions let a reader reconstruct what could have been known when a judgment was made.

For a maintained Signal, a revision should trigger a check of the reasoning rather than an automatic change of state. Ask whether the revised evidence touches the variable on which the judgment actually depends. A large change in an irrelevant variable should not outrank a small change in the controlling one.

Read the date with the number

Publication dates and source dates describe different events. Keeping both makes it possible to compare a current interpretation with the historical record it examines.