Four different kinds of change
A new observation adds a reference period to a history. Economic change is a difference between periods. An agency revision changes an estimate for the same period. A method change alters a definition, benchmark, classification or adjustment. A site correction fixes an error introduced in acquisition, processing or presentation. These events should not share one ambiguous update label.
Within-period revisions require an earlier preserved value for the same producer identity. The release-change workspace now compares two site acquisitions, distinguishing changed numeric values, status changes, extended historical coverage and source-identity corrections. It does not fabricate earlier agency vintages.
What is archived now
The first national snapshot is 2026-09-30; the reviewed version is 2026-09-30-d6e32ee5. Both remain downloadable. The regional archive similarly retains original and reviewed versions. GDP and personal-income updates revise matching past periods; ECI and the unemployment NSA variant require an identity correction, not an ordinary value revision.
Saved indicator and comparison URLs can load either supported national version. The first version is explicitly marked as a legacy view. These are site acquisition records, not a complete as-known-on calendar or proof of historical intraday availability.
How future snapshots are compared
The comparison script matches stable series identifiers and exact reference periods. A previously absent period is a new observation. A different value for a matching period is a candidate revision. Changed units, frequency, adjustment or producer identity halt publication review rather than being treated as ordinary numerical revisions.
The operator reviews each candidate against the relevant producer release and source notes. Approved snapshots remain immutable. The visible site can move to a new approved snapshot, while old versioned files remain available. A failed acquisition or validation leaves the old approved release visible with an appropriate status notice.
Reproduce responsibly
For a citation or saved research view, include the snapshot ID, source code, transformation and observation range. Historical backtests need an information cutoff supported by real vintage records. Do not add later-known revisions or chronology labels to an earlier information set without stating that limitation.