--- title: "Vintages — fixed, named versions of the data | Data Tactical Group" route: "/docs/methodology/vintages" updated: "2026-09-12" --- # Vintages Every DTG release is a fixed, named version of the data. Updated 2026-09-12 Labor market data does not stand still. The Bureau of Labor Statistics revises what it has already published. Industry, occupation, and geographic classifications change. Periods that were projected become surveyed, and periods that were surveyed become final. A figure pulled today and the same figure pulled next year may not be identical, and in most cases the later one is better. That creates a problem for anyone who has to defend a number after they published it. DTG addresses it by versioning the data. ## What a vintage is A vintage is a complete, fixed version of DTG data, built at a point in time and given a name. It contains the entire series — every industry, every occupation, every county and county-equivalent, every quarter from 2004 forward, along with carried-forward and projected periods — expressed in a single industry classification, a single occupation classification, and a single set of county definitions. Everything within a vintage is internally consistent. Detailed industries sum to their parents; occupations sum through the SOC hierarchy, each level to its own parent — though not every broad occupation splits into detailed occupations, so detailed occupations alone do not recover the total; counties and each state’s unallocated area sum to their state, and states to the nation (employment to within rounding; earnings and establishment counts very closely but not always exactly, with the largest differences in the smallest industries); and every year uses the same industry, occupation and geographic codes. Values are not mixed across versions. A released vintage is not altered afterward. When newer source data arrives, or a classification changes, or the series is otherwise rebuilt, the result is a new vintage. The previous one remains what it was. ## How vintages are named Vintages are identified by the year and the release number within that year. `DTG 2026.3` is the third DTG release of 2026. In the event of a corrected rebuild of a release, it will carry a revision marker: `DTG 2026.3r1`. The name identifies the version, not the period the data covers. Every vintage contains the full series. ## Corrections A defect that affects values is corrected by supersession, not by editing. DTG rebuilds the release, and the corrected build carries the revision marker: `DTG 2026.3r1` supersedes `DTG 2026.3`. The defective build is withdrawn from serving and archived — it is not served again and it is not changed, so the rule that a released build is never altered holds through a correction. A superseding build takes its parent’s place in retention, and its release notes name what changed. A correction that affects documentation rather than values does not produce a new build. ## Retention and access DTG retains three vintages: the current one and the two before it. Older versions are removed as new ones are released. The current vintage is available to every account. Access to prior vintages depends on plan; see [pricing](/pricing) for details. ## Why values change between vintages Five things move numbers from one vintage to the next. None of them is an error, and all of them are visible in the data itself. **Preliminary figures become final.** BLS publishes preliminary values for recent quarters and revises them as later reports arrive. DTG reprocesses the current and prior data-file years with each release, so recent quarters improve over successive vintages. **BLS revises earlier periods.** Revisions announced for periods outside the normal window are incorporated when they occur. **Classifications are restated.** When BLS adopts a new industry or occupation classification, or a county definition changes, the next vintage restates the entire series onto the new definitions. Industry values changed by restatement carry flags, described on the [disclosure and provenance](/docs/methodology/disclosure) page. Occupation values are restated on the same principle but are not flagged. **Occupation staffing patterns are re-derived.** Occupation data before 2017 is built from the earliest available survey-based staffing patterns, and those patterns are derived again from their sources with every release. Because the sources themselves are revisable, occupation values in the early years of the series can move between vintages. This follows from the same principle as restatement: each vintage is built in full, from current inputs, rather than carrying earlier results forward. **Projected periods become actual.** A quarter carried forward from survey data in one vintage may be reported from QCEW in the next, and its disclosure code changes accordingly. The same period can be projected, then carried forward, then published, across three successive vintages, and the value will likely differ at each step. The most recent periods therefore change most. Completed QCEW quarters with disclosed values change least. ## Citing a figure A DTG figure is reproducible when the vintage it came from is recorded with it. For work that will be reviewed, challenged or revisited — a published report, a filing, a board document, an application — cite the vintage alongside the figure, along with the industry or occupation, geography and period. Anyone with access to that vintage can retrieve exactly the same number, whatever has changed since. A figure cited without a vintage cannot be reproduced reliably, because the underlying data will have moved. ## Release history DTG does not publish a release calendar and does not commit to a cadence. What is published is observed history: the date each vintage was released, recorded on the [releases](/releases) page alongside its release notes. Past turnaround is a matter of record, not a schedule. ## Release notes Each vintage is accompanied by release notes recording what changed: source revisions incorporated, classification changes applied, and any change to how the data is built. Still stuck? A person can help. [Get support](/support)