Primary sources
Four roles matter, and this site keeps them separate everywhere: datasets that feed the score, datasets shown as context, AI research shown only on AI pages, and datasets planned for later enrichment that are not ingested today.
Used in the score
- JOLTSJob Openings and Labor Turnover Survey, U.S. Bureau of Labor Statistics. Monthly, released about six weeks after the reference month.
- CESCurrent Employment Statistics, U.S. Bureau of Labor Statistics. Monthly, released in the first weeks after the reference month.
- OEWSOccupational Employment and Wage Statistics, U.S. Bureau of Labor Statistics. Annual.
- Employment ProjectionsEmployment Projections, U.S. Bureau of Labor Statistics. Annual, covering the published long-term projection horizon.
- CPSCurrent Population Survey: unemployment by occupation, U.S. Bureau of Labor Statistics. Annual averages table from the monthly household survey.
Shown as context
- Business Employment DynamicsBusiness Employment Dynamics (BED), U.S. Bureau of Labor Statistics. Quarterly, with a publication lag of several months.
- H-1B LCA disclosuresLabor Condition Application public disclosure data, U.S. Department of Labor, Office of Foreign Labor Certification. Quarterly disclosures; file coverage varies.
- Indeed Hiring LabIndeed Job Postings Index, Indeed Hiring Lab. Daily observations; refreshed weekly by the publisher.
- QCEWQuarterly Census of Employment and Wages, U.S. Bureau of Labor Statistics. Quarterly, released a few months after the reference quarter.
- O*NET®Occupational Information Network (O*NET®) Database, U.S. Department of Labor, Employment and Training Administration. Updated on a rolling basis, with periodic database releases.
- ACSAmerican Community Survey, U.S. Census Bureau. Annual, with 1-year and 5-year estimate products.
- CBPCounty Business Patterns, U.S. Census Bureau. Annual.
- WARN noticesState Worker Adjustment and Retraining Notification (WARN) notices, State labor and workforce agencies. Varies by state, from twice weekly to as notices are processed.
- LAUSLocal Area Unemployment Statistics, U.S. Bureau of Labor Statistics. Monthly.
- QWILEHD Quarterly Workforce Indicators, U.S. Census Bureau. Quarterly, typically about three quarters after the reference quarter.
- UI claims (ETA 539)Weekly Unemployment Insurance Claims (ETA 539), U.S. Department of Labor, Employment and Training Administration. Weekly, typically published Thursdays for the prior week.
- IPEDSIntegrated Postsecondary Education Data System: Completions, U.S. Department of Education, National Center for Education Statistics. Annual; provisional then final releases.
AI & Jobs research (never scored)
- Census BTOS: AI useBusiness Trends and Outlook Survey: AI use (Core and 2025–26 AI Supplement), U.S. Census Bureau. Core question biweekly; AI Supplement collected once (Nov 17, 2025 to Feb 8, 2026).
- BLS projections: AI impactsEmployment Projections: AI impacts, U.S. Bureau of Labor Statistics. Annual employment projections; AI methodology updated as BLS research evolves.
- Anthropic Economic IndexAnthropic Economic Index, Anthropic. Irregular research releases; checked quarterly.
- Microsoft AI applicabilityWorking with AI: AI applicability, Microsoft Research (Tomlinson et al., 2025). Static research release; checked annually.
- GPTs are GPTsGPTs are GPTs occupation exposure, OpenAI / OpenResearch / University of Pennsylvania (Eloundou et al., 2023). Static research release; checked annually.
- AIOEAI Occupational Exposure (AIOE), Felten, Raj & Seamans (2021). Static research release; checked annually.
- OECD AI exposureOECD AI Exposure Measure (AI Capability Gap), OECD (AI Papers No. 59, 2026). Research release (2026); checked annually.
Planned, not ingested
- BDSBusiness Dynamics Statistics, U.S. Census Bureau. Planned enrichment. No figure on this site uses it today.
- CareerOneStopCareerOneStop data services, U.S. Department of Labor, Employment and Training Administration. Planned enrichment. No figure on this site uses it today.
Why the schedules differ
Fast data is broad and slow data is specific. Monthly surveys of establishments, such as JOLTS and CES, can publish within weeks because they ask a small number of questions of a sample of employers, and they classify jobs by industry. Occupational estimates require far more detail per job, so OEWS publishes annually. Long term projections are modeled once a year over the published long-term projection horizon.
That is not an inconsistency to be smoothed over. It means the most current thing you can know about an occupation is industry demand pressure, while the most precise things you can know about it, wages and staffing, describe a period that has already closed.
Data through, not updated today
Where a figure appears on this site, the page states the release it came from and when it was imported. Publication dates are shown only when matched to a checked publisher notice; unknown dates remain unverified.
Source freshness review · September 19, 2026
A reference period, a publisher release date, and our import date are different facts. A recent import alone does not establish that a dataset is the latest available release.
| Source | What the review established | Publisher evidence |
|---|---|---|
| Monthly score inputs | The national score uses July 2026. August CES was scheduled for September 4; August JOLTS for September 29. July is consistent with the latest common month, but the feed does not report source-level CES/JOLTS imports, so their lineage is not independently verified. | BLS release calendar |
| OEWS wages | May 2025 estimates; published May 15, 2026. Imported September 17, 2026. The feed incorrectly labels May 1, 2025 as the release date. | OEWS publication |
| Employment projections | 2025–2035 projections; published August 27, 2026. Imported September 17, 2026. The feed incorrectly reports September 1, 2025 as the release date. | Projections publication |
| QCEW | First quarter 2026; news release August 28, 2026. Imported September 16, 2026. January 1 is the start of the reference quarter, not its publication date. A news-release date does not identify a later revision. | QCEW publication |
| O*NET | The imported 31.0 version matches the production version listed in the publisher archive at review time (August 2026). Imported September 13, 2026. An exact publication day is not asserted. | O*NET release archive |
Next scheduled monthly milestone at review time: August JOLTS on September 29, 2026. A new score month should follow only after the upstream pipeline imports, recalculates, and validates it. This review does not advance or restate existing scores.
Additional signals have separate freshness rules: WARN is fetched from supported state agencies; news snapshots expire after 48 hours. At review time the committed news snapshot was empty, so no populated statewide news gauge was verified.
Freshness and release cadence
Source cards distinguish reported imports from independently verified publication dates. An “Imported” badge describes the pipeline, not a guarantee that no newer release exists. Missing release metadata is marked as unreported rather than proof that no data was imported.
National scores and local context
The score is national. Occupation pages show metro employment and wages from OEWS as context, because those estimates exist at metro level, but there is no metro or state level Job Market Health score, and metro lists are ordered by employment rather than presented as rankings of the best places to work.
State pages already show QCEW industry conditions weighted by the occupation’s industry mix. Census ACS, County Business Patterns and Business Dynamics Statistics remain planned sources. State context is separate from the national score.
Published scores versus later revisions
Federal statistics get revised, sometimes months later. When that happens, affected scores are recalculated, so a month can carry both the value first published and a restated value. Scores also carry a confidence rating with its reasons, which is how a month built on partial or stale inputs is flagged rather than hidden.
Provenance and methodology versioning
Every score records the methodology version it was computed under. Changes ship as a new version rather than a silent rewrite of history. The current weighting is described on the methodology page.
Limitations
- Government data is lagged. Nothing here describes this week.
- Estimates can be revised, and suppressed values are left empty rather than filled in.
- Monthly demand data is industry based, so occupation level demand is weighted exposure rather than a measured occupational vacancy count.
- An occupation-level score is a market signal, not a promise about an individual job search. Your experience, location and field of specialization matter more to your outcome than any national average.
- This site is a read-only consumer of published data and does not collect job postings. JobMarketHealth does not republish government statistics as its own. It combines published government data to create original derived labor-market indicators, with the underlying sources, transformations and vintages disclosed.
Next: read the dataset pages for what each source measures, or the guides for how to use these measures in a real job search.