A payroll headline can look decisive until earlier-month revisions, wages, hours and unemployment tell a more complicated story. This guide uses the original US Bureau of Labor Statistics releases and the St. Louis Fed's historical-data archive to build a repeatable reading method. It separates what was known on release day from later corrections, and explains why an employment surprise is not an automatic signal to trade gold, currencies or bonds.
1. Two surveys answer different questions
The Bureau of Labor Statistics publishes the Employment Situation from two distinct surveys. The establishment survey counts nonfarm payroll jobs and reports earnings and hours; the household survey classifies people by employment, unemployment and labor-force participation. Their samples, definitions and counting units differ. A person holding two payroll jobs can contribute two jobs to establishment counts while remaining one employed person in the household survey. A payroll increase therefore need not mechanically coincide with a lower unemployment rate.
Begin by recording the reference month, publication date and source, then separate monthly job creation from unemployment and participation. Do not confuse a vacancy advertised on a recruitment website with a job already counted on a payroll, and do not treat one growing industry as proof that every sector expanded. The official release gives industry details and methodological notes. Reading only a headline number strips away the distinctions needed for a careful assessment.
2. The first estimate is not the final estimate
Establishment-survey numbers are preliminary. More employer and government responses arrive later, and seasonal factors can be recalculated. In the September 2026 report released on 2 October, the first estimate of nonfarm payroll change was plus 29,000. July's previously reported plus 21,000 was revised to minus 10,000, a downward change of 31,000; August's plus 162,000 became plus 133,000, a downward change of 29,000. Together the revisions lowered the two earlier months by 60,000 jobs.
It would be incorrect to subtract 60,000 from September's 29,000 and call the result the actual September payroll change. The revisions apply to July and August, not September. Keep three columns in a research sheet: first-released figure, next-release revision and latest available value. A backtest using today's revised history as though it had been available at the earlier trading timestamp introduces look-ahead bias and can create imaginary predictive performance.
3. Annual benchmark revisions have another meaning
BLS also benchmarks establishment estimates against broader administrative employment counts, drawn mainly from unemployment-insurance records. On 28 August 2026 it announced a preliminary March 2026 benchmark difference of minus 79,000 for total nonfarm employment, approximately minus 0.1 percent. That is an adjustment to the estimated level for a reference month, not a sudden loss of 79,000 jobs in August. It is also separate from the July and August monthly revisions described above.
The agency explicitly said that this preliminary benchmark was not immediately incorporated into the official monthly series. The final benchmark was scheduled for inclusion with January 2027 data published in February 2027. Label each correction by type, affected reference period, release date and whether it is preliminary or final. Adding a benchmark-level revision to a monthly flow, without a coherent reconstruction of the series, is a category error rather than an insightful market calculation.
4. Wages, hours and unemployment provide context
The 2 October release reported September unemployment at 4.2 percent. Average hourly earnings for private nonfarm employees rose five cents, or 0.1 percent over the month, to $37.81; the twelve-month increase was 3.0 percent. Average weekly hours remained at 34.4. These observations describe different parts of the labor market. A single wage figure cannot establish future inflation or a central-bank decision, and nominal wage growth does not automatically equal growth in purchasing power.
Check whether gains are broad-based or concentrated in industries, and compare wage changes with prices and productivity before making a claim about inflation pressure. A shorter workweek can sometimes provide context before hiring changes, but it is not a reliable standalone recession signal. Avoid mixing seasonally adjusted and unadjusted series, and avoid confusing monthly growth with twelve-month growth. Neither payroll growth nor unemployment alone can summarize the experience of every household.
5. A worked surprise calculation, not a price forecast
Consider invented figures for teaching, not a historical market event. Suppose the consensus forecast is 120,000 additional jobs and the first release reports 80,000. The headline surprise is minus 40,000 relative to consensus. Suppose earlier-month revisions are minus 15,000 and minus 10,000: the combined revisions equal minus 25,000. Store the surprise and revisions separately. Calling minus 65,000 the number of new jobs would improperly mix a current-month forecast error with changes to earlier months.
Some investors may interpret weaker hiring as support for less restrictive monetary policy, yet the market also considers earnings, unemployment, inflation, real yields, expectations and liquidity. A first reaction in gold or foreign exchange can reverse when automated orders meet thin liquidity or widened spreads. Write conditional scenarios instead of promising a direction: what if hiring is weak but wages are firm, or hiring is strong but revisions are sharply negative? None of those patterns guarantees a trade outcome.
6. Build an auditable release log and respect execution risk
For each release record the timestamp and time zone, reference month, current payroll estimate, previous two months before and after revision, unemployment, hourly earnings and hours. Save the archived BLS release URL and the date you accessed it. The St. Louis Fed's ALFRED system preserves historical data vintages: what a series looked like when information was first available or later revised. Ordinary FRED displays usually favor the latest history. This distinction matters when evaluating a strategy with realistic information availability.
Check the official release calendar before acting and translate US Eastern Time carefully into your local clock, including daylight-saving changes. During fast markets, stop orders may fill away from the indicated trigger, spreads may widen and leveraged positions can suffer losses beyond the planned price move. Account for commissions, slippage, margin and platform failures. A screenshot showing one successful reaction is not evidence of a repeatable strategy. Use the jobs report to test and revise hypotheses, not to buy a guaranteed signal.
Official resources
- U.S. Bureau of Labor Statistics — The Employment Situation, September 2026, released 2 October 2026 (first estimates and July/August revisions) ↗
- U.S. Bureau of Labor Statistics — Preliminary CES benchmark revision for March 2026, released 28 August 2026 (not yet incorporated in monthly estimates) ↗
- Federal Reserve Bank of St. Louis — ALFRED archive and historical data vintages (checked 29 October 2026) ↗
- Federal Reserve Bank of St. Louis — FRED versus ALFRED and real-time historical values (checked 29 October 2026) ↗
- U.S. Bureau of Labor Statistics — Employment Situation technical note (survey methods, checked 29 October 2026) ↗
No online income or trading system guarantees profits. Results depend on skills, costs and circumstances.
Disclaimer ↗