August nonfarm payrolls hit 162,000 against a 53,000 consensus, while June and July were revised up by a combined 55,000 — turning a reported month of losses into gains. The prior two reports revised earlier months down by 103,000 and 74,000. Sixty-two percent of August's gain came from food service and government education, the two hardest August series to seasonally adjust.
The US economy added 162,000 jobs in August against a Dow Jones consensus of 53,000 -- a beat of 109,000, roughly three times the forecast. In the same release, the Bureau of Labor Statistics revised July from -23,000 to +21,000, and June from +20,000 to +31,000. A month that had been reported as job losses became a month of job gains, retroactively, in a paragraph most readers skip.
Those two facts belong in the same sentence, and almost nowhere are they put there. The July revision was 44,000. The June revision was 11,000. Combined, 55,000 -- half the size of the surprise everyone reacted to. And this month the revisions ran in the flattering direction. The two prior reports revised earlier months down by 103,000 and 74,000.
Set the figures next to each other, which is the comparison the coverage does not make.
The revisions are not small noise around a stable signal. They are the same order of magnitude as the thing markets treat as the signal. A 103,000 downward revision is almost exactly the size of the 109,000 upside surprise that produced this month's headlines -- and it arrived with a fraction of the attention, because revisions are always printed alongside a fresh number that competes with them for the same reader.
None of this is a claim that the BLS is doing something wrong. Revision is designed into the series. The first print is an estimate drawn from a survey with incomplete responses; establishments report late, and the sample fills in over the following two months. The agency says so, and attributes the adjustments to updated survey responses and recalculated seasonal factors. The mechanism is public and functioning as intended.
The problem is not the revising. It is that the first print is the version that moves markets and headlines, and the first print is the least reliable version of the number.
A forecaster predicting nonfarm payrolls is predicting a figure that will be rewritten twice, by amounts comparable to the forecast error being used to score them.
Composition matters here more than in a typical month. Restaurants and bars added 59,000 jobs -- more than a third of the entire monthly gain on their own. Government education added 42,000. Manufacturing contributed 16,000.
Food service and education employment in August are the two hardest series in the calendar to seasonally adjust, for the same underlying reason: both have enormous genuine seasonal swings that the adjustment has to net out. Schools staff up as terms begin, on a timetable that varies by district and shifts from year to year. Restaurant employment turns as the summer season ends, on a timetable that depends on weather and tourism patterns.
Seasonal adjustment handles this by estimating what a normal August looks like and reporting the deviation from it. When the raw swing is very large and its timing moves around, a modest error in the estimate of 'normal' produces a large error in the adjusted figure. Roughly 62% of August's headline gain came from precisely the two categories where that error is largest.
That does not mean the jobs are not real. It means the confidence interval around this particular month's headline is wider than around a month whose gains came from, say, professional services or construction.
Information-sector employment declined, and the detail is worth reading rather than summarising. Computing infrastructure providers, data processing, web hosting and related services shed 8,000. Publishing lost 7,000. Broadcasting and content providers lost 5,000. Twenty thousand jobs across the sector most exposed both to AI substitution and to the ongoing reset in digital advertising.
That is a small number against a 162,000 headline and should not be over-read from a single month. But it survives the seasonal-adjustment objection better than the headline does: information-sector employment does not carry a large August seasonal, so the adjusted figure is closer to the raw one.
Separately from the monthly revisions, the BLS publishes an annual preliminary benchmark revision that reconciles the survey estimate against near-universal administrative records from unemployment insurance tax filings. The most recent one estimated that total nonfarm employment for the year through March 2026 was overstated by 79,000, or about 0.1%.
By historical standards that is a small benchmark revision, and it is worth saying so, because recent years have produced far larger ones. The 79,000 figure is evidence that the level of the series is roughly where the administrative data says it should be -- even while the month-to-month changes remain volatile. The final version is incorporated into official estimates with the January 2027 report, published in February 2027.
Level and change are separate questions, and they currently have separate answers. The level looks well anchored. The monthly change is what keeps getting rewritten.
Two series in the same release move far less and get revised far less than the payroll count does.
Neither is dramatic, and that is the point. Unemployment flat at 4.1% and wages growing 3.1% describe a labour market that is neither tightening sharply nor deteriorating quickly. That description has held steady across several months in which the headline payroll figure swung from reported losses to a beat three times consensus.
The honest summary of the August employment report is roughly this: the level of employment is fine, the direction is mildly positive, more than half the gain came from the two categories where the measurement is weakest, and the number will be different in October.
That is a less satisfying sentence than 'blowout jobs report,' which is why it is not the one being written. But anyone hiring, trading or planning against this series is better served by watching unemployment, wages and the three-print trend than by reacting to any single first estimate -- including this one.
The first print is an estimate from a survey of employers that is incomplete on publication day. Establishments report late, and the sample fills in over the following two months, so each month's figure is revised twice. The BLS also recalculates seasonal adjustment factors as new data arrives. Revision is a designed feature of the series, not a malfunction -- but it does mean the version that moves markets is the least complete one.
DrafterDaily's finance coverage works through how the headline numbers are constructed, and what survives the caveats.
More finance analysisReported talks would put roughly $5 billion of public credit into Fluidstack through the Office of Strategic Capital. The instrument is the story: a loan avoids appropriations, delivery obligations and asset ownership — and puts taxpayers above equity priced for a very fat right tail.
The August PPI moved rate-hike odds to roughly 70%. The rise came from goods, and over a third of the goods increase traced to diesel. The services line closest to the Fed's actual target did not move at all.
Nothing was cancelled on 4 September. No lender was ordered to drop FICO. What the FHFA removed was a 50-lender ceiling — and the market repriced the durability of pricing power, which is a bigger move than losing a customer.