July's Jobs Report Went Negative — And Quietly Revised Away 103,000 More

At 8:30 on Friday morning the Bureau of Labor Statistics published a number, and by 8:31 several hundred people had finished explaining what it meant. The number was −23,000. Total nonfarm payroll employment, July 2026, seasonally adjusted, down twenty-three thousand.

That's the number that got the coverage. It is not the interesting number.

The interesting number is further down the same release, in the flat administrative prose BLS uses for everything including its own corrections: "The change in total nonfarm payroll employment for May was revised down by 66,000, from +129,000 to +63,000, and the change for June was revised down by 37,000, from +57,000 to +20,000."

Sixty-six thousand plus thirty-seven thousand. One hundred and three thousand jobs that May and June used to have, and now don't.

Nobody sent a correction. There is no correction to send. This is the system working exactly as designed, which is the part worth your next four minutes.

What you read in June, and what actually happened¶

Line them up. This is the whole article in six numbers.

As first reported, month by month: May +129,000 · June +57,000 · July −23,000. Add them, divide by three: +54,300 jobs a month. A cooling labor market that is nevertheless still a labor market. You could run a business on that read. Plenty of people did.

As the record now stands: May +63,000 · June +20,000 · July −23,000. Add them, divide by three: +20,000 jobs a month.

Nothing happened to the economy between those two rows. The only thing that changed is how much of it BLS had counted.

Fifty-four thousand a month is a slow expansion. Twenty thousand a month is functionally a flat line. Whether that's enough depends on how fast the working-age population is growing — a genuinely contested question at the moment, and not one I'm going to settle in a subordinate clause. What isn't contested is narrower and more useful: those two figures describe meaningfully different labor markets, and only one of them is on the record. If you made a hiring decision in June off the +129,000 print, you made it off a number that no longer exists.

Why the number moves, and why that's the good news¶

Here is the mechanism, because the mechanism is the reason none of this is a conspiracy.

The payroll figure comes from the Current Employment Statistics survey — a sample of employers who report their own payroll counts. BLS publishes the first estimate roughly three weeks after the reference period, using whatever reports have arrived by then. Estimates are then revised twice, in each of the two following months, as more employers file. After that the figure is held constant until the annual benchmarking process rebuilds it against near-universal administrative records.

So the first print isn't a measurement. It's an estimate built on a partial sample, published fast because a slow accurate number is less useful than a fast approximate one. The revision isn't a correction of an error. It's the arrival of the rest of the mail.

That's the turn, and it's worth being blunt about it: the problem was never that BLS revised the number. The problem is an information diet where the first print arrives with sirens and the revision arrives with nothing at all — where a figure with a known ±51,000 tendency gets reported to the last digit and then quietly restated a month later into a silence.

“

A number that never gets revised is a number nobody is checking. The revision is the audit trail. It's the only evidence you have that someone went back and looked.

”

Any operator who has ever closed a month knows this instinctively. Your first revenue figure on the 1st is not your final revenue figure on the 15th, because invoices land late and cards settle and somebody expensed a flight. If your books never moved after you closed them, you wouldn't trust them more. You'd trust them considerably less. It's the same discipline that makes double-entry worth caring about in the first place — the correction is the feature.

The line under the headline: wages against prices¶

Two more figures from the same release, and one from the release before it.

Average hourly earnings for private nonfarm employees came in at $37.62 in July, up two cents on the month and 3.2% over the year. The most recent inflation print — June CPI, published July 14 — had all-items prices up 3.5% over the year.

Those two periods don't line up exactly, and I'm not going to pretend they do: one is the year to July, the other the year to June, and July's CPI doesn't land until August 12. But the direction is clear enough to act on. Nominal wages are growing roughly in line with, and probably a touch behind, prices. The paycheck is not getting meaningfully bigger in real terms.

Which connects directly to the thing this desk has been tracking all month: transactions at small businesses have now fallen year over year for nine consecutive months while sales held up on ticket size alone. A consumer whose pay is flat in real terms and who is visiting fewer places while spending more at each is not a mystery. That's the same person, described twice, by two different agencies.

What to actually do with this¶

The temptation with macro data is to read it as weather — something that happens to you, that you comment on, that changes nothing you do on Tuesday. Here's the version that changes Tuesday.

1. 1

   Never make a decision on a first print

   This is the practical takeaway and it costs you nothing. When a jobs number lands, note it and wait. It has a known ±51,000 tendency and two scheduled revisions ahead of it. If a headline number is going to move your hiring plan, let it settle for two months first — the plan will still be there.

2. 2

   Watch the three-month average, and recompute it monthly

   One month is noise; the revisions prove it. Keep a running three-month average and rebuild it each month with the current figures rather than the ones you wrote down at the time. Right now that number is +20,000 a month, and it's the honest summary of where hiring is.

3. 3

   Read a soft labor market as a hiring opportunity, not just a demand warning

   Both things are true at once and most people only register the first. A market adding 20,000 jobs a month is a market where the person you couldn't afford in 2022 may now answer your email. If you have the cash flow to carry a hire through a slow patch, this is a materially better market to hire into than a hot one.

4. 4

   Stress-test your pricing against flat real wages, not against inflation

   "Inflation is 3.5%, so I'll raise prices 3.5%" quietly assumes your customers got a 3.5% raise. They got 3.2%, before tax, on average, and averages hide a lot. Price against what your customer can actually absorb — which is a different exercise from indexing to CPI, and one worth doing properly.

5. 5

   Fix your own revision problem first

   If your internal numbers never get revised, you are not more accurate than BLS. You are less audited. Close the month, then close it again two weeks later, and keep both. The gap between them is the most useful diagnostic you will generate all year.

Frequently asked questions¶

The correction that never gets printed¶

The thing I keep returning to isn't the −23,000. It's the ninety seconds of Friday morning where a wire reporter had to decide whether the revision was part of the story, and the structural reality that in most outlets it wasn't.

There's no villain in that. Revisions are boring, they're a month stale on arrival, and "the number we told you in June was 66,000 too high" is a sentence with no headline in it. So the first print does all the work in the public record, and the truer number arrives later to an empty room.

You cannot fix that. What you can do is stop treating the first print as the last word — on payrolls, on your own month, on anything where the count came in before the counting finished. The businesses that get hurt by a turn in the labor market are rarely the ones that read the data wrong. They're the ones that read it once and never went back.

The mail is still arriving. It always is.

It's just business — and the books get restated.

Sources

1. U.S. Bureau of Labor Statistics — Employment Situation Summary, July 2026 (released August 7, 2026)
2. U.S. Bureau of Labor Statistics — CES Revisions to Establishment Survey Data (revision schedule and mean absolute revision)
3. U.S. Bureau of Labor Statistics — Consumer Price Index Summary, June 2026 (released July 14, 2026)
4. U.S. Bureau of Labor Statistics — CPI release schedule (July 2026 data publishes August 12, 2026)