The first print is a model.

At 8:30 on Friday morning, September 4, the Bureau of Labor Statistics printed the number that moved markets for the rest of the day: 162,000. That was the gain in payroll jobs in August, about three times the 53,000 economists had expected, and investors quickly raised the odds of a rate hike at the Federal Reserve's September meeting to about 60 percent.¹⁶ The same release, in its technical note, says how much weight that number can carry. For the monthly change in payrolls, the 90 percent confidence interval is on the order of plus or minus 122,000.¹

Put those two facts together and the honest reading of Friday's headline is a range: somewhere between about 40,000 and 284,000 new jobs, with 162,000 as the best single guess. The first print is the figure that reaches the headline, the phone alert and the trading desk. The estimate with its error bar is what the Bureau actually published: a number produced by a model from a partial sample, with a stated margin and a schedule of revisions already on the calendar. They share the same digits. They are not the same thing.

My argument is that anyone who acts on a jobs report, whether setting rates, pricing a bond or planning next year's hiring, should read the first print as the estimate it is, with its error bar, its revision schedule and the trend around it. This does not claim that the number is wrong, that the Bureau does careless work, or that August was weak. By the release's own yardstick, August's gain was larger than its margin of error, which is more than June or July could say. The claim is only about how much one month can bear.

What the number is

The payroll figure comes from the Current Employment Statistics survey: about 119,000 businesses and government agencies, representing roughly 622,000 worksites.¹ It is not a count of every job in the country. The Bureau takes the change in employment among businesses that reported both this month and last, weights it, and applies that relative change to last month's estimate to carry it forward.⁵ Then it adds a modeled amount for the jobs created and lost as businesses open and close, a component the Bureau describes as covering business births and deaths not captured by the sample.⁵ And it does all of this before the sample is complete. The estimates for the two most recent months, the release says, are based on incomplete returns, which is why they are labeled preliminary.¹

So the first print is three things at once: a sample, a ratio and a forecast. None of that is hidden. It is spelled out in the technical note at the back of every release, which is roughly where most readers have stopped reading.

Every print is a vintage

The clearest evidence is what happened to the months just before August. May was first printed on June 5 at 172,000 jobs. A month later it was revised to 129,000, and a month after that to 63,000.³² June was first printed at 57,000, revised down to 20,000, then back up to 31,000.³²¹ July's first print, on August 7, was a loss of 23,000. On September 4 it became a gain of 21,000.²¹ The sign changed. Anything written on August 7 about the month the economy started losing jobs was written about a number that no longer exists.

Figure 2 lays those prints side by side. Each gold dot is a first print. The blue bar around it is that print's 90 percent interval, plus or minus 122,000. The pencil marks are the later versions of the same month. Every revision so far has landed inside the bar. That is not a promise: the interval describes the error that comes from sampling, and revisions also reflect the reports that arrive late. But the bar was printed on the first day, and it told a careful reader how far the number could plausibly move.

A revision already in the queue

There is a second, slower kind of revision. Once a year the Bureau re-anchors the survey to a fuller count of employment, a step it calls the benchmark. On August 28, a week before the August report, it published its preliminary estimate of the next one: payroll employment in March 2026 was 79,000, or 0.1 percent, higher than the fuller count supports. For private employers alone the overstatement was 178,000. The final figure will be built into the official numbers with the January 2027 report, published in February.⁴

So Friday's 162,000 sits on a level that the Bureau has already said is probably a little too high. That does not shrink August's change by any known amount, and 0.1 percent is small; the Bureau notes that its benchmark revisions over the past ten years have averaged 0.2 percent in absolute terms.⁴ The point is narrower. The level, not just the monthly change, is an estimate, and the correction to it was public before the August number came out.

The industry inside the total

The same caution applies with more force inside the total. Health care added 13,000 jobs in August, well below its average gain of 32,000 a month over the prior year.¹ One month at 13,000 means little on its own; the error bar that makes the total uncertain makes its parts more so. The trend is more telling. The July report put health care's prior twelve-month average at 36,000, and the June report at 38,000.²³ The average itself has drifted down across three releases, and the preliminary benchmark says private education and health services were overstated by 96,000 in March.⁴

Narrower still: Becker's Hospital Review, reading the detailed tables the same morning, reported that offices of dentists lost 4,600 jobs in August while offices of physicians added 2,000.⁷ A detailed industry is estimated from a smaller slice of the same sample, so one month there deserves even less weight than the national figure. A practice owner who reads "dental jobs fell" on Friday and changes a hiring plan on Monday is acting on the first print of a first print.

Reading a number that is published early

Any system that has to publish before it knows works this way. A flash estimate, a month-end close full of accruals, a forecast in a board deck: each is a best guess made on a deadline, and each will be replaced. No. 9, Yesterday's answer can change, argued for keeping the old versions when that happens. This is the reading-side half of the same idea. A number that will be revised should be read with its margin and its revision dates attached, and a decision built on it should rest on the trend rather than the latest point.

The Bureau, to its credit, publishes everything a careful reader needs: the interval, the preliminary label, the revision of the two prior months, the benchmark date. It even says how big a move has to be to mean something in the noisier household survey: about 650,000 people.¹ August's household employment gain of 569,000 fell short of that bar.¹ The two surveys measure different things in different ways, as an earlier dispatch on reading the denominator explained, but both arrive with their margins attached. The bar gets dropped somewhere between the release and the headline.

Objections, considered

Revisions are small on average, so this is pedantry. Over a year, they are small: the benchmark revisions have averaged 0.2 percent. Over a month, they are not. With the three-month average of payroll gains at about 71,000, an interval of 122,000 on a single month is larger than the signal it is meant to describe, and July's first print had the wrong sign.¹

Markets moved on the number, so the number mattered. It did. A first print is information about what everyone else will do, and a trader who ignored it would lose money. Reading it as an estimate is not ignoring it. It is pricing the chance that the next two releases tell a different story, which, as May and July show, they sometimes do.

Error bars on everything make it impossible to decide. The opposite. A decision that names its range can be made now and revisited on a schedule. The useful question is not whether August was exactly 162,000 but what range of outcomes would change the plan, and whether the trend has moved into it.

The first print is the Bureau's best guess on the day, and the error bar is part of the guess.

Before the next jobs report

• Read the range with the number. For a monthly payroll change, add and subtract about 122,000 before deciding what it says.
• Name the vintage. Every figure quoted in a plan or a deck should carry the date of the release it came from.
• Know what is scheduled to change it. Two monthly revisions and an annual benchmark, all on the calendar in advance.
• Decide on the trend. If a decision turns on a single month, say in advance what the next two prints would have to show to reverse it.

Back to Friday's slip. The gold line, +162,000, is what printed first and what the market heard. The blue lines under it were printed in the same minute: the interval, the incomplete returns, the modeled businesses, the revisions due on October 2 and the month after, the benchmark due in February. The slip is still printing. The right time to read August is with all of it on the page.