Benchmark Revisions Land on One Month, and 79,000 Understates the 178,000 in Private Payrolls

On August 28, 2026 the Bureau of Labor Statistics published this sentence: “The preliminary estimate of the Current Employment Statistics (CES) national benchmark revision to total nonfarm employment for March 2026 was -79,000 (-0.1 percent), the U.S. Bureau of Labor Statistics reported today.” A headline that day rendered it as a change in job growth — “BLS revises U.S. job growth down 79,000 in benchmark update”.

Three sentences in the same release rule that reading out. The revision is calculated for one month. It has not been applied to any published number. And the figure for the private sector is more than twice as large.

None of that makes the estimate unimportant. It makes it a different object than the headline suggests, and the difference is arithmetic rather than interpretation.

One Month, Not One Year

The release is explicit about scope: “The preliminary benchmark revisions in table 1 are calculated only for March 2026 for the major industry sectors.” What the number measures is stated separately: “It serves as a preliminary measure of the total error in CES employment estimates from March 2025 to March 2026.”

Those two sentences do different work. The second says the error accumulated over twelve months. The first says the estimate of it is attached to a single month’s level. Converting the second into the first is what the benchmark process does later, by a fixed rule set out in the CES National Benchmark Article: employment estimates for the months between the most recent March benchmark and the previous year’s benchmark are adjusted using a linear “wedge-back” procedure, and “This procedure assumes that the total estimation error accumulated at a steady rate since the last benchmark.”

A steady rate means division. Spread across twelve months, 79,000 is 6,583 a month (this article’s calculation). That is the amount by which each over-the-month change inside the window would move, not the amount by which a given month’s level would move. The level shifts by one twelfth at April 2025, two twelfths at May, and by the full amount at March 2026.

For scale, the Employment Situation released on September 4, 2026 reported that August’s gain of 162,000 was “higher than the average monthly gain of 31,000 over the prior 12 months.” A wedge of 6,583 a month is 21.2 percent of that average (this article’s calculation). The two windows do not line up — the benchmark covers April 2025 through March 2026, the trailing average covers September 2025 through August 2026 — so this is a comparison of size, not a restatement of either figure.

What happens after the benchmark month follows a separate rule. For the nine months that follow it, BLS “applies previously derived over-the-month sample changes to the revised March level to get the revised estimates. New net birth-death model forecasts are also calculated and applied during post-benchmark estimation.” The level carries forward.

The whole sequence is still conditional, because none of it has happened. The release says so twice: “Official establishment survey estimates are not updated based on this preliminary benchmark revision,” and “The final benchmark revision will be incorporated into official estimates with the publication of the January 2027 Employment Situation news release in February 2027.” Every payroll figure in circulation today is unbenchmarked with respect to this estimate.

The Private Number Is 178,000, and the Percent Column Hides It

Table 1 of the release puts two rows above the sector detail and one below it. Total nonfarm is -79 thousand at -0.1 percent. Total private is -178 thousand, also at -0.1 percent. Government, printed last, after the sector rows rather than above them, is +99 thousand at +0.4 percent.

The private revision is 2.25 times the total nonfarm revision and the two print as the same rounded percent (this article’s calculation). The table carries a footnote on that column: “Values of 0.0 indicate a value between plus or minus 0.05 percent.” One decimal place cannot separate the two rows.

The offset is government. A +99 thousand revision there absorbs 55.6 percent of the -178 thousand private revision (this article’s calculation), which is why the top line reads smaller than the private line beneath it.

The sector rows push in the same direction. Summing the ten private supersectors in table 1, downward revisions total 412 thousand and upward revisions total 234 thousand; the net of the two is the published -178 (this article’s calculation). Counting rows whose revision exceeds the total nonfarm figure in magnitude, four of the ten private supersectors clear it, as do three of the four components inside trade, transportation, and utilities (this article’s calculation). The two largest of those are retail trade at -154.6 thousand and transportation and warehousing at +135.1 thousand.

Trade, transportation, and utilities shows the compression most sharply. Its four components are wholesale trade at -86.2, retail trade at -154.6, transportation and warehousing at +135.1, and utilities at +8.1. They sum to -97.6, which the table prints as -98. Gross movement inside that single supersector is 384.0 thousand, or 3.93 times the net (this article’s calculation).

BLS attaches its own caution to these rows: “As is typically the case, many of the individual industry series show larger percentage revisions than the total nonfarm series, primarily because statistical sampling error is greater at more detailed levels than at an aggregated level.” That is a warning against treating any one sector line as a finding. It is not a warning against noticing that the aggregate is a residue.

How Far a Preliminary Estimate Has Moved Before

The preliminary figure measures a gap between two counts, and the release is direct about it: “The preliminary benchmark revision reflects the difference between two independently derived employment counts, each subject to their own sources of error.” The question that follows is how far the final has historically moved from the preliminary. BLS does not publish that comparison in one place, but both series are on the record.

Three recent benchmarks, each preliminary announcement set against the final in table 4 of the CES National Benchmark Article:

• March 2023 — preliminary -306,000, final -187,000. Gap of 119,000.
• March 2024 — preliminary -818,000, final -598,000. Gap of 220,000.
• March 2025 — preliminary -911,000, final -861,000. Gap of 50,000.

The mean gap across those three is 129,700 (this article’s calculation), which is 1.64 times the entire March 2026 preliminary estimate. In all three the final came in smaller in magnitude than the preliminary, though three observations is a base rate thin enough to name as such rather than to lean on.

The longer record sits in table 4 itself, which gives final revisions for March 2015 through March 2025. Nine of the eleven are negative; the two exceptions are 2017 at +135,000 and 2022 at +506,000. The widest of the eleven in either direction is the 2025 final, at -861,000. At 79,000, the 2026 preliminary is smaller in magnitude than nine of those eleven; only 2018, at -16,000, and 2021, at -7,000, were smaller.

The release supplies its own yardstick in percent terms: “The annual benchmark revisions over the last 10 years have an absolute average of 0.2 percent of total nonfarm employment.” Averaging the absolute percent column in table 4 across 2016 through 2025 returns 0.19, which rounds to that figure and confirms the window (this article’s calculation; two of those years are published only as less than 0.05 percent, so the true average sits between 0.19 and 0.20).

In levels, which the release does not give, the same ten finals average 300,100 in absolute size (this article’s calculation). The March 2026 preliminary is 26.3 percent of that.

Put the three findings together and the shape is plain. The estimate is small by the standard of the last decade. It sits inside the range that preliminary-to-final movement has covered across the last three years, though one of those three gaps — 50,000, in 2025 — was smaller than the current estimate. And it will not touch a published series before February 2027.

What to Watch Next Week

• Whether a monthly figure appears anywhere. The release does not publish one. Any account that converts the estimate into a per-month number should be reproducing 79,000 divided by 12, and should be naming the April 2025 to March 2026 window it applies to.
• Whether the private line gets separated from the headline. The -178,000 and the -79,000 answer different questions, and a 99,000 government offset is what stands between them.
• The trailing average in the next Employment Situation. The 31,000 average monthly gain reported on September 4 rests on unbenchmarked data, and will keep resting on it through every release between now and February 2027.
• The anchor source. The counts behind the benchmark are, in the release’s words, “derived primarily from state unemployment insurance (UI) tax records that nearly all employers are required to file with state workforce agencies.”
• February 2027 as the only entry point. That is the date at which any of this reaches an official estimate, and the date at which -79,000 either survives or is replaced.

Concrete Framework

1. Read the reference month before the number. March 2026 is a level for one month, not a span of hiring.
2. Divide before comparing with monthly job growth. 79,000 over twelve months is 6,583, and it applies only inside the wedge window.
3. Take the private line on its own. It is -178,000, and the gap to the headline is a +99,000 government revision.
4. Treat sector rows as noisier than the total, on BLS’s own sampling-error sentence, and check whether components net or cancel. Trade, transportation, and utilities nets -97.6 out of 384.0 of gross movement.
5. Carry an uncertainty band drawn from the last three preliminary-to-final gaps: 119,000, 220,000 and 50,000, averaging 129,700. Two of the three are larger than the current 79,000 estimate and one is smaller.
6. Date the entry point. Nothing published changes before February 2027, so a series compared with itself today is comparing unbenchmarked with unbenchmarked, which is internally consistent.

Where This Doesn’t Apply

The distinctions above describe a measurement process, not a labor market call, and several conditions switch them off.

• A level revision is not nothing. Once applied, every month from April 2025 through March 2026 sits lower by a rising fraction of the amount. The nine months after March 2026 start from the revised March level and then take a freshly calculated net birth-death forecast on top of it. The argument here is against one particular restatement, not against the revision mattering.
• The wedge is an assumption, stated as one. BLS writes that the procedure “assumes that the total estimation error accumulated at a steady rate since the last benchmark.” The 6,583 a month follows from that assumption. It is not a measurement of when the error occurred, and if the error was concentrated in a few months the distribution is wrong even though the March level is right.
• Sector figures carry the larger error BLS names. The retail and transportation numbers are useful for showing that the aggregate is a residue. They are not sector conclusions, and the release says why.
• Three preliminary-to-final gaps are three observations. They also span an unusual stretch. The ten-year average absolute size of 300,100 includes benchmarks from 2020 through 2022, when the underlying conditions were not ordinary, and an average that includes them will not describe an ordinary year well.
• None of this reads on the level of employment itself. A smaller benchmark revision than last year’s says the two counts agreed more closely, not that hiring was stronger. Those are different statements about different objects.