Summary of the BLS preliminary benchmark revision, August 28, 2026

  • The preliminary revision to March 2026 total nonfarm employment is −79,000, or −0.1 percent, below the 10-year absolute average of 0.2 percent.
  • The three prior preliminary revisions were −306,000, −818,000 and −911,000 — the run of record-sized cuts ends here, though the revision is negative for a fourth year.
  • Private employment was revised down 178,000; government was revised up 99,000.
  • Information (+3.0 percent) and transportation and warehousing (+2.0) were revised up most; wholesale (−1.4) and retail trade (−1.0) down most.

Each year in late summer, BLS checks its survey-based payroll estimates against something close to a census: state unemployment-insurance tax records covering nearly every employer. For three straight years that check cut the count by hundreds of thousands, feeding an argument — made in our own July jobs coverage among others (July deep dive) — that the monthly jobs numbers overstated the labor market. This year’s check found the estimates were about right.

The record-sized cuts end, at least for this vintage

A −0.1 percent miss is ordinary — smaller than the decade’s average, and the opposite of what forecasters looked for: Goldman Sachs, per pre-release coverage, expected the first upward revision in three years, between +50,000 and +450,000. The practical consequence runs through this year’s payroll argument: the survey entered 2026 with no accumulated overcount behind it, so the weak stretch that cut September hike odds — including July’s −23,000 — started from an honest base. Whether the months since March hold up is next year’s check; if anything, this one found government employment undercounted.

The industry detail rearranges the picture more than the total does. Information — the supersector containing data hosting alongside publishing and telecom — was revised up 87,000, or 3.0 percent, the largest percentage move in the table; transportation and warehousing up 135,000. Retail trade was revised down 155,000 and wholesale 86,000. The table stops at the supersector level, so which industries inside information drove its 87,000 is not published here.

The revision covers only the year through March, and it is itself preliminary

The preliminary revision measures one thing: the accumulated error in the survey between March 2025 and March 2026. It says nothing about months since March, and it is itself preliminary — the final figure arrives in February 2027 with the January jobs report. Last year the preliminary of −911,000 finalized at −861,000; the year before, −818,000 became −598,000. The direction has held even when the size moved.

Chair Warsh, speaking at Jackson Hole the same morning, called labor markets “consistent with full employment” and low monthly job gains the natural result of barely growing labor supply. The benchmark did not contradict him. The August jobs report lands September 4, the last one before the September 15–16 meeting — the first in three years to arrive with the survey’s most recent annual audit clean.

Sources: BLS, Current Employment Statistics Preliminary Benchmark Announcement, August 28, 2026 (archived), with the release text and Table 1 also preserved in the GeoMean data files. Prior years’ preliminary and final revisions from BLS CES Technical Notes Table 5 (cestn.htm), compiled in the GeoMean benchmark-revision history, 1979–2025. Warsh remarks: Federal Reserve; companion note: Warsh at Jackson Hole. The revision applies to the March 2026 reference month; official estimates are unchanged until February 2027.