The jobs report has
a second draft.

On July 3, 2025, the first estimate for June showed147,000 added jobs. By the third estimate, published September 5, the figure was a 13,000 decline. A later release, on March 6, 2026, put the decline at 20,000.

Those are three readings of the same employment month. The difference matters: “payrolls grew” and “payrolls shrank” are different descriptions of June. Neither, on its own, can tell us why the change happened—or settle whether a recession was underway.

A fast answer, with more evidence to come

The jobs report solves an awkward problem. People need to know what is happening before every employer has reported it. The first payroll estimate gives a timely reading from the information available. Later responses, corrections and another pass through seasonal adjustment can change that reading. It is useful precisely because it arrives early—and that makes its publication date part of its meaning.

There is another, broader revision. Each year, BLS benchmarks the payroll survey against more comprehensive employment records, principally unemployment-insurance tax records. Seasonal adjustments can change earlier months as well. A revision that crosses that update cannot be explained simply as “late surveys came in.” The chart above describes the arithmetic between releases; it does not divide responsibility among those processes.BLS explains the revision process.

Most useful is the contrast

June is worth noticing. So is the month after it. July 2025 began at 73,000 added jobs and reached 72,000 at the third estimate: a difference of just 1,000. The March 6 vintage puts it at 64,000. All three retain the same estimated direction.

Revisions also move upward. March 2024 went from 303,000 added jobs to 310,000 at the third estimate. Its March 6, 2026 reading is lower, at 228,000. “The next estimate” and “a much later estimate” are different comparisons. A revision has both a starting point and a destination.Check the BLS first and third estimates.

The blank is part of the story

October 2025 has no blue dot. The lapse in federal appropriations disrupted publication. BLS lists no first preliminary estimate for that month. October data first appeared in the December 16 release, but BLS classifies that estimate assecond preliminary. Calling it “first” here would make a tidy chart and a misleading comparison.

The interruption also delayed August’s third estimate until December 16. Normally, the third estimate arrives two months after the first; the dates here show when it actually arrived. A blank is not zero, and a regular-looking calendar is not evidence of a regular release schedule.See the release archiveandthe December 16 explanation.

A year is more than twelve headlines

The January 9, 2026 vintage showed 584,000 jobs added during 2025. The February 11 benchmark release lowered that to 181,000. After the next release revised December again, the March 6 vintage showed 116,000.

Those totals each subtract December’s employment level from the previous December’s level in one dated edition. Adding twelve first-release changes would stitch together twelve different versions of history. It would answer a different question.The benchmark releaseandMarch 6 releaseshow the changes.

The practical habit is small: keep the date attached. Read the first estimate for its timely information. Return for the later draft. And when the draft changes, let the explanation change with it—without pretending the earlier reading never existed.