Jobs, Revised

Two years of payrolls: first prints against latest vintages.

Coverage
Sep 2024 – Aug 2026
Vintage
Sep 4, 2026
Series
Total nonfarm, s.a.
Source
BLS CES via FRED
View
Order
Theme
First print Final / latest Revision Δ (final − first)
+2,656k
Headline total, sum of first prints
+1,318k
Final total, sum of latest vintages
−1,338k
Revision drag, ≈1.34M jobs revised away
17 / 24
Months revised down; 5 up, 1 pending
Exhibit 01 · Monthly change

The headlines overstated growth by half

First prints summed to +2,656k across 24 months; latest vintages sum to +1,318k. Select a row for the full vintage path and industry drivers.

Coverage
Sep 2024 – Aug 2026 · 24 ref months
Vintage
Sep 4, 2026 (Aug release day)
Units
Thousands, seasonally adjusted
Source
BLS Employment Situation · PAYEMS
Ref month1st print2nd print3rd printFinalΔContributorDetractor

“Final” is the Sep 4, 2026 vintage, not literally final — Aug 2026 is preliminary and July already flipped −23k → +21k on revision.

No October 2025 release was published (shutdown); its CES figure arrived with November on Dec 16. October household data were never collected.

Findings

Executive summary — initial release vs final

  • Headlines overstated growth by ~2×. Summed first prints +2,656k; summed latest vintages +1,318k. About 1.34M reported jobs did not survive revision.
  • The bias flipped with the year. Late 2024 revised both ways and stayed strong (Dec 256k → 237k). From Jan 2025 nearly every month revised down — Jan +143k → −48k, Mar +228k → +67k, Jun +147k → −20k, Aug +22k → −70k. Calendar 2025: ~+1,208k first-reported vs ~+116k latest.
  • The benchmark explains much of it. BLS (Feb 11, 2026) cut 2025 growth +584k → +181k and the Mar 2025 level by 862k (NSA), citing CES birth-death overcount. The latest vintage runs harsher still for early 2025 after further monthly revisions.
  • 2025 was essentially flat; 2026 re-accelerated unevenly. BLS: 2025 averaged ~+15k/mo post-benchmark vs +49k/mo pre-benchmark. 2026 brought a genuine Feb contraction (−92k → −156k) and a soft July (−23k → +21k).
  • One engine, one anchor. Health care (with social assistance) leads ~18 of 24 months — in Aug 2025 (+31k) it was the only gainer. Federal government is the modal detractor from Feb 2025 on (e.g., Jan 2026 −34k).
Fall 2025 data hole. No October 2025 release (43-day shutdown; household survey never collected). October CES (−105k as published, −140k latest) arrived jointly with November on Dec 16; September (+119k) was 7 weeks late. October unemployment is permanently missing.
Next

Suggested additions

  1. Levels + 3-month average — smooths strike, shutdown, and seasonal noise.
  2. Unemployment + participation — joblessness (Sep 2025: 4.4%) vs labor-force exit (Jun 2026: 4.2%).
  3. Private vs government split — isolates the policy-driven drag from private momentum.
  4. Benchmark flags + birth-death contribution — which “finals” moved on QCEW counts vs routine revision.
  5. Wages + hours, JOLTS + claims, diffusion index — was flat employment still tight on pay, and how narrow were the gains?
  6. Vintage toggle — Dec-2025 “2025 = +584k” vs Feb-2026 “+181k” vs today “~+116k”.

Sources & methodology

  • BLS Employment Situation archives — headline + industry narrative per month (empsit_10042024 through empsit_08072026). Archive: bls.gov/news.release/archives · Schedule: release calendar.
  • BLS benchmark & shutdown notices — Mar 2025 level −862k (cesbmk); 2025 growth +584k → +181k (Jan 2026 release); revised fall-2025 dates (lapse notice).
  • FRED / ALFRED, PAYEMS (BLS CES source) — latest levels via FRED; 1st/2nd/3rd prints as same-vintage over-month level differences from ALFRED vintages dated the day after each release. Reproduces every BLS headline exactly (e.g., Sep 2024: 159,105 − 158,851 = 254).
  • Industry drivers — first-listed gainer/loser in each release’s BLS narrative. Totals are sums of rounded monthly changes.