The labour market is two surveys, and in October 2025 only one of them happened

The American labour market is measured twice every month by two entirely different methods, and the headline number comes from the more fragile of the two. That fragility became visible in October 2025, when one survey produced its figures and the other produced nothing at all — permanently. This page reads both at the Bureau of Labor Statistics, shows exactly which series have that month and which do not, measures the gap between the headline rate and the broader one, and counts how long unemployment kept rising after each recession ended.

The month one survey does not have

Eight series were requested from the Bureau of Labor Statistics public API in a single call on 13 August 2026, covering 2025 and 2026. Five of them returned a dash for October 2025. Three returned a normal value. The pattern is not random and is not about subject matter.

Everything missing is collected by asking someone in a defined reference week. Everything present is collected from records that still exist afterwards. The BLS itself explains the reference period in another context, quoted by the National Bureau of Economic Research: the monthly employment estimates pertain to the reference period containing the 12th of the month — for payroll employment, the pay period containing the 12th; for household employment, the week containing the 12th.

A payroll record for the pay period containing 12 October 2025 still exists in November and can be collected then. A household that was not interviewed in the week containing 12 October 2025 cannot be interviewed about it later in any way that produces the same statistic. That month of the household survey is gone permanently, and with it the unemployment rate, the U-6 rate, the participation rate and the employment-population ratio for October 2025.

The consequences reach further than the month itself. Any twelve-month comparison for October 2026 has no base. Any rolling twelve-month calculation spanning that month inherits the hole — the real-time Sahm rule, which uses a twelve-month rolling minimum of the unemployment rate, has no value for October 2025 for exactly this reason. And any chart drawn without checking will simply join September 2025 to November 2025 with a straight line and show nothing.

Where the labour market stood when this was written

The gap between the two published rates is the first thing to notice. U-3, the headline, was 4.1% in July 2026. U-6, which the BLS publishes alongside it, was 7.9% — 3.8 percentage points higher, and almost double. Both are official, both are current, and quoting either without naming it is the most common error in this subject.

The second thing is the direction of the participation rate. It was 62.5% in September 2025 and 61.4% in July 2026, a fall of 1.1 pointsover ten published months. The unemployment rate over the same span went from 4.4% to 4.1% — it fell. Those two movements are arithmetically compatible, because someone who leaves the labour force stops being counted as unemployed, and that is precisely why a falling unemployment rate is not on its own evidence of anything. The employment-population ratio, which has no such escape hatch, went from 59.7% to 58.9% over the same period.

The whole population of the unemployment rate

The seasonally adjusted unemployment rate runs monthly from January 1948 and the file read on 13 August 2026 has 942 values, one of which — October 2025 — is empty. Its lowest reading is 2.5% in May 1953 and its highest 14.8% in April 2020, a range of 12.3 points across seventy-eight years.

The most useful property of the series is one the NBER states in writing and almost nobody applies. It calls the unemployment rate a trendless indicator that moves in the opposite direction from most other cyclical indicators, notes that it sometimes rises before the peak of economic activity, when activity is still growing but below its normal trend rate of increase, and that it often continues to rise after activity has reached its trough. Here is that second property, measured over every recession since 1948.

In nine of the twelve the maximum came after the recession had officially ended, by one to nineteen months, and in only three did it fall in the trough month itself. The 1980 row carries a caveat that has to be stated rather than smoothed: its maximum of 10.8% in November 1982 lies inside the separate recession that began in July 1981, so the 28-month figure is an artefact of two recessions seventeen months apart and is not comparable with the others. Declaring that is the point; averaging it in would not be.

The NBER gives its own examples of the same effect, which are worth having beside the table: the unemployment rate reached a low level of 4.4 percent in May 2007 and rose to 5.0 percent by the peak of economic activity in December 2007, and after the March 1991 trough it continued to rise for 15 months.

The faster series, and what they cost in precision

Weekly initial claims are the fastest labour market number published, and the file read here has 3,109 weekly observations from 7 January 1967. The reading for the week ending 1 August 2026 is 199,000. The lowest in the whole file is 162,000 in the week to 30 November 1968, and the highest is 6,137,000 in the week to 4 April 2020— a single week roughly thirty-eight times the size of the quietest week ever recorded.

The NBER is unusually direct about how much weight to give this series: the monthly BLS employment estimates are a more reliable indicator of labour market developments than unemployment insurance claims, so we normally place little weight on the claims data. It then names the exception: we judged that the special circumstances of March 2020 caused the claims data to be unusually informative, because the weeks ending 21 and 28 March 2020 saw historically unprecedented levels of new claims, suggesting a collapse of employment between the reference week and the end of the month. A weekly series was worth reading precisely when the monthly one could not move fast enough, and not otherwise.

Job openings, from the JOLTS programme, gave 7,359 thousand for June 2026 and did have an October 2025 value of 7,170 thousand. It is also the slowest of the series here: in a request made on 13 August 2026, its latest month was June 2026 while payroll employment already had July.

Five rules for reading these numbers

• Name the rate. U-3 was 4.1% and U-6 was 7.9% in the same month, from the same publisher.
• Read the participation rate beside it. Between September 2025 and July 2026 unemployment fell 0.3 points while participation fell 1.1 and the employment-population ratio fell 0.8.
• Know which survey a number came from. One of the two has no October 2025 and the other does, and the reason is how each is collected.
• Do not date a cycle with unemployment. Its maximum came after the recession ended in nine of the last twelve episodes.
• Check the latest month per series, not per subject. On the same day, payroll employment had July 2026 and job openings had June 2026.

What these sources do not say

• The BLS website could not be read. Three requests returned 403 on 13 August 2026. No official definition of any rate, no notice about the missing month and no release schedule appears here, because none was opened.
• The household survey for October 2025 cannot be recovered. A reference-week interview missed is not a delayed statistic, it is an absent one.
• The unemployment rate is not a measure of hardship. It is a ratio to the labour force, and the participation and employment-population figures beside it moved in the opposite direction over the same ten months.
• No wage, earnings or inequality figure appears here. None was read for this page.
• Nothing about any other country. Labour force definitions differ between statistical agencies and no foreign source was opened.
• No claim about what employment data do to any market. No price series was read for this page.

How to verify this yourself

1. Post one request to the BLS public API listing LNS14000000 and CES0000000001 for 2025 and 2026. No key is required. Compare the October 2025 rows.
2. Add LNS13327709 to the same request and compare July 2026 with the headline rate. The gap should be 3.8 points.
3. Add LNS11300000 and LNS12300000 and compare September 2025 with July 2026. Both fall while the unemployment rate also falls.
4. Download UNRATE and, for each NBER recession, find the highest reading between the peak and two years past the trough. Compare each date with the trough month.
5. Download ICSA and find the maximum. It should be 6,137,000 in the week ending 4 April 2020.
6. Check the latest month returned for each series in the same request. They are not all the same.

Where to go next on this site

• Recession indicators — the Sahm rule built on this series, and the hole it inherits.
• The economic calendar — when each of these surveys is published, and by whom.
• Stagflation — what the record shows when this rate rises while prices rise too.
• All macroeconomics guides — the area index, one click away.

Sources opened for this page

Each source below was opened and read on 13 August 2026. Verified on 13 August 2026.

• US Bureau of Labor Statistics — Public Data API v2, series LNS14000000, LNS13327709, LNS11300000, LNS12300000, CES0000000001, CES0500000003, CUUR0000SA0 and JTS000000000000000JOLUsed for: the July 2026 values of 4.1%, 7.9%, 61.4%, 58.9%, 158,858 thousand and 37.62 hours, the June 2026 job openings of 7,359 thousand, the September 2025 comparison values, and the finding that October 2025 is a dash for every household-survey and price-collection series while the establishment-survey and JOLTS series carry values. Read 13 August 2026.
• Federal Reserve Bank of St. Louis — FRED series UNRATE, U6RATE, CIVPART, EMRATIO, PAYEMS and ICSAUsed for: the 942 monthly unemployment values from January 1948 with the extremes of 2.5% in May 1953 and 14.8% in April 2020, the empty October 2025, the peak unemployment month of each of the twelve post-war recessions, and the 3,109 weekly claims observations with their low of 162,000, high of 6,137,000 and latest value of 199,000. Read 13 August 2026.
• National Bureau of Economic Research — Business cycle dating procedure, frequently asked questionsUsed for: the statement that the unemployment rate is a trendless indicator moving opposite to most cyclical indicators, the 4.4% to 5.0% example of 2007, the fifteen months of rising unemployment after the March 1991 trough, the reference-period definition containing the 12th of the month, and the assessment of unemployment insurance claims with the March 2020 exception. Read 13 August 2026.
• National Bureau of Economic Research — US Business Cycle Expansions and ContractionsUsed for: the twelve post-war peak and trough months against which the unemployment maximum of each episode was measured, and the note that the business cycle data were last updated on 14 March 2023. Read 13 August 2026.
• www.bls.gov — three requests to the CPI home page, the CPI news release and the release scheduleUsed for: the record that all three returned HTTP 403 on 13 August 2026, so no BLS definition of U-3 or U-6, no notice about October 2025 and no release schedule was read for this page. Read 13 August 2026.

Change log

• 13 August 2026— page rewritten from scratch against the BLS public API, six FRED labour series and the NBER dating FAQ, all opened that day. Removed from the previous version: the unemployment rate presented as the measure of the labour market with no mention of U-6, which was almost double it in the same month; a falling unemployment rate described as improvement while participation was falling faster; a claim that unemployment signals the end of a recession, contradicted in nine of the last twelve episodes; jobless claims treated as a primary indicator, which the NBER explicitly says it normally weighs lightly; and no mention that the household survey has no October 2025 at all.

Vextor Capital is not authorised under MiFID II as an investment firm, is not registered with the SEC, FINRA, the CFTC or the NFA, is not an investment adviser, holds no position in any instrument named on this page, and receives no payment from any bank, broker, exchange, index provider or data vendor. Every outbound link on this page points to a statistical agency, a central bank or a public data series.