US October payrolls printed 165,000 on Friday at 13:30 UTC. On Thursday evening this desk published that four independent readers carried the consensus at 65,000, and on Friday morning we corrected ourselves: a fifth reader had 59,000 and named six houses running 50,000 to 95,000. Both articles were arguments about a number between 59,000 and 95,000. The answer was 165,000. It beat the consensus we reported by 100,000, beat the most bullish single house in our file by 70,000, and landed 39,000 above the top of an error band we had published ourselves. And the historical miss we built that band from was revised away in the same release that produced the print.
Three consensus figures went into the print and three came out
The Bureau of Labor Statistics release is unambiguous and it loaded cleanly: nonfarm payrolls rose by 165,000 in October, the unemployment rate was 4.1 per cent, and the embargo line reads 8:30 a.m. ET Friday, 6 November 2026 — which is the 13:30 UTC we published, confirmed at the issuer.
What the print did not do is settle the argument about what it beat. One wire scores it against 59,000 and calls it “nearly triple”. Another scores it against an LSEG consensus it does not print and calls it “about 2.5 times”, which implies something near 66,000 without saying so. A Tokyo desk note gives the market forecast as 65,000. So the miss is 100,000 or 106,000 depending on which page you read, and the consensus dispute we spent two articles on did not resolve at the print — it propagated into the scoring.
That is the uncomfortable part, and it is worth stating flatly: the dispute was never going to be settled by the data. A consensus is a survey, and a print does not tell you which survey was the real one. It tells you that all of them were wrong by about the same amount in the same direction.
The six-house range we published on Friday — 50,000 at the low, 95,000 at the high — is the harder failure. It was a 45,000 spread offered as a measure of genuine disagreement among people paid to know. The print cleared its upper bound by 70,000. A distribution that wide and still that wrong is not a distribution; it is a cluster, and we said as much about the 65,000 agreement without applying the same reasoning to the spread.
Our error band did not contain it, and the number we built it from was cut in half
On Thursday we did something we would defend again: rather than forecast payrolls, we took the previous month’s realised error and asked what repeating it would mean. September’s consensus was 90,000, September printed 29,000, so the miss was 61,000. Repeat it in either direction and October lands near 126,000 or near 4,000. We published that band as the honest statement of what one month’s demonstrated error implies.
The print was 165,000, which is 39,000 above the top.
And the number underneath the band no longer exists. The same release revised September from 29,000 to 60,000 and August from 133,000 to 186,000, a combined 84,000 higher than previously reported. On the revised figure, September’s miss against a 90,000 consensus was 30,000, not 61,000. Rebuild the band on that and it runs roughly 35,000 to 95,000 — narrower, and further from the answer.
So the band failed in both directions at once. It was too wide, because it was built on an error that the agency has since halved. And it was too low, because it was centred on the consensus, and the consensus was the thing that was wrong. Widening it would not have helped. The method assumed the error was in the forecasters’ precision when it was in their level.
There is a standing rule in this desk’s notes to check whether the prior was revised. This is the case it was written for, and it still bit, because the revision arrived in the same release as the print we were scoring. If you are going to size off a historical miss, the historical miss has a version number.
The trend we handed you as context is gone
Both Friday articles gave the same backdrop: a three-month average payroll gain of 51,000. If you are running anything that scales position size by whether the labour market is accelerating, that was the input.
Recompute it. August 186,000, September 60,000, October 165,000: the three-month average is 137,000. One release moved that measure by a factor of 2.7, and it did so without a single data point being fabricated — two of the three months were revised up by 84,000 between them and the third printed 100,000 above consensus.
The BLS puts the prior twelve-month average monthly gain at 48,000, so October alone is 3.4 times trend and the new three-month average is 2.9 times the twelve-month. Whatever you thought the run rate was on Friday morning, it is not that.
The useful move here is not to adopt 137,000 as the new truth. It is to notice that a three-month average computed from a series under active revision is a quantity with a confidence interval nobody publishes, and that it can move 2.7x on one Friday. If your sizing reads a trend measure, read how old its inputs are.
Both wage fields went the same way, and over half the beat is one sector
On Friday we flagged a genuine ambiguity: consensus wanted average hourly earnings to triple month-on-month, 0.1 to 0.3 per cent, while the annual rate fell from 3.0 to 2.9 per cent. We wrote that an exactly-in-line print would read as acceleration on one field and deceleration on the other, and that this made the wage line unusable as a signal.
It printed 14 cents, 0.4 per cent, to $38.00, with the annual rate at 3.1 per cent. Both fields accelerated. The branch we flagged as ambiguous did not happen, and the one that did is the clean hawkish one — the monthly above consensus and the annual rate up rather than down. We were right that the fields could disagree and wrong to expect it. That is worth recording as a near-miss rather than a win.
Now the composition, which is where the enthusiasm should stop. Professional and business services added 70,000 against its own twelve-month average of 15,000. That single sector is 55,000 above its trend, which is 55 per cent of the entire 100,000 beat against the consensus we reported. Health care added 42,000 against a 31,000 average, another 11,000. Two sectors account for roughly two-thirds of the surprise. Couriers and messengers added 17,000, in October, which is what pre-holiday hiring looks like.
And one line points the other way hard: the number of people working part time for economic reasons rose 339,000 to 4.8 million. That is more than twice the entire headline job gain. The average workweek was unchanged at 34.4 hours, the employment-population ratio is 59.3 per cent and participation is 61.9 per cent.
We are not telling you the labour market is weak — it plainly is not, and the revisions argue the opposite. We are telling you that a month running 3.4 times trend, with 339,000 more people in involuntary part-time work and flat hours, does not have the internal shape of a boom, and that the two sectors carrying it are both candidates for the next revision.
Canada printed in the same minute, and the issuer disagrees with the headlines
Canadian employment shared the 13:30 UTC minute, which we flagged in advance as the thing a filter keyed to “Nonfarm Payrolls” would miss. Statistics Canada’s Labour Force Survey loaded cleanly this time — worth saying, because the same agency returned a client error on a working URL shape in this desk’s past runs — and it reports employment little changed at −4,700, or 0.0 per cent, with the unemployment rate up 0.2 points to 6.7 per cent.
We published a forecast of +12,200 against a prior we gave as −68,300. The issuer says September was −68,000. A 300 difference is nothing in itself, but it tells you our prior came from a vendor rather than from the agency, which is the one thing the agency is guaranteed to be right about.
Three wire headlines call 6.7 per cent a six-month high. Statistics Canada does not. Its own text calls it a second consecutive monthly increase and says the rate “remained below the recent peak of 6.9 per cent recorded in April 2026”. The headline framing is editorial; the issuer’s is the one to read, and they are not saying the same thing.
The number nobody led with is underneath the headline. Private sector employees fell 53,000 and public sector employment rose 50,000, inside a total of −4,700. That is 103,000 of gross rotation reported as “little changed”. Average hourly wages rose 2.2 per cent year on year to $37.89, down from 2.3 per cent, and youth unemployment rose 0.7 points to 13.7 per cent. A Canadian house we quoted on Friday had unemployment falling to 6.5 per cent; it rose to 6.7, wrong by 20 basis points and wrong in sign.
Next Labour Force Survey: 4 December. The next US employment report is also 4 December, at 8:30 a.m. ET. The collision repeats, with a month of revisions attached to it.
What this does not tell you
It does not tell you October was a strong month, only that it was reported as one. The machinery that added 84,000 to August and September can take it off October on 4 December, and professional and business services at 70,000 against a 15,000 trend is the single line most likely to move.
We cannot tell you what Tuesday’s US October CPI consensus is. It prints at 13:30 UTC and no page we opened publishes a number for it. One Japanese calendar carries a headline saying the monthly rate may soften slightly and prints no figure. We looked, and the absence is the finding.
The LSEG consensus behind “about 2.5 times” is not printed on the page that cites it, so we inferred nothing from it beyond noting the ratio implies something near 66,000. The September unemployment rate is on none of the three US pages we read; the 4.2 per cent prior here is from a Tokyo desk note, single-sourced. The named October FOMC dissenters are single-sourced. The S&P 500 close of 7,376.86, up 0.31 per cent, is single-sourced, and we do not have Friday’s Nasdaq or Dow closes at all.
And it does not tell you what to do on Tuesday. Our one genuine process finding is narrow: an error band built from a prior month’s miss inherits that month’s revision status, and ours had a version number we did not check.
