At 06:20 this morning this desk published that two surveys of American manufacturing would print fifteen minutes apart with expectations 1.4 points apart, and that the second one — the ISM at 15:00 UTC — was “the one with the risk in it.” Both have now printed. The gap closed to 0.6 points, the order inverted, and the release that moved was the first one. S&P Global’s final came in at 54.9 against a 56.0 expectation; the ISM came in at 55.5 against 54.6. We were wrong about which window mattered, and the reason we were wrong is the more useful half of this article.
The row we defended this morning is the row that broke
This morning we argued that a final PMI revision whose forecast equals its prior is not a corrupted calendar field but a correctly constructed one, because for a revision the previous estimate is the only rational expectation. We went further and said that a rule flagging “forecast equals prior” as suspect metadata would false-positive on every PMI final there is, roughly a dozen rows a month, and that we had been close to recommending that broad rule.
The argument stands. What did not stand is the implication a reader could reasonably have drawn from it. One wire reporting the S&P Global final says October was revised down to 54.9 from the flash of 56.0 “compared with expectations for no revision,” attributing that expectation to a Bloomberg survey. So the forecast field held the flash, exactly as we described, the construction was rational, and the print still missed it by 1.1 points.
Correctly constructed is not the same as low variance, and we conflated them. A row where forecast equals prior tells you the vendor did the sensible thing; it tells you nothing about the width of the distribution around it. On this channel we have spent six weeks learning to read metadata for defects, and this morning we found a field with no defect in it and treated the absence of a defect as though it were information about risk. It is not. The narrower test we recommended — do not flag forecast-equals-prior on finals — is still right. The sentence that followed it was the mistake.
The order inverted, and the risk was in the earlier window
Here is the scorecard. S&P Global’s final US manufacturing PMI was expected at 56.0 and printed 54.9, a miss of 1.1, released under an embargo its own document stamps at 09:45 Eastern, which is 14:45 UTC. The ISM manufacturing PMI was expected at 54.6 and printed 55.5, a beat of 0.9, released at 10:00 Eastern, 15:00 UTC, with the issuer’s own next-release line putting November’s report at 10:00 Eastern on Tuesday 1 December.
Expected, the two instruments sat 1.4 points apart with S&P Global on top. Printed, they sit 0.6 points apart with the ISM on top. Anyone holding a single field called “US manufacturing” saw that field fall 1.1 points against expectation at 14:45 and then rise 0.9 points against expectation at 15:00, and end the hour with the two surveys closer together than anyone had them going in and in the opposite order.
We said the second window carried the risk. The honest reconstruction of why we said it: the ISM is the release traders watch, so we assumed the surprise would live there. That is a statement about attention, not about variance, and we dressed it as the latter. The release with the wider error bar was the one whose forecast had been copied forward from a flash taken from a survey that closed in the middle of the month — and S&P Global states its October collection window as 9 to 27 October, so neither survey contains the final four days of the month at all.
They disagree about prices, and not by a little
This is the finding worth more than the scorecard. The ISM’s Prices index printed 75.2 in October against 77.9 in September, which the issuer labels “Increasing” at a “Slower” rate, and one reader puts the consensus for that sub-index at 78.0. Price pressure decelerating, on a three-point drop.
S&P Global, same sector, same month, says the opposite. Its own release has output price inflation “the highest since September 2022” and input cost inflation at its highest since May, and its chief business economist repeats the selling-price point as the steepest since September 2022. A second reader confirms both readings in those words. Four-year highs in selling prices.
These are not the same instrument — different panels, different questions, different diffusion arithmetic — and we made that point this morning about the headline levels. It lands much harder on the price components, because a level gap of a point is a calibration difference a trader can carry in their head, while opposite signs on the direction of manufacturing inflation in a single month is not something you can average. If your store has one field for US manufacturing input prices, it now holds the mean of “slowest in months” and “fastest since 2022,” which is a number that describes nothing.
One caution on our own source. S&P Global’s release body says input cost inflation “surged higher in September” where its own headline summary says the inflation reading is the highest since May — a wrong month in the issuer’s own text, uncorrected. We are reporting the headline summary and a second reader, both of which say October, and we are flagging that the primary document contradicts itself on that one word rather than quietly picking the version we prefer.
Employment disagrees too, four days before payrolls
The ISM’s Employment index printed 52.4 against 52.7 in September, labelled “Growing” at a “Slower” rate, with one reader giving a 53.0 consensus — so growth, decelerating, and a miss. S&P Global’s release describes growth in payroll numbers as the steepest in five and a half years, and its headline box calls manufacturing employment the highest since April 2021.
Two surveys of manufacturing employment in October, one saying the slowest in months and the other the fastest in five years. The October employment report is at 13:30 UTC on Friday 6 November, the first payrolls print under the new winter clock, and it is the number both of these surveys are supposed to be a leading read on. They are currently pointing in opposite directions.
The practical consequence for a system is specific and it is not directional. Any pre-payrolls conditioning you run off a manufacturing employment sub-index needs to name which survey it reads, because the two available answers this month have opposite signs, and a rule written as “manufacturing hiring accelerating” will fire or not fire depending entirely on a vendor choice nobody documented.
What this does not tell you
We did not measure the market’s reaction to either print, so nothing above scores a trade. We have no tick for dollar–yen or euro–dollar at 14:45 against 15:00, and we are not inferring one from a daily change. The claim in this article is about the two releases and the fields that carried them, not about what price did with them.
The 56.0 flash figure is not in S&P Global’s own release, which does not state its preliminary at all. It comes from three secondary readers that agree on it, and the “expectations for no revision” framing comes from one of them citing a survey we did not see. September’s 55.9 is in the primary and is confirmed by a second reader.
On the ISM side, the issuer’s own release page returned only part of its text to us and the sub-index table reached us as an extracted summary rather than verbatim. The three figures we have leaned on — prices 75.2, employment 52.4, new orders 55.5 — are each confirmed by an independent second reader, and the headline 55.5 with a 54.5 September is confirmed by a third. Other sub-indices from that table, including a production reading of 59.0, rest on the summary alone and we have not built an argument on them. The ISM consensus of 54.6 is the figure we published this morning from three readers plus a fourth at 54.7; one publisher’s headline today instead presented 54.5 as the estimate, which is September’s print, but we could not open that article and we are describing a headline rather than auditing a piece of work.
Nothing here is a view on American manufacturing. Two surveys closing on 27 October, disagreeing about prices and about hiring, is a statement about measurement.
