At 15:00 UTC today a calendar we read lists six separate rows: the October ISM manufacturing survey, three of its sub-indexes, and September construction spending. The 10am New York option cut is at 15:00 as well, for the first Monday since the clock change. Fifteen minutes earlier the final October manufacturing PMI prints. Nothing at all is scheduled at 13:30, the slot the United States data block moved into on Sunday. And the row most traders will have flagged as the day’s second event — a Federal Reserve speaker at 16:15 — is a man whose role at that event, per the issuer’s own agenda, is to moderate somebody else’s keynote in a room that has barred the press.
Six rows, one minute, and the cut in the same minute
Here is the day as one calendar prints it, converted where needed and stated in UTC. 14:45 the final October manufacturing PMI, prior 55.9 against an expected 56. 15:00 the ISM manufacturing survey, prior 54.5 against an expected 54.6. 15:00 ISM manufacturing employment, prior 52.7. 15:00 September construction spending, prior +0.9 per cent. 15:00 ISM manufacturing new orders, prior 55.3. 15:00 ISM manufacturing prices, prior 77.9 against an expected 77. Then 16:15 the Federal Reserve appearance, 16:30 the three-month and six-month bill auctions, 19:00 the senior loan officer survey, 20:00 the Treasury’s refunding financing estimates.
The release time is now four-sourced across four time zones, which is as settled as this channel gets. One calendar gives 03:00 PM under a column headed GMT. A Japanese broker’s morning note gives 24:00 Japan time. A dedicated event page gives 10:00 EST and glosses it as 15:00 London. A European market briefing prints its whole schedule in Prague time and puts the ISM at 16:00 CET, which in November is UTC plus one. Every one of those is 15:00 UTC. Our own slot this morning had this at 15:00 off three readers; it is four now, and the matter is closed.
The thing worth doing something about is the clustering, not the hour. A sub-index is not a separate event in the sense a news filter means, because all four ISM numbers arrive in one press release at one instant — but they are four separate rows in a calendar, and a system that counts rows to decide how wide a window should be will read six events at 15:00 and open something enormous. Count releases, not rows. There are two releases at 15:00 today: the ISM report and construction spending. That is still one more than most desks have on the board.
And the empty 13:30 slot is the quiet hazard. Every United States release moved an hour later on Sunday, which means the window a lot of systems still hold at 12:30 or 13:30 now guards nothing at all today, and will guard payrolls on Friday.
The prior is closed at 54.5, and one calendar still has July
This morning our Systems desk published tonight’s ISM consensus as 54.6, 54.6 and 54.7 across three readers and noted that the prior at 54.5 came from a single reader. That single-sourced figure is now closed. A second, independent reader gives September 2026 at 54.5 twice over — in its body text, which says the index edged down to 54.5 in September from 54.6 in August, and in its own calendar row, where 54.5 sits in the previous-value column for tonight’s release. Two readers, two placements, same number. Call it settled.
What it is not is issuer-confirmed. The Institute for Supply Management is the issuing body and its report pages returned a server error on both the index and the September release. So every ISM figure in this article is a secondary reading of a document we could not open, and we say so again in the footer because it is the kind of thing that should cost a sentence every time it happens.
Now the part that answers a question we asked out loud twelve hours ago. Our Systems piece said it had not checked whether any vendor calendar’s stored rows actually updated across the weekend, and called that tomorrow’s test. We did not get the row it meant — the release time — but we got the adjacent one, and it is worse. A dedicated event page for tonight’s ISM carries, in its labelled “Prior” field, the value 55.6 per cent, marked July 2026. July was three releases ago. The page itself was last updated on 25 August, which is why. Its own sidebar, further down the same page, correctly describes the 1 September release printing 54.6 against a 55.2 consensus and down from July’s 55.6 — so the page contains the right history and a stale headline field at the same time, and the consensus field reads “Not yet published”.
A trader scoring tonight’s print against that page would mark 54.6 as a 1.0-point miss rather than a one-tenth beat. The number is not wrong, which is what makes it dangerous: 55.6 was genuinely the ISM manufacturing PMI, in a month nobody is trading tonight.
The 54.6-versus-54.7 spread is a column, not a disagreement
Two of the three consensus readings in circulation are 54.6 and 54.7, and we spent part of this morning treating that tenth as a mild disagreement between forecasters. It may not be one.
One vendor’s ISM page publishes two separately labelled columns for tonight’s release in the same row: Consensus 54.6 and TEForecast 54.7. The first is its reading of the market expectation. The second is its own house model. They differ by exactly the tenth that separates the readers.
We are not claiming anyone copied anyone. We have no evidence of that and will not invent it. The claim is narrower and more useful: the two numbers in circulation for tonight’s ISM are precisely the two numbers one widely-scraped table prints side by side under different labels, so before you treat a one-tenth gap as forecaster dispersion, establish that your two readers sampled the same column. The same page does this again elsewhere — it also carries a quarter-end model figure of 55.00, which is a third number, for a different object, in the same place.
This is the sixth distinct field in about seven weeks where this desk has found the problem to be the field rather than the figure: a consensus in the wrong field, a rate without its instrument, an average without its window, a clock without its zone, an event without its type, an event without its time — and now two forecasts of different provenance in adjacent columns with no provenance label on either when they are quoted onward. The numbers in this market are mostly fine. The labels are where the money leaks.
Construction spending has three consensus figures and two are the same vendor
The other 15:00 release is the clean demonstration. September construction spending has a prior of +0.9 per cent, which is agreed, against a July figure that one page describes as a downwardly revised 0.1 per cent decline. The expected value is not agreed at all:
- +0.5 per cent — a Japanese broker’s morning event table.
- +0.6 per cent — one vendor’s construction spending indicator page, labelled Consensus.
- +0.4 per cent — the same vendor’s master calendar page, in the consensus position for the same release.
Both of that vendor’s pages agree on its house model figure at +0.2 per cent, and disagree with each other about the market consensus by 20 basis points. One publisher, two pages, two versions of one field, read within the same minute. We have logged this failure shape before with a publisher running two versions of a dataset under two bylines; this is tighter, because it is two surfaces of one product.
Practically: a 20-basis-point spread on a monthly series whose prior is +0.9 means the release can print +0.4 and be simultaneously an in-line result and a two-tenths miss depending on which page your scoring rule reads. If you have a rule that trades construction spending surprises, it is not currently measuring a surprise. Our own position is to carry the range and score nothing on it tonight.
The 16:15 row is a moderator, and two Federal Reserve releases are on nobody’s calendar
Two FX-facing calendars have a Federal Reserve row at 16:15 today. One labels it “Fed Goolsbee Speech”. The other, printing in Prague time, lists “17:15 Fed speaker Goolsbee”, which is the same instant. A Japanese broker’s note says something different: that the Chicago Federal Reserve president is serving as moderator of a discussion.
The Japanese reader is the one that is right, and the issuer settles it. The Federal Reserve Bank of Chicago’s own event page for its Financial Markets Group Fall Conference, dated 2 November 2026, 9:00 am to 5:00 pm Central, gives a 10:15 am item titled “Keynote and Fireside Chat”. The keynote speaker is listed as Andréa M Maechler, Deputy General Manager of the Bank for International Settlements. The moderator is listed as Austan Goolsbee, President and CEO of the Federal Reserve Bank of Chicago. 10:15 Central Standard Time is 16:15 UTC, which matches both vendor rows to the minute — the hour was never the problem.
Three further facts from the same pages, each of which independently reduces what that row is worth. The conference is invitation-only and conducted under the Chatham House Rule, with no access granted to media — so there is no mechanism by which a quotable policy line reaches a wire in real time. The Federal Reserve’s own FOMC membership page, last updated 28 October 2026, lists him among the alternate members for 2026, not the voting ones. And the Board’s own November 2026 calendar lists no speaking engagement whatsoever on 2 November.
What the Board’s calendar does list for today is a senior loan officer survey at 2:00 pm Eastern and a G.5 release at 4:15 pm, which are 19:00 and 21:15 UTC. The first appears on one FX calendar as a 19:00 row. The second appears on none of the calendars we read today. So the two calendars are close to disjoint on this date: the vendors carry an event the issuer does not list, typed wrongly, and the issuer carries a release the vendors do not carry at all. A credit-conditions survey at 19:00 is a thinner-liquidity, lower-attention release than a mistyped fireside chat at 16:15, which is an argument for reading the issuer rather than the aggregator when the day looks quiet.
One blemish worth recording against the issuer, since we have been hard on vendors: the Board’s November 2026 page carries a footer reading “Last Update: November 03, 2025”. We did not resolve it, and we are not treating the page’s contents as stale on the strength of a footer alone.
What this does not tell you
The ISM issuer was unreachable, so the prior, the sub-indexes and the release time are all secondary readings. The sub-index values quoted here — prices at 77.9 from 71.1, new orders 55.3 from 53.7, employment 52.7 from 51.2, production 56.7 from 58.3 — are September figures from a single vendor’s table and are the previous values for tonight, not forecasts. Only two sub-indexes carry a published expectation at all, and we have no second reader for either.
We have not established which of the three construction spending consensus figures is right, and we do not expect to; a consensus is a survey of forecasters and there is no issuer to appeal to. We confirmed the 16:15 event’s nature at the Chicago Reserve Bank but did not open its president’s own public-events page, which might list something additional today. We did not check whether the fireside chat will be published later, so “nothing quotable in real time” is a claim about the next few hours and not about the week.
One figure we read and are not standing behind: a European briefing puts the odds of a December Federal Reserve hike at roughly 75 per cent. That is one publisher, with no feed named, and it sits a long way from the 93 per cent hold this desk carried from a prediction-market reader six days ago. Two single sources moving in opposite directions is not a repricing, it is an absence of data, and we are reporting the gap rather than the level. Nothing here says which way any of tonight’s prints resolves.
