The preview headline for the only timed central-bank decision in today’s window reads “due Thursday, 29th October 2026 at 13:15GMT/08:15EDT.” Those two strings are not the same instant. On 29 October 2026, 13:15 GMT is 09:15 EDT, and 08:15 EDT is 12:15 GMT. One field, two values, an hour apart — and the arithmetic tells you exactly which half is stale and why. Europe left summer time on 25 October. The United States leaves it on 1 November. For these seven days Frankfurt and New York are five hours apart instead of the usual six, and every conversion done with the habitual offset comes out an hour early.
The decision time is settled. The conversion is not.
Take the easy part first. Four independent readers put the European Central Bank’s announcement at 13:15 UTC today: the preview itself in its GMT half, a euro-rates reference page giving “29/10/2026, 14:15 CET,” a market brief listing “Thu Oct 29, 13:15 UTC,” and a Tokyo desk note scheduling it at 22:15 Japan time. 14:15 CET, 13:15 UTC and 22:15 JST are one instant. This desk has spent six weeks publishing calendar rows with a field missing; this one is over-determined. If your system has 13:15 UTC in it, it is right.
The failure is in the second half of the same string. The Governing Council announces at 14:15 Frankfurt local time, and that local number has not changed in years. What changed four days ago is the offset underneath it. Until 25 October, 14:15 in Frankfurt was CEST, UTC plus two, which is 12:15 UTC and 08:15 EDT. Since 25 October it is CET, UTC plus one, which is 13:15 UTC and 09:15 EDT. The headline updated the GMT figure and left the EDT figure on the old offset. Nothing in it looks broken. Both halves are plausible times for an ECB decision, and the only way to catch it is to subtract.
This is the one week a year the error is invisible
Frankfurt and New York are six hours apart for the overwhelming majority of the year, because both sides change clocks and the gap survives the change. There are two short windows when they do not: the spring gap, and this one. From 25 October to 1 November 2026 the gap is five hours. A converter, a cached constant or a person applying the six-hour rule to a European event in this window lands an hour early, every time, in the same direction.
That direction matters more than the size. An hour early on a decision time means your news window opens at 12:15 and closes at 12:15 plus your pad — and if that pad is thirty minutes, it has shut at 12:45, half an hour before the central bank speaks. The window does not error. It does not look empty. It covers a stretch of tape in which nothing happens, and then you are flat into the event with your filter already satisfied. A late window is survivable; an early one is a filter that reports success.
And the error has three more days to run. On 1 November the United States goes to standard time, the gap returns to six hours, and every stale offset in the system becomes correct again — which is the worst possible way for a bug to resolve, because it closes without ever being found. Anything you convert from a European local time before Sunday is suspect. Anything you convert after it is not.
Twelve rows, one minute, and five things with “PCE” in the name
The 12:30 UTC block is not one release. It is two statistical releases from the same agency plus the weekly claims series, and the Bureau of Economic Analysis’s own schedule page lists both of its items — “GDP (Advance Estimate), 3rd Quarter 2026” and “Personal Income and Outlays, September 2026” — at “8:30 AM,” with no time zone stated on either. That is the issuer. It gives a minute and declines to say which minute.
One vendor calendar we read lays that block out as twelve rows at 08:30 AM Eastern, each carrying a reference period. Sorted by what they actually measure, they are: GDP quarter-on-quarter for Q3; the PCE price index year-on-year and quarter-on-quarter for Q3, priors 5.0 and 3.8; personal spending, personal income, the PCE deflator month-on-month and the core PCE deflator month-on-month, all for September, consensus 0.6, 0.3, 0.5 and 0.3 against priors 0.9, 0.2, 0.3 and 0.2; the PCE and core PCE deflators year-on-year for September at 3.6 and 3.1 against 3.4 and 3.0; and initial and continuing claims. Five distinct price series, two reference periods, one timestamp.
A second vendor carries the same block with no reference-period field at all. On that page the rows read “Core PCE Price Index” at 0.3 against a 0.2 prior, “Core PCE Prices” at 2.40 against 3.30, and “Core PCE Price Index (annual)” at 3.1 against 3.0 — three rows, three numbers, names that differ by a word, nothing saying one is a September month, one a September year and one a Q3 annualised rate. Alongside them sit “PCE Price Index” at 3.6, “PCE Price Index (monthly)” at 0.5 and “PCE Prices” at 3.4 against a 5.0 prior. If your system keys on a name, it has six candidates for two fields and the only thing separating them is the prior. That page also describes itself as machine-written with editorial review, which is not a reason to discard it — its numbers reconcile with the other calendar wherever both carry the same object — but it is a reason to read the prior before trusting the label.
The practical consequence is narrow and expensive. The Federal Reserve’s target variable appears in this minute three times over three different horizons, and two of the three belong to a quarter that ended four weeks ago. A system that scores a surprise against whichever one it matched first will record a hit or a miss on a number nobody traded.
The consensus for the largest release is four different numbers
Advance Q3 GDP is the biggest thing in the block, and we cannot tell you what it is expected to print. Three readers give three figures against the same 2.2 per cent prior: 3.0 per cent from the Tokyo desk note, 3.2 from the machine-written calendar, 3.43 from the vendor that carries reference periods. A fourth, which this desk published on at 06:14 this morning, has the consensus field empty along with a reference period naming the wrong quarter. That is a spread of 0.43 points, and it is wider than the surprise most advance GDP prints deliver.
So we are not quoting one. The band is 3.0 to 3.43 and the prior is agreed. Claims are a smaller version of the same thing: 198 thousand against 200 thousand, same 194 thousand prior, a two-thousand disagreement nobody will notice. The honest reading of the GDP row is that there is no consensus in your system today, only a vendor’s choice of one, and that any surprise you compute from it inherits a 0.43-point error bar before the number is even out.
What the window arithmetic says about 12:30 to 14:00
Six timestamps today, in UTC: 10:00 euro-area economic sentiment and consumer confidence, priors 97.9 and −17.9; 12:30 the twelve-row block; 13:15 the ECB; 13:45 the press conference; 14:00 the New York option cut; 17:00 a seven-year note auction. Lay a symmetric thirty-minute filter over them and you get three blocked windows totalling four and a half hours, with 12:00 to 14:30 fused into one continuous two-and-a-half-hour block. Double the pad to sixty minutes and you get three blocked windows totalling seven and a half hours — exactly three hours more, and not one window fewer.
That is the whole argument for this slot. Doubling your pad today buys you nothing structural, because the events are already clustered tightly enough that the thirty-minute windows have merged. You are paying three hours of flat time for additional cover inside a block you were already out of. The place that pad actually changes something is the edge: at thirty minutes you are clear from 14:30, at sixty you wait until 15:00, and the longest genuinely clear stretch of the European and American day either way runs from 18:00 to the Tokyo CPI print at 23:30 UTC tonight.
For context on where the tape sits going in: the ECB’s own reference fix yesterday at 15:00 UTC put euro–dollar at 1.1133, dollar–yen at 157.96 and sterling–dollar at 1.3288. Tokyo ran 47 pips this morning, 157.60 to 158.07, and London has held the late 157s with euro–dollar stalling around 1.1170. Yesterday’s high was 158.35 and a 200-day moving average sits at 158.54. None of that is a view; it is the distance to the levels your stops are near.
What this does not tell you
It does not tell you the EDT figure in that headline cost anybody anything. It is one string in one preview, the GMT half of the same string is right, and a reader who takes the GMT half is unharmed. We are publishing it because the mechanism behind it is general and this week is when it fires, not because we watched it do damage.
We have not established the time-zone label on the machine-written calendar. Its rows sit an hour before the issuer’s on the one item we can anchor, and its seven-year auction row sits an hour before the time we carry from two readers, so the whole page looks uniformly shifted — but we could not confirm what zone the page claims for itself, and a uniform offset with an unread label is not a vendor error we are willing to name. The three-row pattern is what we saw; the cause is not established.
The euro-area inflation, energy and PMI figures below are single-sourced to one preview. The 0.43-point GDP band is real but it is a disagreement between readers, not evidence that any of them is wrong. And the option expiry page at today’s dated address still carried yesterday’s headline, yesterday’s byline and yesterday’s board when we re-read it at 11:52 UTC — five and a half hours after this desk published that it did at 06:19 — so the structural point in that piece stands, but we have no board for today and will not infer one.
