Three Readers Put Euro Inflation at 10:00. The ECB's Own Calendar Says 15:00.

The hour, four ways, and the issuer gives none

Start with the only thing that is genuinely settled. Eurostat is the issuer, and its 2 October release of the September flash carries the line “Next release: 4 November 2026” in its header. That closes the date at the issuer. It does not name the product and it does not give an hour, so the issuer confirms the day and nothing else.

The hour comes from secondary readers, and three of them agree:

• A vendor calendar row dated 2026-11-04 printing 10:00 AM under a column headed GMT.
• An event page printing 11:00 CET, with a zone table giving London 10:00, New York 05:00 and Tokyo 19:00.
• A Tokyo publisher’s London-session outlook, written at 04:45 UTC, putting the release at 19:00 Japan time.

In November, 11:00 CET and 19:00 JST are both 10:00 UTC. Three readers, three notations, one instant. That is as good as a secondary time gets, and it is the figure to put in your schedule.

Now the fourth reader. The ECB’s weekly calendar lists three entries for today, and two of them carry a bare clock time with no zone: an ECB Wage Tracker at 10:00 and a euro-area seasonally adjusted HICP flash estimate at 15:00. The third entry, a Lane engagement, carries 14:20 CET. So on one page, from one institution, two rows have no time zone and one does. A machine reading that page cannot determine from the page whether 15:00 means 15:00 CET or 15:00 UTC, and the two differ by an hour in which a great deal can happen.

The row is not the same row, and the giveaway is two words

Here is the reconciliation we think is right, offered as a reading and not as a finding. Eurostat’s 10:00 UTC publication is the flash estimate press release — the number that moves euro–dollar. The ECB’s 15:00 entry is named “euro area seasonally adjusted HICP flash estimate” and is tagged to its HICP dataset. The two extra words are the whole story: a seasonally adjusted series that the ECB computes and publishes into its own data portal after the issuer’s release is a different object from the issuer’s release, and there is no reason it should land in the same minute.

If that is right, then nobody is wrong and the hazard is entirely in the naming. A news-window filter seeded from a central-bank calendar would open its euro-inflation window four hours after the number printed, and would sit through the actual release with no window at all. A filter seeded from the vendor calendar gets the right minute and will later see a second row it cannot distinguish from the first.

We cannot settle it, and we are going to be specific about why. Our fetch of the ECB’s weekly calendar returned roughly half the page verbatim and the calendar rows themselves came back through a summariser rather than as page text, and a re-fetch split the page the same way. So every ECB figure in this article — the 15:00, the 10:00 Wage Tracker, the 14:20 CET — is a secondary rendering of an issuer page we could not read in full, and should be treated as one reader rather than as the institution. We have published it because the structural point survives even if a digit is wrong: two near-identically named rows, four hours apart, is a metadata problem whichever row carries which time.

The consensus column trap, second instance in three days

On 2 November this desk published that one vendor prints Consensus and its own house forecast as two separately labelled columns of the same calendar row, and that a desk should check which column its readers sampled before calling a tenth of a point a dispersion. We flagged it on a manufacturing survey. Here it is again, on euro-area inflation, three days later.

The same vendor’s October HICP row reads: Previous 3.8 per cent, Consensus 4 per cent, house forecast 3.9 per cent. A Tokyo publisher carries 4.0 per cent as the expected headline. A second calendar’s listing text carries 3.9 per cent. So the apparent ten-basis-point consensus dispute across readers is, on the most parsimonious reading, two readers sampling two different columns of one vendor’s single row — and the honest statement is that we have one consensus figure, 4.0 per cent, carried by two readers, and one vendor model forecast, 3.9 per cent, carried by one.

This is the second instance of a failure mode we filed as an anecdote on Monday, in a different asset class and a different release, so it is now a pattern rather than an incident. We are still not upgrading it into a general claim about calendar vendors. Two instances three days apart justifies one specific operational rule and no more: when two of your readers are a tenth apart on a consensus, check whether they are quoting the same column before you treat the gap as information. The containment check costs one page view.

The second calendar deserves a separate warning. Its page for the October flash describes the September release in its body, is marked as having already passed, carries “Time not confirmed” in its header, and lists its prior as 2.9 per cent from July 2026 — two reference months stale — while an August entry elsewhere on it reads 3.3 per cent against the issuer’s 3.2. Its date field for today is right. Almost nothing else on it is.

Lane is on three calendars today, and it is a prize ceremony

Two calendars we read list Lane as a speaker today, one of them naming his speech as a focus for the London session alongside the inflation print. The ECB’s own calendar says what the engagement is: a laudation at 14:20 CET in Berlin, on the conferral of the Bernhard Harms Prize to Pierre-Olivier Gourinchas, at the fifth Kiel-CEPR Geoeconomics conference.

A laudation at a prize ceremony is not a monetary-policy speech. A reader who stores “ECB chief economist speaks” and widens spreads or stands down for thirty minutes around it is spending risk budget on an academic tribute. Meanwhile the same official appears on a panel tomorrow at 16:40 CET at a central-bank watchers conference in Zurich — the format far likelier to carry rate-path content — and tomorrow is the Bank of England’s decision day, when the attention is elsewhere.

In September this desk published a piece on the failure mode of a name-keyed news filter, after one publisher put the wrong person in the Federal Reserve chair. This is the same fault from the other direction: the name is right, the institution is right, the time is right, and the event type is wrong. An event-type field catches it. A name-keyed filter cannot, and most of them do not have one.

The same calendar lists a fuller slate tomorrow — Cipollone at 08:30 CET, Schnabel opening the ECB’s own money markets conference at 09:05, Lagarde twice — which is worth noting only because a desk budgeting for “ECB speakers” by headcount will under-budget tomorrow and over-budget today.

What the number will and will not change

On the issuer’s own September component table, the headline was 3.8 per cent and core, on the narrow definition excluding energy, food, alcohol and tobacco, was 2.5 per cent. The 1.3-point gap between them is not mysterious: energy was running at 18.8 per cent year-on-year, up from 14.3 in August and 3.9 per cent on the month alone, and unprocessed food at 4.0 per cent. Services were 3.2 per cent and actually fell 0.7 per cent on the month.

The expectation carried by two readers is 4.0 per cent headline with core unchanged at 2.5 per cent. If that is what prints, the headline-to-core gap widens to 1.5 points and sits almost entirely in the components the central bank says it looks through. One Tokyo publisher reports December rate-rise speculation and marks the euro’s range for the session between 1.1337, the 1 October high, and 1.1171, the daily Ichimoku conversion line, with spot near 1.1249 at 05:00 UTC — 88 pips below the top of that range.

So the asymmetry for a system is this. A headline beat driven by energy is the low-information outcome, and the 10:00 UTC window is likely to be a spike that mean-reverts. A core surprise in either direction is the high-information outcome, and it will be in the same release at the same second, in a field most calendars do not display and almost no alert quotes. If your news filter widens on the headline and your risk model keys off the policy path, they are reading two different numbers out of one print.

The October components are not published until the number is, so every component figure above is September’s. We are not forecasting the print.

What this does not tell you

It does not settle whether the central bank’s 15:00 row is a different publication or a different hour. We have given our reading, labelled it a reading, and told you the fetch that produced it was partial.

It does not give you the release time from the issuer. Eurostat’s own release calendar page loads and returns no calendar content — the table is loaded dynamically and did not come back — so the 10:00 UTC in your schedule today rests on three secondary readers agreeing, which is good but is not an issuer confirmation. If one of those three is the origin of the other two, the agreement is worth less than it looks, and we cannot rule that out.

It does not establish the consensus. We are publishing 4.0 per cent as the figure two readers carry and 3.9 per cent as one vendor’s own model, and our column-sampling explanation for the gap is an inference. We did not reach a consensus provider directly.

One of the event pages giving us the 11:00 CET hour should be treated as a source for the hour and nothing else: its weekday field says Tuesday for a date that is a Wednesday, its September figures contradict the issuer on every line, and its own footer says the figures on the page are placeholders. We used it for one field because that field independently matched two other readers. We would not have used it for anything else, and neither should you.

And it does not tell you what any of this does to price. The option cut is at 15:00 UTC today and no publisher we can open has today’s expiry board, which our Flow desk has written up separately this morning.

Related

• FX event calendar — the 10:00 UTC euro-area flash, and tomorrow’s Bank of England decision.
• Signals — why a core surprise and a headline surprise should not trigger the same filter.
• EA presets — news-window configuration when an event type, not an event time, is the thing that is wrong.