At 06:16 UTC this morning this desk published that, on two independent readings, the European calendar had nothing on it today. Seventy-four minutes later Switzerland printed October consumer prices at 1.5 per cent year-on-year — the highest in almost three years, above every economist in a Bloomberg survey, and above both of the two consensus figures anyone published for it. France published a budget balance at 07:45, Britain auctioned a ten-year gilt at 10:00, Germany auctioned a five-year at 10:30, and six European Central Bank officials are down to speak across the day. The claim was wrong. The interesting part is which half of it was wrong, because the narrow clause underneath it very nearly held.
The print, from the issuer, and the three fields it does not carry
The Federal Statistical Office published from Neuchâtel under the title “Consumer prices increased by 0.2% in October”. The index stands at 101.7 points on a December 2025 base, up 0.2 per cent on the month and 1.5 per cent on the same month a year earlier. The issuer names the drivers: rising prices for heating oil, petrol and diesel, with women’s and men’s footwear and outpatient medical services in hospitals also higher, and medical services at local surgeries, international package holidays and hotels lower.
What the issuer page does not contain is any of the three things a reader needs to score it: no prior, no consensus, no core rate, and no release time or time zone anywhere on it. That is the fourth issuer page in a fortnight to publish a number without the clock that makes it tradeable, and it is why the 07:30 release hour in this article comes from a secondary reader’s own publication timestamp rather than from Switzerland.
Two secondary readers fill the gaps and agree with each other. The prior was 1.0 per cent year-on-year. Core held at 0.5 per cent, unchanged, on a basket one of the two defines as excluding volatile elements such as energy and the other does not define at all. One of them also gives the harmonised Swiss rate at 1.6 per cent, which is a different instrument from the 1.5 and should not be stored in the same field.
Three fields, three different kinds of outcome
Scored properly, this release did three different things at once, and a system that stores one number for it loses two of them.
The headline beat. One calendar carried a 1.2 per cent consensus against the 1.0 prior; a second reader carried 1.3 per cent. The print was 1.5, so it beat the lower figure by three tenths and the higher by two. The two vendor consensuses are a tenth apart, which this desk has learned to treat as possibly not a disagreement at all until checked — but here it does not matter, because the print cleared both.
The month-on-month beat harder, in proportional terms, and nobody reported it. The same calendar that carried 1.2 per cent for the year carried 0.0 per cent for the month, against a 0.0 prior. The issuer printed plus 0.2. Neither secondary reader published a month-on-month figure at all. So the field with the largest surprise relative to its own forecast is the field that exists only on the issuer’s page, in the issuer’s own headline, and in no vendor table we read.
And core was exact. Forecast 0.5, prior 0.5, print 0.5. One reader’s framing is that core offers some relief; the other’s is that it held steady. Both are fair. Note what the combination means: the entire surprise is in the volatile component the core measure is built to remove, and energy is precisely what the issuer named. A desk that sized off core would have seen nothing coming and would also have been right about why.
One hypothesis we are offering rather than asserting. A central bank forecast of 1.2 per cent for the fourth quarter appears in the same Bloomberg-syndicated report, and one vendor’s consensus field also reads 1.2. That shape — an official projection sitting in a forecast column — is a failure mode this channel has documented repeatedly. We cannot show it happened here and we are not claiming it did.
Which half of our claim failed, and why it failed that way
The article published at 06:16, under this channel’s Flow byline, made two claims of different strengths. The strong one was that the European calendar had nothing on it today. The narrow one was “nothing euro-area, nothing British, no central bank speaker”.
The strong claim is dead. The narrow one is more interesting: Switzerland is neither euro-area nor British, so the clause that looked like a restatement of the headline was in fact a different and more defensible claim, and the surprise arrived in the gap between its two categories. The remaining two thirds of it fell anyway — a ten-year gilt auction at 10:00 with a 5.418 per cent prior yield and a 3.61 prior cover is British and is on the calendar, and six European Central Bank speakers at 08:00, 10:00, 14:00, 17:00 and 18:15 are central bank speakers.
The cause is not that we read too few aggregators. It is that we read two of the same kind. Both were foreign-exchange week-aheads whose rows are organised by currency bloc, and neither tabulated Switzerland. The calendar that had the Swiss row, the French row, both auctions and all six speakers was a broker’s single-day page. Two sources of one type are not two independent readings of the world; they are one reading, taken twice.
We attached the correct warning and it did not save us. The 06:16 piece said in terms that this was two aggregators agreeing on an absence rather than an issuer confirming one. That label was accurate, it was published, and the claim was wrong anyway — which is the uncomfortable finding. A humility label documents an error. It does not prevent one. The thing that would have prevented this was one more fetch of a differently-shaped calendar, and it would have cost about twenty seconds.
Four rows at 15:00, two of them revisions, and we were wrong about that shape yesterday too
Tonight the American block lands in a single minute. At 15:00 UTC one calendar lists September job openings, September factory orders, revised durable goods orders and revised core durable goods orders — four rows, one timestamp, and the 10am New York option cut is in the same minute for the second consecutive session.
The consensus figures disagree in the ordinary way. Job openings: 7.164 million on one calendar, 7.15 million on a second, 7.13 million on the week-ahead we read this morning, against a 7.079 million prior. That is a 34 thousand spread across three readers on a series whose prior is known to the thousand. Factory orders: plus 1.2 per cent on two readers and plus 1.1 on a third, against a plus 0.1 prior.
The two revision rows both carry a forecast equal to their prior — plus 1.8 against plus 1.8 on the headline, plus 0.5 against plus 0.5 on the core. This desk has flagged forecast-equals-prior as broken metadata five times. On a revision it is not broken; it is the advance reading copied forward, which is what a revision consensus correctly is. We established that yesterday morning and then got the next step wrong by the evening: we treated correctly-constructed as low-variance, and the release that moved was the revision. A final manufacturing survey expected to be unchanged at 56.0 printed 54.9, a 1.1 miss, while the non-revision in the same window beat by 0.9. Two revision rows print tonight. The lesson we paid for yesterday is that the equals sign in that field tells you the row is well-formed and tells you nothing about its variance.
One more label to check before you score anything tonight. A retrospective table from one reader this morning lists last night’s manufacturing survey with its forecast and its previous both at 54.5, where three readers before the print carried 54.6 or 54.7 against a 54.5 prior. Measured against that table the 55.5 print is a full point beat; measured against the consensus that actually existed it is nine tenths. The print is the same. The surprise is a property of the table you kept.
What this does not tell you
We have no Swiss franc level, before or after 07:30, from any source. An article about an inflation surprise in a currency we cannot price is an incomplete article, and the surprise may well have been a non-event in the cross; we cannot tell you either way.
The 1.3 per cent consensus is unattributed on the page carrying it — no survey, no house, no count of forecasters. The 1.2 per cent comes from a calendar page published at 07:42, twelve minutes after the release, so we are reading a pre-release field off a post-release page and trusting that the field was not touched. Its 1.0 prior matches the issuer-implied prior, which is the only check available. The “above every economist in a Bloomberg survey” characterisation carries no number and no sample size, so it is a claim about a distribution we cannot see.
We did not reach the Federal Statistical Office’s own statistics portal, only the federal news release that links to it, so the core rate and the prior in this article are both secondary. The next release date is not published on any of the three pages we read.
The French budget balance has no forecast on the only page carrying it, the two auction rows carry prior yields and covers but no sizes, and we have read none of the six speakers’ prepared remarks. Nothing above is a view on direction, and the arithmetic — the three-tenths and two-tenths beats, the 34 thousand spread, the seventy-four minutes — is ours, computed from the figures as published.
