At 23:12 UTC last night this desk published a week-ahead table with one line for today’s only European release: 09:00 German Ifo business climate. The hour is wrong. The Ifo Institute’s own release schedule gives 10:30 local time for today and for every month out to December 2027 — and Germany left summer time yesterday, so local time is now UTC plus one. The release is at 09:30 UTC, in roughly three hours, not 09:00. We were thirty minutes early on the first post-clock-change European event of the week, which is exactly the event we spent Friday and Saturday writing about the risk of getting wrong. Worse, the issuer writes the time as “10:30 CET/CEST” — both zone labels, one time, in the single week of the year when they differ by an hour. And the two secondary readers we carry both say 09:00, which is how a wrong minute survives: they agree with each other.
Thirty minutes, and we published it twice
The correction is small and it is ours. Our 23:12 table said 09:00 and flagged the entry as “not researched”, which was honest and did not stop us publishing a number. The number came from a Japanese week-ahead schedule giving 18:00 JST. Japan does not observe summer time, so that conversion is zone-safe and we have been recommending it since Saturday — it is a good method and it inherited a bad input. 18:00 JST is 09:00 UTC, and the publisher simply has the wrong half-hour.
The issuer settles it. The Ifo Institute publishes a forward release calendar, and every date on it — 26 October, 24 November, 17 December, and on through 2027 — carries the same local time of 10:30. 10:30 CET is 09:30 UTC. If your news filter pads fifteen minutes either side of an entry stamped 09:00, you are flat from 08:45 to 09:15 and live at 09:30 when the print lands. That is the worst of both: you paid for the pad and you took the event.
The operational point is narrower than “check your calendar”. It is that we had a method for the one class of fact most likely to be wrong this week, we applied it, and it returned a wrong answer because the method only protects against the zone and the error was in the minute.
The vendor got the hard part right and the easy part wrong, in all nine rows
One aggregator’s Ifo page gives today’s release as 09:00 with no time zone stated at all, and its nine rows of history are more interesting than its forecast. January, February and March 2026 are stamped 09:00. April through September 2026 are stamped 08:00. That one-hour step between March and April is the European clock change, correctly applied, in a column that evidently does not itself observe summer time.
So the vendor handles the genuinely difficult part — a zone boundary that moves every year and caught us out this morning — without error. Against the issuer’s 10:30 local, though, every one of those nine rows is thirty minutes early: 08:00 where 08:30 belongs during summer time, 09:00 where 09:30 belongs outside it. The zone logic is right and the stored local time is 10:00 rather than 10:30, consistently, every month.
We cannot rule out the alternative, and we are not going to pretend otherwise: the Ifo may have moved its own publication time to 10:30 recently, in which case the vendor is carrying a stale time rather than a wrong one. We have not found a dated announcement either way. What decides our behaviour is that the issuer’s forward schedule says 10:30 for fifteen consecutive future releases, and a forward schedule is the only kind of evidence that can be right about today.
A consistent error is harder to find than an erratic one. Nobody audits a field that has given the same answer every month.
The consensus is 89.9 or 90.0, and the check we published yesterday fires on one of them
Two readers give today’s consensus differently. One says 89.9, explicitly unchanged from September. The other says 90.0. September printed 89.9, up 1.1 points from August’s 88.8, so on the first reading the forecast equals the prior exactly and on the second it does not.
That matters to us because of a two-line rule this desk published at 23:14 UTC last night: refuse to fire on a field whose stored consensus equals a previously published value of the same series. We wrote it after a consumer-sentiment final whose stored forecast was its own preliminary to the decimal, which turned a routine revision into the day’s largest beat. Applied to the very next European release, twenty hours later, that rule flags today’s Ifo if your file holds 89.9 and does not flag it if your file holds 90.0. Whether a suppression rule suppresses depends on which vendor you happen to subscribe to, which is not a property a suppression rule should have.
It is also probably a false positive, and the distinction is the thing worth keeping. “Consensus equals prior” has two causes that look identical in a database: a lazy default, where nobody forecast the field and the prior was copied in; and a genuine forecast of no change, which for a sentiment survey after a 1.1-point jump is a perfectly reasonable call. Our rule cannot tell them apart. One piece of evidence does break the tie here: across the nine months of that aggregator’s own history, its stored Ifo consensus never once equals the prior. A series that has never defaulted is unlikely to have started today, which argues the 89.9 is somebody else’s honest flat forecast rather than a copied field — and therefore that our rule would misfire.
And then the deflation, which is ours too. That same history gives six beats, one in line and two misses in nine months, a mean surprise of plus 0.22 and a mean absolute surprise of 0.67 points. The two vendors disagree by 0.1. The disagreement is roughly one-seventh of the typical surprise. We have spent weeks on this channel treating 0.1-point consensus gaps as findings; on this release, that gap is noise inside noise, and the only reading with no vendor in it is the comparison against September’s 89.9.
Friday’s prices had no consensus at all, and the headline covers three different lookbacks
The opposite failure ran on Friday. The American flash purchasing-managers survey landed at 13:45 UTC and, read through its consensus fields, it was a non-event: composite output 57.6 against a 57.6 forecast, exactly in line; manufacturing 56.0 against 55.5 and 55.6, a beat of 0.4 to 0.5; services 57.9 against 57.9 and 58.0, in line and a miss of 0.1 at once. Our 23:14 article made that ambiguity its subject and flagged the 58.0 as single-sourced and load-bearing. It is now carried by three independent readers. That leg is closed.
The informative part of the release has no forecast field anywhere. Per the compiler’s own press release, selling prices across goods and services rose at the sharpest rate since August 2022; input costs at the fastest pace since November 2022, excluding September’s rise; and one reader adds that manufacturing selling prices rose at a pace not seen since July 2022, with supplier delivery times lengthening the most since July 2022 and backlogs the most since May 2022. Respondents named oil-based inputs, energy, fuel, freight, steel, aluminium and copper, often on tariffs or shortages, and service providers named wages, utilities and insurance. The compiler’s chief business economist called it the sharpest selling-price increase for over four years and said consumer price inflation is likely to run further above the Federal Reserve’s target in coming months.
“Over four years” is doing more work than it looks. From October 2026, August 2022 is 50 months and July 2022 is 51, so the selling-price claim clears four years comfortably. November 2022 is 47 months — three years and eleven months, under four years — and it carries an exclusion clause. The two price series therefore have lookbacks four months apart, one of them not actually four years and one of them conditional on skipping the preceding month, and a single “four-year high in prices” headline covers both. If your file stores one number for “PMI prices”, it is storing the longer, cleaner, goods-weighted one and dropping the caveat on the other.
We are not going to tell you what the Federal Open Market Committee does with this at 18:00 UTC on Wednesday. A hold is widely expected and we have sourced no numeric probability this run, so we are not publishing one. What we will say is that the part of Friday’s release that bears on Wednesday is the part no surprise-driven rule can see, because there is nothing to compare it to.
What this does not tell you
It does not tell you the Ifo matters. A business-climate survey forecast to come in unchanged, with a typical absolute surprise of 0.67 points, is not obviously worth a position either way; we are correcting a timestamp, not recommending a trade. Nor does it tell you the vendor is wrong rather than stale — we could not find a dated announcement of an Ifo time change, and we are acting on the issuer’s forward schedule because it is the only document that speaks about today.
The nine-month surprise record, the no-zone-stated observation and the 90.0 consensus all come from one aggregator’s page, read directly rather than scraped; the 89.9 comes from one Japanese publisher. Neither consensus is two-sourced, which is itself the point of the third section and is also a limit on it. The manufacturing selling-price comparison to July 2022, and the delivery-time and backlog superlatives, are single-sourced to one reader; the August 2022 and November 2022 comparisons are the compiler’s own.
And our own correction is a correction of a table, not of an article. No piece we published carried 09:00 for the Ifo in its body. We are publishing it because a reader who took the week-ahead at face value is thirty minutes out, and because the failure says something about a method we recommended on Saturday: a fixed-offset second source protects the zone and nothing else.
Related
- FX events calendar — the week’s central-bank windows in UTC
- Signals
- EA presets
- One Index, Three Signs — the article whose closing rule misfires on today’s Ifo
- Four PMIs in Seventy-Five Minutes — the 55.6 consensus figure, as we published it on 23 October
