Yesterday this desk published that a scheduled US release had printed at 14:30 UTC and that nine hours later the only place the number existed was the issuing bank’s own web page. We set two falsifiable tests for this morning: would FXStreet ever populate the actual, and does any aggregator carry the figure at all. Both are now settled, and the answers go in opposite directions. Sixteen hours after the print, FXStreet still shows n/a. But three of the six calendar pages we checked do carry 9.8 — so the generalisation we were tempted by is wrong, and we are saying so. What we found instead is worse for your process than a missing number: the three pages that have the print give three different consensus figures, which means whether the release was a beat or a miss depends on which vendor your system asked.
The test we set, and what came back
The release is the Federal Reserve Bank of Dallas Texas Manufacturing Outlook Survey for September, published 28 September at 14:30 UTC. General business activity 9.8, down from 11.6. Six of the seven sub-indices underneath it rose, two by double digits. That is read off the issuing bank and it has not changed.
Six calendar pages, all read this morning between roughly 06:10 and 06:25 UTC, sixteen hours and a quarter after the print:
- FXStreet — event page shows “Sep 28, 14:30 UTC+00:00”, previous 11.6, consensus n/a, actual n/a. Its “Next Release” field also still reads Sep 28, 14:30, so the page has not rolled forward either.
- Investing.com — release history table shows the 28 September row with actual and forecast both blank, previous 11.6. Its next-release row also still points at 28 September.
- MQL5 — actual 9.8, forecast 7.2, previous 11.6.
- Myfxbook — actual 9.8, consensus 1, previous 11.6.
- Trading Economics — 9.80 for September against 11.60 for August, last updated 28 September.
- FX Blue — most recent release shown is 28 July 2025, actual 0.9, previous −12.7, with the next event listed as 25 August 2025.
So the honest scoreboard is three carrying it, two showing the row with no value, and one serving a calendar fourteen months out of date. We said yesterday that if none of four aggregators had it, the claim would generalise from “two calendars” to “the aggregator layer”. It does not generalise. We are not going to keep a conclusion that our own test refused.
What did generalise: the time field and the value field are separate products
FXStreet had this release’s time correct to the minute before it happened. It does not have the value sixteen hours afterwards. Yesterday we put that lag at nine hours and noted the field might fill overnight. It did not. The claim therefore strengthens rather than narrowing: this is not a publication delay measured in hours, it is a field that has not been written.
Investing.com is the more instructive of the two, because it is not a small vendor and its August row is fully populated — 11.6, previous 1.3. So the pipeline that fills this series exists and worked last month. Whatever failed, failed for one month on a page that has the rest of the history.
The process consequence is narrow and concrete. If you read release times from a calendar page and assume the actual will appear on the same page, you have a dependency you have never tested, because the two fields do not come from the same place and do not fail together. A system that blocks entries around 14:30 on a Dallas Fed day and then reads the outcome from the page it got the time from will block correctly and then read nothing. Verify schedule coverage and print coverage separately, at different sources, and do not let the reliability of the first one vouch for the second.
FX Blue is the third failure mode and the one most likely to hurt you quietly. A page that is missing this month’s number tells you it is missing. A page serving July 2025 as its latest release will hand your parser a complete, well-formed, internally consistent row with an actual, a previous and a date — and the date is the only field that gives it away. If your ingestion does not assert that the release date is recent, a fourteen-month-old print will enter your system looking exactly like a fresh one.
Three consensus figures, and the beat is between 2.6 and 10.8 wide
Here is the part we did not expect. The pages that carry the actual do not agree on what was expected. MQL5 gives a forecast of 7.2. Myfxbook gives a consensus of 1. Both show the same actual and the same previous.
Against 7.2, a print of 9.8 beats by 2.6. Against 1, the same print beats by 8.8. And against the previous of 11.6, the same print is a fall of 1.8. Three fields, three signs of answer: better than expected, much better than expected, and worse than last month. All three are true statements about one release.
There is a third figure we will not publish as a source. This desk has twice now seen a consensus of −1.0 for this series in a search-result snippet we have never been able to open, and a snippet is not a read. Against −1.0 the beat would be 10.8. We mention it only so you know the range of what is floating around and that we are keeping it out of our own numbers for the second consecutive run.
Note also that we were straightforward yesterday about being unable to say whether 9.8 was a beat or a miss, because we had no readable consensus. We now have two. That closes the question and immediately reopens it in a more awkward form: the answer is yes, it beat, by an amount that is either trivially small or very large depending on which vendor you asked. If your sizing is a function of surprise magnitude, your position size on this release is a function of your data supplier, not of the economy.
Trading Economics is worth one line of credit here. Its own text says the index “edged down” in September after August’s jump, and that broader business conditions improved while company outlook deteriorated and uncertainty rose. That is an independent second reader arriving at the same structure we published yesterday off the primary document: the headline fell and most of what sits under it did not.
The other window we told you to watch has re-armed, larger
Yesterday we also settled an option-expiry test with a null result. We had published a 4.5-billion-euro EUR/USD strike at 1.1400 for the 14:00 UTC cut and called it the standout. Price never went there: the 28 September high was 1.1395, five pips short, so there was nothing to observe.
This morning the same strike is back for the 29 September cut at 4.9 billion euros, about nine per cent larger, and the publisher again calls it sizeable and puts it roughly forty pips from spot. Tokyo traded EUR/USD in 1.1355 to 1.1374 and left it at 1.1357 at 15:00 JST, which is forty-three pips below the strike — further away this morning than yesterday’s session high got. The hundred-hour moving average is around 1.1384, twenty-seven pips above spot and between the two.
For scheduling purposes that is the whole question: does a 14:00 UTC cut belong in your event list when the strike is forty-three pips away and yesterday’s identical setup produced nothing? Our answer is that it belongs in the list as a conditional, not as an event. A strike matters when price is at it. A strike forty-three pips away on a pair whose one-week implied is 5.91 per cent is a level price may not reach in the session at all, and blocking your London and early New York hours for a cut that was irrelevant twenty-four hours ago at a smaller distance is paying real opportunity cost for a notional that only looks decisive on a page. Arm the filter if spot comes inside, say, fifteen pips of 1.1400 before 13:30. Otherwise leave it out.
The London context around that, from Gaitame’s European outlook: EUR/USD’s year-to-date low is 1.1325 set on 24 June, the twenty-five-day high is 1.1411, and 1.1300 is the round-number support below. Spot is closer to the June low than to the strike. Three ECB speakers are scheduled through the European session — Kazimír, Nagel and Escrivá, the last described as dovish and as participating without a vote this month — followed by euro-area September confidence indices, and then US consumer confidence and JOLTS at 14:00 UTC, the same minute as the cut.
What this does not tell you
It does not tell you FXStreet and Investing.com will never carry this print. Both fields could fill this afternoon. Our claim is about their state sixteen hours after a release, which is the window in which a system trading or filtering on that release would have needed them, and that claim is now twice measured rather than once.
It does not tell you which consensus is right. We have no basis for preferring 7.2 to 1, and we are not going to pick one to make the paragraph tidier. We are also aware that a single-digit consensus field rendered as “1” could be a rounding or truncation of something else, and we cannot rule that out from the page. If you need a surprise magnitude for this series, you need to source the survey’s consensus deliberately and know who compiled it.
It does not establish that FX Blue is stale for every series — we checked one page, and one stale page is evidence about that page. It is, however, enough to justify a date assertion on anything you ingest from anywhere.
And it does not tell you the 1.1400 strike will be ignored again today. A null result yesterday is one observation, not a pattern, and a 4.9-billion notional is a real thing that will matter if price arrives. Our claim is about whether it earns a scheduled block in advance, not about whether it can move price when tested.
One thing we owe you plainly: two of the six pages here were reached through search results rather than from a source we already trusted, and the whole article is a claim about vendor reliability built from vendor pages. We have not audited these companies. We read six pages on one release on one morning and we are telling you exactly what each one showed.
Related
- FX events calendar — release times in UTC, kept separate from release values.
- Signals — what we do with a surprise whose magnitude depends on the supplier.
- EA presets — news-window filters, date assertions on ingested data, and conditional option-cut blocks.
