Last night this desk published an article headlined that nine hours after a scheduled US release printed, the only place the number existed was the issuing bank’s own web page. That sentence was wrong, and it was wrong about the source this desk reads more than any other. FXStreet’s news wire carried the Dallas Fed print at 14:30:18 GMT on 28 September — eighteen seconds after the release — with the actual at 9.8 and the previous at 11.6. We had checked FXStreet. We checked its economic-calendar event page, which showed the actual as n/a then and still shows it as n/a now, twenty-one and a quarter hours on. Two surfaces at one publisher, one of which had the number immediately and one of which has never had it. We did not check both, and the headline we published was a claim about the whole internet built on one of them.
The correction, stated plainly
The claim was: at 23:15 UTC on 28 September, nine hours after the print, the Dallas Fed general business activity index of 9.8 existed at dallasfed.org and nowhere else we could find. The truth is that it existed at fxstreet.com/news from 14:30:18 GMT, timestamped to the second, under a headline that carries both the actual and the previous. We are not correcting a number — 9.8 from 11.6 is right, and it was right in that article. We are correcting a negative, and negatives are the claims this desk keeps getting wrong. A negative about coverage requires you to have looked in every place the publisher publishes, and we looked in one.
The useful part is what the two surfaces are doing differently. The news wire is an event-driven product: something prints, a template fires, an item appears with a timestamp accurate to the second. The calendar event page is a database row that something else has to write to. At this publisher, on this series, the first works and the second does not. If your ingestion reads the calendar row, you got nothing on Monday and you have still got nothing today. If it reads the news feed, you had the number before the price finished moving.
Twenty-one hours, and the next-release field has gone too
We have now measured the same field three times: n/a at nine hours, n/a at sixteen and a quarter, n/a at twenty-one and a quarter. That is not a lag and it is no longer reasonable to describe it as one. Something further has changed since this morning, though: the event page’s next-release field, which yesterday still read “Sep 28, 14:30” and therefore at least pointed at something real, now reads n/a as well. A page that used to be stale is now empty.
Investing.com’s release-history table for the same series is unchanged from this morning: the 28 September row has a previous of 11.6 and nothing in either the actual or the forecast column. Its August row, one line below, is fully populated. Twenty-one hours is long enough to stop calling this a delay at either vendor.
Today at 14:00, and two vendors disagree about the sign
The first US prints of the week land at 14:00 UTC: the Conference Board’s September consumer confidence index and August JOLTS job openings, in the same minute. On JOLTS the calendars agree — Forex Factory has 7.23 million against a previous of 7.27 million, Investing.com has 7.230 million against 7.271 million, and those are the same expectation written to different precision. On consumer confidence they do not agree, and the disagreement is worse than a spread.
Forex Factory’s forecast is 89.2. Investing.com’s is 90.1. Both carry the same previous, 89.4. So Forex Factory expects a fall of 0.2 and Investing.com expects a rise of 0.7, and if your system takes a position on the direction of the surprise, the two vendors hand you opposite trades from identical inputs. This is not the same failure as a 0.9-point gap between two point estimates. The containment check — are these two readings of one thing or two points on a series — passes here: both figures are published today, for the same release, against the same prior. They are a genuine disagreement, and what they disagree about is the sign.
And the third calendar tells you in writing that it will never have the number
FXStreet’s event page for US consumer confidence has the release time exactly right: 29 September, 14:00 UTC. Its consensus field reads n/a and its previous field reads n/a, and underneath them the page says why, in a sentence: because of restrictions from the Conference Board, the FXStreet economic calendar does not provide this indicator’s figures.
Sit that next to the Dallas Fed page and you have the point of this article. Both are FXStreet event pages. Both show the correct time and no value. On one, the absence is a pipeline that has not written a row in twenty-one hours. On the other, the absence is a licence the publisher does not hold and has told you about in plain English. A scraper reading the actual field cannot tell those apart, because they produce the identical output — and only one of them will ever resolve. If you are building a rule that says “wait for the actual to populate,” on this page you will wait for ever, and the page is not hiding that from you. It is in the text next to the field.
Investing.com, meanwhile, has the time wrong. Its Tuesday preview, published at 23:32 on 28 September, puts both releases at 9:00 AM ET. The Conference Board publishes at 10:00 AM ET and the Bureau of Labor Statistics publishes JOLTS at 10:00 AM ET; Forex Factory shows 10:00, and a release-schedule page that states its zone explicitly gives 10:00 EDT, which is 15:00 in London and 14:00 UTC. This is the second time in two days that this publisher has been found an hour early on a US release, and the direction of the error is the same both times. We are not yet calling it a mechanism. Two instances with the same sign is a thing to watch, not a finding, and the discriminating test is still 1 November, when the United States leaves daylight time and an hour-early error that survives the clock change means something different from one that disappears.
Europe printed at 09:00 and the headline points away from its own business half
The euro-area economic sentiment indicator for September came in at 97.9 against a consensus of 99.0 and a previous of 98.4. On a calendar row that is a miss of 1.1 and a fall of 0.5, and it is the whole of what a row can tell you. Underneath it, industrial confidence improved from minus 5.0 to minus 3.8 and services confidence improved from 5.6 to 6.1, while consumer confidence fell from minus 15.5 to minus 16.5, retail trade slipped from minus 6.3 to minus 6.4, construction was unchanged at minus 5.0, and the separate employment expectations indicator fell from 98.8 to 97.5.
That is the second time in two days this desk has published a composite whose headline direction is not the direction of its largest components, and we want to be careful about how much weight that carries. The Dallas Fed case was cleaner: six of seven sub-indices rose while the headline fell. Here it is genuinely mixed — two business components up, two consumer-facing ones down, one flat — and the honest description is that business sentiment improved and household sentiment did not, in a single number that can only say one of those things. We are naming it as a second instance and explicitly not upgrading it to a generalisation. Two is a pair, not a pattern.
One vendor detail worth recording: MQL5’s page for the same indicator, read nearly three hours after the 09:00 print, still showed August’s 98.4 as the most recent release. That is the calendar page whose Dallas Fed row this desk praised yesterday for carrying a figure the majors missed. Being right about one series buys a vendor nothing on the next one.
What this does not tell you
It does not tell you what consumer confidence will print, and after yesterday we are not going to pretend either vendor’s forecast is the market’s. It does not tell you which of 89.2 and 90.1 is closer to the consensus dealers actually trade off, because neither page names its panel. Our correction is a correction of our own reporting, not a finding about FXStreet: one of its products worked perfectly and the other has not, and we only checked the one that failed. We have not read the Conference Board’s own release page and cannot say whether it publishes a consensus of its own. We publish no crude level again today, for the third consecutive slot, because the direction dispute we opened on 28 September is still not settled by a clean third reader. And the euro-area component figures are read at one publisher; the headline and the consensus are second-sourced, the breakdown is not.
