The New York Fed’s Empire State manufacturing survey prints at 12:30 UTC. Three readers give three consensus figures — 14.75, 15.0 and 14.0 — against a prior of 20.6 that the New York Fed itself calls its highest reading in more than four years. So the number you are trading is not a number, it is a band about one point wide, and every member of that band says the same thing: roughly six points are expected to come off a four-year high. The part you can actually settle is the prior, and we settled it at the issuing institution. The part you cannot settle is the minute, because the New York Fed does not promise one: its own release note says the survey goes out “at or shortly after 8:30 a.m.”
The band, and why we are publishing it instead of a number
Three independent readings of the September consensus, as of this morning:
- 14.75 — Trading Economics’ consensus field, and the figure FXStreet carried yesterday. Two readers, one number.
- 15.0 — Fisco’s New York preview, published 19:34 JST this evening.
- 14.0 — Trading Economics’ own model forecast, which sits on the same page as its 14.75 consensus and is a different object: a house projection, not a survey of houses.
The honest statement is that the market expects something between 14 and 15. This desk has spent three weeks being wrong about single consensus figures — the Canadian CPI dispute closed on Monday with both vendors half right and neither wrong — and the lesson we took from it was to publish the band and name the sources rather than pick a winner. A one-point spread on a survey index that moved five points last month is noise. It is worth exactly one sentence, and this was it.
What is not noise is the direction. Every reading in the band implies a deceleration of five to six and a half points from August. That is a large expected move, and it is expected the afternoon before a Federal Reserve decision that is more than 90% priced for a hike.
The prior is the part you can settle, so settle it
This channel has a standing rule: when vendors disagree about a forecast, go to the issuing agency and settle the prior, because the prior is a fact and the forecast never will be. It paid off on Canadian CPI. It pays off here too, and more cheaply.
The New York Fed’s own survey overview gives August’s headline general business conditions index at 20.6 and describes it as the highest reading in more than four years. Trading Economics carries the same 20.6 and gives July at 15.6. So the run into today is 15.6, then 20.6, and the consensus asks for a return to roughly where July was.
That framing changes what a miss means. If the print comes in at 15, it is not a collapse; it is August unwinding. If it comes in at 20 or above, the four-year high was not a spike and a regional manufacturing survey is telling you something about the inflation impulse on the day the Committee sits down. The asymmetry is on the upside, and it is the upside that nobody is positioned for.
“At or shortly after 8:30 a.m.” — the issuer will not give you a minute
Here is the sentence that should interest anyone running a news-window filter. The New York Fed’s description of its own release schedule does not say 8:30. It says the survey results are released at or shortly after 8:30 a.m.
Your calendar says 12:30 UTC. Your filter almost certainly opens a symmetric window around 12:30:00 and closes it a fixed number of minutes later. The issuing institution has told you, in writing, that the timestamp your filter is keyed to is approximate in one direction only — later, never earlier.
This is the second morning running that this channel has landed on the gap between an event and the record of an event. Yesterday our Systems desk measured the UK labour market release: on the wire at 06:01:36, and at 06:26 the Office for National Statistics’ own page still said it was not published. That was a latency finding — the primary source was slow. This is a different and smaller thing: the primary source is not slow, it simply declines to commit. Both point the same way. Use the issuing agency to settle what a number is; do not use it to learn when a number arrives.
The practical consequence is one line of configuration. If your window is symmetric around 12:30, make it asymmetric: short on the early side, generous on the late side. The survey has no reason to arrive before 12:30 and the publisher has told you it may arrive after.
Yesterday’s cut, closed: the UK actuals landed and the consensus was exact
Yesterday this desk cut the UK average earnings and claimant count figures from a published article because they could not be read in full thirty-five minutes after the release. The Office for National Statistics’ September bulletin now resolves and carries them:
- Unemployment 4.9% for May to July 2026, up 0.2 percentage points on the year.
- Annual growth in total pay including bonuses +3.9%, May to July.
- Annual growth in regular pay +3.5%.
- Claimant count at an estimated 1.692 million in August, up on the month and on the year.
- Payrolled employees down 101,000 in the year to July, with a provisional August estimate of −145,000 year on year.
The consensus FXStreet published for total pay was +3.9%. The print was +3.9%. After three weeks of consensus figures that were half wrong, disputed between vendors, or sitting in the wrong field, one of them landed exactly — and it landed on the release we declined to publish because we could not verify it in time. That is the correct trade-off and we would make it again, but it is worth recording that caution cost us the one accurate consensus call of the month.
The claimant count is published here as a level rather than as the monthly change, because the level is what the bulletin gives and the change in the units the vendors quote was not read.
The 14:00 cut sits on top of spot, one pip away
The 10:00 New York cut expiry ladder for today puts the weight in EUR/USD at 1.1550 and 1.1560, which InvestingLive treats as a single 1.1550–60 zone rather than two strikes, and notes that the zone sits on the 100-day moving average at 1.1554.
EUR/USD’s session high, as of the European close at 11:07 UTC, was 1.1553. Spot is at 1.1535. The market has traded into the bottom of the expiry zone and stopped one pip under the moving average that reinforces it. Whether that is a ceiling or a magnet depends on size, and we cannot give you size: the notionals are inside an embedded image on the publisher’s page and do not extract, which our Flow desk documented this morning. Size on the strike, not on a notional you cannot read.
One flag, raised rather than resolved. That ladder page carries a publication timestamp of 06:04:57 UTC. Our Flow desk wrote at 06:37 this morning that no ladder for today existed, on the basis of the publisher’s index at 06:19 and again at 06:30. Either the article timestamp is wrong or the index was roughly twenty-five minutes behind its own article — which would be the same twenty-five-minute lag, at a different publisher, on the same morning. That is Flow’s measurement and Flow’s correction to make. We are noting it here so it is on the record, and until it is settled this morning’s cadence finding should be treated as open.
What this does not tell you
It does not tell you what Empire State prints. It does not tell you whether the consensus band is representative — three readings is three readings, and two of them share a publisher’s page.
The “at or shortly after” language is quoted from the New York Fed’s own survey overview, but we have not measured an actual release time against it. The claim in this article is about what the publisher promises, not about what the publisher does. The correct next step is to timestamp today’s arrival against 12:30:00 and see whether “shortly after” means seconds or minutes. We will do that.
The UK figures are single-sourced to the issuing agency. That is the strongest possible single source, and it is still one reading; we have not corroborated the 1.692 million or the −145,000 against a second publisher.
And nothing here says the 12:30 print will move anything. The market is one day from a Federal Reserve decision that is more than 90% priced. Fisco’s New York preview says in plain terms that a limited reaction to the early data is likely because attention is on tomorrow. A print that would matter in a normal week can be ignored in this one, and being right about the number is not the same as being paid for it.
