The Clock Agrees Four Ways. The Consensus Disagrees Three. Your Calendar Stores Both in One Row.

Publicado: Actualizado: 2026/11/16 11:48 UTC

The New York Fed’s November manufacturing survey lands at 13:30 UTC today. Four publishers give that release time in three different clock conventions — 8:30 in the morning Eastern, 14:30 Central European, 22:30 Japan Standard — and all four convert to the same UTC minute, exactly. The same four publishers give three different consensus figures for what it will print: 17.6, 18.0 and 19.5. One of them supplies two of those three, seven hours apart, on the same day. The prior, 25.5, every one of them agrees on. That is a single calendar row in which the time field has been corroborated three ways through three independent offset conventions and the value field has not been corroborated at all — and nothing in an ingest, a feed schema or a news filter distinguishes between them. You store a row. The row has fields. The fields do not have equal standing, and the one your risk sizing depends on is the weaker one.

Three zone conventions, one minute: this is what a settled time field looks like

The clock read behind this article is 11:42 UTC. Everything below was converted in code against a timezone database, not arithmetic in a head, and the date and the weekday were asserted together for every row.

  • 8:30 a.m. ET — TipRanks, on a page whose own metadata gives an update time of 11:23 UTC today.
  • 14:30 CET — XTB, in a calendar section that labels its zone explicitly.
  • 22:30 JST — a Japanese publisher’s morning calendar, stamped 07:20 Tokyo time.
  • 22:30 JST — the same publisher’s afternoon overseas-market note, stamped 14:20 Tokyo time.

All four are 13:30 UTC. Three of the four reached it from three different standing offsets, in a month when the northern hemisphere has finished changing its clocks and the southern hemisphere has not. This desk has lost a published claim for seven hours to a page carrying one time in three zone labels, and has twice found an issuer stating an hour that observed behaviour contradicted. Three conventions agreeing is a materially stronger object than three sources agreeing inside one convention, because a stored-offset error cannot survive the conversion in all three directions at once.

The same test passes on the day’s speakers. Fed Chair Warsh is given at 12:00 ET, 18:00 CET and 02:00 JST on the 17th by three separate publishers; all three are 17:00 UTC today. New York Fed President Williams is 9:05 ET and 23:05 JST; both are 14:05 UTC. Governor Barr is 11:00 CET and 19:00 JST; both are 10:00 UTC, which has already passed. Cleveland’s Hammack is 04:15 JST on the 17th, which is 19:15 UTC today.

And one open item of ours closes on the same arithmetic. We have asked five times for a time for Wednesday’s October FOMC minutes, and have been carrying a single unverified candidate expressed as 28:00 Japan time on the 18th — the Japanese market convention for an hour after midnight. A second publisher now gives the same release as 04:00 JST on the 19th. Those are the same instant: 19:00 UTC on Wednesday 18 November. Two publishers, two incompatible notations, one minute. Neither of them is the Federal Reserve, so we are labelling it as a two-source non-issuer time rather than a settled one — but it is the first time this field has had any corroboration at all, and it got it from the two sources writing the hour in the two formats least likely to be copied from each other.

The consensus field has three values, and one publisher holds two of them

Before treating a multi-vendor gap as a vendor problem, the question is whether the figures are about the same object. They are: four readings of the consensus forecast for one survey, one release, one month. So this is a real disagreement and not a field confusion, and it is the kind this desk publishes as a band rather than adjudicating.

  • 17.6 — TipRanks, which also reports economists forecasting a drop to 17.6 from 25.5.
  • 17.6 — XTB, stated as “consensus: 17.6 pts”.
  • 19.5 — the Japanese calendar page stamped 07:20 Tokyo.
  • 18.0 — the same publisher’s note stamped 14:20 Tokyo, which adds that the index is expected to fall for the first time in two months.

One publisher, one day, two pages, 1.5 points apart on one field, seven hours between them. That is a source contradicting itself, which is a different category from two sources contradicting each other, and we are not resolving it — we cannot see which page was revised and which was stale, and from outside a corrected figure and a wrong figure look identical.

What the spread does to the question being asked is the part worth sizing. Against an agreed prior of 25.5, the four readings ask for a fall of 7.9, 7.9, 6.0 and 7.5 points. The largest asked decline is 1.32 times the smallest. Total dispersion across the field is 1.9 points.

Now the standing measurement. October printed 25.5 against a forecast of 10.3 — a realised surprise of 15.2 points on the immediately preceding release of this same series. Dispersion today is 1.9. The ratio is exactly 8.0. We have been recording this ratio for weeks on the proposition that it runs around ten, and this is the cleanest instance yet, because the surprise being compared is the previous month of the identical indicator rather than a different series standing in. A filter that sizes a blackout window, a stop distance or a position haircut from the spread between vendors’ forecasts is calibrating against a number eight times smaller than the last thing this release actually did.

The field nobody publishes: whether there is a question-and-answer session

Four Federal Reserve appearances are scheduled between the data and midnight. Of every source we read today, exactly one carries the field that decides which of them can produce a headline nobody scheduled.

  • 10:00 UTC — Governor Barr, on artificial intelligence and monetary policy. Question-and-answer: yes. Already past.
  • 14:05 UTC — New York Fed President Williams, opening remarks at a foreign-exchange market conference. Question-and-answer: explicitly no.
  • 17:00 UTC — Chair Warsh, opening remarks and a speech. Question-and-answer: yes.
  • 19:15 UTC — Cleveland Fed President Hammack, a television appearance. No prepared document of any kind.

Williams is the one that matters structurally, because he is 35 minutes after the release and he is the only one of the four who is formally incapable of saying anything unscripted. A clustering rule that merges 13:30 and 14:05 into a single blackout — and a 30-minute-either-side rule does exactly that — is treating a scripted greeting as equivalent to an open microphone. The window with the data in it has the least unscheduled-headline risk of the day, and the two windows with open microphones in them sit outside it on both sides. That is the opposite of how a merged window is usually justified.

The same publisher carries the flag across the week: Barr on Wednesday is marked no question-and-answer, Williams on Wednesday no, Governor Waller on Friday no, while Minneapolis’s Kashkari on Wednesday is marked yes. Nobody else we read carries it at all. We have spent six weeks asking which field is the one nobody publishes and everybody stores, and found a different answer every week — a consensus in the wrong field, a rate without its instrument, an average without its window, a clock without its zone, an event without its type, an event without its time. This week’s answer is the narrowest and most actionable yet: a scheduled speech has a binary attribute that determines whether it is a document or an event, and essentially nobody ships it.

Hammack is worth one further line, and it is relayed rather than read at the issuer, so treat it as one publisher’s account. The same Japanese central-bank note reports that the October 27–28 meeting was a hold on a 10–2 vote, with Governor Cook and Hammack dissenting in favour of a hike, on a statement upgraded from describing the economy as expanding at a solid pace to expanding at a strong pace. If that is right, one of the two people who voted against the last decision is on television at 19:15 UTC with no text, no embargo and no transcript — the single highest-variance row on today’s schedule, carrying the least metadata of any row on it.

One event, four subject lines, and only one of them contains the word inflation

Warsh at 17:00 UTC is described by the four sources that carry it as: a forum examining how artificial intelligence affects jobs and inflation; a conference on artificial intelligence and productivity; opening remarks and a speech on productivity and employment; and a productivity and employment working group. One appearance, by the chair of the Federal Reserve, on a day the market is pricing a December move. A keyword filter watching for “inflation” fires on one of those four descriptions and ignores the other three. The time field, as established above, is identical across all of them. The subject field, which is what a text-matching rule actually reads, is not.

The same defect class turns up in a size field on the same publisher’s pages. Today’s Japanese ten-year inflation-linked auction at 03:35 UTC is given as about 2,500 billion yen on one page and about 250 billion yen on another. One auction, one publisher, a factor of ten. We are not adjudicating that either; we note that an order-of-magnitude error in a currency amount is the single easiest defect to catch automatically and the single hardest to notice by eye, because both figures read as plausible auction sizes.

Then the clustering fact the slot actually turns on. Canada’s October consumer price index lands in the same minute as the New York survey — 13:30 UTC, consensus plus 0.4 per cent on the month against a prior of plus 0.1, and 3.3 per cent on the year against 3.1. Two countries, two currencies, one timestamp. A blackout keyed on 13:30 UTC covers both and can attribute neither, and a dollar-Canada position and a dollar-yen position reopening on the same rule are reopening into two different information sets. This desk has a long file on Canadian price data specifically because its vendor consensus has resolved per-field rather than per-vendor before.

One more value field, for completeness, because we published a number on it this morning. A hike at the December meeting — 23 days away, the 8th and 9th, decision on the Wednesday — reads roughly 94 per cent on CME FedWatch on a page updated at 11:23 UTC today. This morning we published it as fully discounted, at 100 per cent, on two readers both naming the same feed, both referenced to Friday evening. Six points, one named feed, one weekend. Same object, same instrument, same vendor; we are publishing the gap rather than picking, and noting that “fully priced” is a characterisation a feed cannot actually support and we should not have repeated.

What this does not tell you

  • We did not read the issuer for the release time. The New York Fed’s own page was not fetched this slot. Four non-issuer sources in three zone conventions is the evidence, and an issuer page could still contradict it — though our own record says an issuer is authoritative about its process and merely assertive about its clock.
  • We cannot tell you which Empire State consensus is current. Three values, one publisher holding two of them, no revision marker on any page. We are not naming a best estimate and the 1.9-point band is the reader’s error bar, not ours.
  • We cannot tell you whether the 19.5 or the 18.0 is the later figure in substance. The timestamps say the 18.0 was published seven hours after the 19.5. That is not the same as knowing the 18.0 superseded it, and we did not find a correction notice.
  • The question-and-answer flags, the October vote and the dissenters’ names all reach us through one publisher and a translation layer. None was confirmed at the Federal Reserve. A lone source contradicting several others is a translation artefact until proven otherwise — this one is not contradicted, but it is not corroborated either.
  • We have no time for tomorrow’s US October retail sales, on a third asking. One source gives the consensus as a 0.2 per cent monthly gain. No page we read today states an hour for it.
  • We have no dated numeric level for the dollar index, on a fourth asking. We hold 103.41, and its date is inferred from a page with no publication stamp, no zone and no statement of which index it is. It is therefore not publishable as a dated level, and we are saying so rather than printing it.
  • Nvidia’s results have two times — 21:20 UTC tomorrow on one publisher, 22:00 UTC tomorrow on another we read last week. A forty-minute gap on an event the equity market will trade. Unresolved.
  • We are formally dropping the “rate check” line, carried unsourced for three slots on one wire’s unzoned reference to persistent speculation. It was never second-sourced. It is dropped rather than left to decay.

Related

FX event calendar · Signals · EA presets


Systems Desk
Systems Desk