A quarter-point hike in December is, on the one feed anybody names, fully discounted. That single fact decides what this afternoon can and cannot do. Six Federal Reserve speakers are scheduled between 14:00 and 23:30 UTC, the only piece of United States government data arrives at 13:30, and a $25 billion 30-year auction goes off at 18:00. None of them can move the December base case, because there is no room left in it. What they can move is a tail the market prices at about ten per cent, and a January meeting nobody has priced at all. Meanwhile the two calendars you would use to frame the 13:30 print disagree about its consensus by 12,000 claims, and both carry a number from the wrong week in the field marked actual. We settled that at the Labor Department in one fetch, and we settled one speaker’s clock at his own bank, where the majority turned out to be wrong.
27.6 basis points is one observation wearing three hats
A wire carrying a Federal Reserve speaker schedule, published 10 November at 14:58 UTC and date-checked by us before anything was taken from it, prints three figures for December from the same named source, CME FedWatch: a quarter-point hike “fully discounted”, a ten per cent probability on a half point, and 27.6 basis points of implied tightening against a previous reading of 24.7.
Those are not three signals. Take the first two as stated — 25 basis points certain, plus a ten per cent chance of a further 25 — and the implied total is 27.50 basis points. The page printed 27.6. Solve the other way and 27.6 requires a half-point probability of exactly 10.40 per cent, which is within four-tenths of a point of the ten per cent on the same page and, as it happens, identical to the 10.4 per cent this desk published yesterday from a separate wire reading of the same feed.
So the three numbers reproduce one another to a tenth of a basis point. If your system stores the 25-basis-point probability, the 50-basis-point probability and the implied-basis-points figure as three independent inputs, it is counting one observation three times, and any confidence it derives from their agreement is manufactured. Reproduce the aggregate from its components before you treat them as separate evidence. When the arithmetic closes this cleanly, it is not corroboration — it is the same number.
One honest limit: we cannot reproduce the 24.7 previous figure the same way. A ten per cent half-point probability with a certain quarter point gives 27.5, not 24.7, so the earlier reading must have carried live no-change risk. On a 1.4 per cent half-point probability, 24.7 implies about 96.0 per cent on the quarter point and about 2.6 per cent on no change. That is a reconstruction, not a reading, and we offer it as one.
What a fully priced meeting leaves six speakers to move
If December is decided, the reaction function for this afternoon is narrow and specific, and it is worth being blunt about which questions are live.
- The December base case cannot move. A quarter point is already in. Hawkish commentary cannot add it twice.
- The ten per cent tail can move, and it is the only thing in December that can. A half point is the single unpriced December outcome, and the condition its most explicit advocate attached to it has already fired: Governor Waller said in London on 10 November that a half point may be needed if October core consumer prices came in around 0.3 per cent or higher, and core CPI printed 0.3 per cent. The condition fired and the market pays about ten per cent for it. That gap is the position, not the direction.
- January is where the information is, and almost nobody will ask about it. Waller’s most useful sentence was about sequencing: it is not necessary to raise rates at consecutive meetings, and the precise timing does not matter so long as the promised moves arrive within a reasonable period. A December probability is silent on January by construction. A filter that scores every speaker against a December number is measuring the one meeting that cannot be repriced.
- Six speakers in nine and a half hours are not six independent draws. They are six people who have read the same CPI print and the same Waller speech. Treating each as fresh information and sizing six separate event windows is how you end up flat for the afternoon without ever deciding to be.
The schedule, in UTC: Hammack 14:00, Barr 15:00, Goolsbee 16:15, Daly 18:10, Schmid 18:30, Williams 23:30. Three European Central Bank speakers sit inside the same block, Vujčić at 15:30, Nagel at 17:30 and Lagarde at 19:30.
The one speaker time we could check at an issuer, and the majority had it wrong
We published this morning that two aggregators were fifteen minutes apart on Goolsbee — 16:15 against 16:30 — and told readers to use the earlier one if they were blacking out. That was caution, not evidence. Now there is evidence.
The Federal Reserve Bank of Chicago’s own speaking-engagements page carries the entry: 10:15 am CT on 12 November, at a financial news network’s investment conference. United States daylight time ended on 1 November, so Central Time is six hours behind UTC and 10:15 CT is 16:15 UTC. The earlier figure is right.
Look at what that does to vote-counting. Two sources said 16:30, including one that printed both 16:30 GMT and 11:30 EST, internally consistent with each other and wrong by fifteen minutes against the bank. One source said 16:15, and it matched the issuer. This file created a standing mark after four non-issuer sources agreed and a dissenting issuer was right; today is the mirror image, with the lone dissenter among the aggregators matching the issuer instead. The lesson is not that minorities win. It is that counting sources is not a method, and the only tie-break that has ever worked is going to the institution whose calendar it is.
There is a second field on that page worth more than the time. The issuer calls the event a moderated discussion, virtual, with a named journalist asking the questions. One aggregator calls it a speech. Those are different objects for anyone running a news filter: a prepared speech has a text, often released on the hour, and the market-moving sentence is in it. A moderated interview has no document at all, and the sentence that moves you is an unscripted answer to a question somebody else chose. This morning we published that the advice to prefer issuers is advice about documents, and that a remark made in a room has none at the moment it moves the market. A scheduled moderated interview is that same hole with a calendar entry on it: you know when it starts, and there is still nothing to read.
Three rows of one calendar, three field defects, settled at the Labor Department
Initial jobless claims at 13:30 UTC are the only United States government release today. We went looking for a consensus and found three, from three publishers: 203,000, 191,000 and 195,000. That is a 12,000 spread, 6.35 per cent of the prior level, and the three forecasts ask for +14,000, +2,000 and +6,000 against the same prior. One of the three pages states a consensus of 195,000 in its prose and says in its own table that the consensus is not yet published. We are not going to adjudicate that. The spread is your error bar and we are naming it rather than picking.
The prior, though, is settleable, and this is the discipline that pays: when vendors disagree about a forecast, go to the issuing agency for the prior. The Department of Labor’s own release, embargoed “8:30 A.M. (Eastern) Thursday, November 5, 2026”, is unambiguous, and comparing it against one widely scraped calendar convicts three adjacent rows:
- Initial claims. The calendar shows 189,000 as an actual against a week it labels 7 November. The issuer says 189,000 is the week ending 31 October. Right number, wrong week — and it sits in the actual column for a release that has not happened yet.
- The four-week average. The calendar shows 194,750, again labelled 7 November, again the issuer’s 31 October figure. Its previous field reads 192,500, which matches neither the issuer’s prior average of 197,000 nor the 196,750 that figure was revised from. No value at the issuer corresponds to it.
- Continuing claims. The calendar shows 1,718,000 against a week it labels 31 October. The issuer says 1,718,000 is the week ending 24 October. Its previous field reads 1,710,000, and 1,709,500 is the issuer’s insured-unemployment four-week average for that week, not the prior level of 1,695,000. A rounded average has been placed in a level field.
The continuing-claims shift is the one that will cost you money rather than embarrassment. The Labor Department publishes continuing claims a week behind initial claims by design — one release, 31 October for one series and 24 October for the other, a built-in seven-day lag. A vendor that aligns both to a single date has deleted that lag. Any ratio, spread or difference your system computes between the two series off that calendar is differencing two different weeks, and it will look perfectly reasonable while doing it. The revisions matter too: the issuer revised the prior initial-claims week up by 1,000 and revised prior continuing claims down by 6,000, from 1,701,000 to 1,695,000. Score today’s print against an unrevised prior and you have mismeasured the change before you start.
Note what this does not require. We did not need a second aggregator, and a second aggregator would not have helped, because the defect is in the week label rather than the value — two vendors can agree on 189,000 and both file it under the wrong Friday. The only instrument that finds this is the issuer.
What this does not tell you
- We do not know where the market prices a half point now. Our ten per cent is from 10 November and our 10.4 per cent from yesterday; neither is a reading taken this afternoon, and we did not find a feed-named figure stamped today.
- We have not established what any speaker will say. Nothing here is a view on the December decision, and three of the six have no stated stance on the page we read.
- The 24.7 basis-point previous figure is reconstructed, not read. Our 96.0 and 2.6 per cent decomposition is arithmetic on an assumption and could be wrong about the assumption.
- We did not check the other five speakers’ times at their own banks. Only Goolsbee’s was settled at an issuer, and the others rest on two aggregators agreeing — which is exactly the evidence that failed here.
- We do not have a consensus for continuing claims from any source. One page carried a figure for a week in August, which is not it.
- We did not read today’s wire schedule at the issuer for Barr, who appears on one aggregator and our own earlier table and not on the wire schedule at all.
- The 30-year auction result lands at 18:00 UTC, after this note. We have said before and repeat: the when-issued yield every “tail” figure is measured against is a dealer quote the Treasury does not publish, so whatever tail you read tonight is somebody else’s subtraction. For a size-matched baseline use the 13 August $25 billion new issue at 5.216 per cent, not the 8 October $22 billion print at 5.618, and say which you mean.
- One calendar we read carries 5.618 per cent against today’s 18:00 auction row. That is October’s result, not a forecast and not today’s. A system scraping that row before 18:00 inherits a result that does not exist yet.
Related
- FX events calendar · event windows in UTC
- Signals
- EA presets
