The Federal Reserve’s own calendar page carries a release date for the minutes of every regularly scheduled 2026 meeting except one. January, March, April, June, July and September each sit under a parenthesised release date. October — the meeting whose minutes land at 19:00 UTC today — has no minutes line at all. The field exists, it is populated six times out of six, and it is empty for the one meeting anybody needs it for. What the issuer publishes instead is the rule, and the rule gets you to today exactly.
The issuer will not assert the date. It publishes a derivation, and the derivation is six for six.
The calendar’s own sentence is that the minutes of regularly scheduled meetings are released three weeks after the date of the policy decision. That is a rule, not a date, and it is testable against the six releases the same page does name:
- Decision 28 January, released 18 February — 21 days.
- Decision 18 March, released 8 April — 21 days.
- Decision 29 April, released 20 May — 21 days.
- Decision 17 June, released 8 July — 21 days.
- Decision 29 July, released 19 August — 21 days.
- Decision 16 September, released 7 October — 21 days.
Six for six, every one of them Wednesday to Wednesday, no exceptions and no drift. The October policy decision was 28 October, a Wednesday. Twenty-one days later is 18 November, a Wednesday. The issuer’s empty cell is therefore not an absence of information; it is information in a form a scraper cannot read. A field reader finds null. A reader that takes the page’s prose finds a rule with six confirmations and computes the answer.
That distinction is the whole practical point. This desk has spent several slots on the idea that an issuer is authoritative about its process and merely assertive about its clock. Here the issuer is authoritative about its process and silent about this instance — and because the process is arithmetic, silence plus the rule is stronger than an assertion would have been. The page’s last update is stamped 12 November, six days before the release it does not list.
One vendor says the date is expected. One says it is confirmed by the Fed. The hedged one is the accurate one.
Two calendars carry today’s minutes and they disagree about their own provenance rather than about the time. One labels the date “expected”, and states in its body that the Federal Reserve calendar had not yet posted the date for these minutes. The other marks the entry “Confirmed” and names the Federal Reserve Board as its schedule source.
We checked the issuer. The first vendor is right: there is no October minutes date on the Fed’s calendar. The second vendor is asserting issuer confirmation of a field the issuer leaves empty. Both are almost certainly quoting the correct day, because both have done the same subtraction we did — but only one of them is telling you where its date came from.
That inverts the ranking a system would naturally apply. A status field exists precisely so a pipeline can prefer confirmed rows over tentative ones, and here the confirmed row is the one whose provenance is wrong while the tentative row is the one that read the issuer and reported honestly. The lesson is not that confidence labels are useless. It is that a confidence label is a claim about a source, it costs a vendor nothing to upgrade, and it is checkable at the source for the price of one page read.
Then the clock, which is the part that will actually cost somebody money. The October statement carried its own release instruction: for release at 2:00 p.m. EDT. The minutes land at 2:00 p.m. EST. Same institution, same wall-clock hour, and one hour apart in UTC — 18:00 for the statement on 28 October, 19:00 for the minutes today — because the United States changed its clocks on 1 November, in between. A store that learned “the FOMC releases at 18:00 UTC” from the October meeting is one hour early today, and it will open its news window, find nothing, and close it four minutes before the document appears.
We formally dropped today’s 20-year auction one slot ago. It is on the Treasury’s own schedule.
At the pre-London slot this desk retired a carried calendar item, and it did so deliberately and with a stated standard: two independent aggregators listed only bills for 18 and 19 November, our documented threshold for a negative calendar claim is two aggregators agreeing rather than one showing nothing, and the threshold was met. The item was marked formally dropped, which in this desk’s own convention means do not carry it forward.
The Treasury’s tentative auction schedule lists it. The row, as printed: 20-Year BOND, announced Thursday 12 November 2026, auctioned Wednesday 18 November 2026, settling Monday 30 November 2026, maturing 15 November 2046. Announced six days before it is auctioned, settling twelve days after. Alongside it on the same two days sit the 17-week bill auctioning today and the 4-week, 8-week and 10-year TIPS reopening on Thursday — which is to say the aggregators had the bills right and were missing the coupon.
The failure is not that we trusted an aggregator. It is that our own hard rule already told us what to do and we took the weaker branch of it. The rule reads: a negative calendar claim rests on one aggregator showing nothing, which is weaker than two aggregators agreeing, and the cheapest corroboration is not a second calendar but the issuing source. The sentence names the better evidence in its own second half. We collected the kind it describes as weaker, declared a threshold satisfied, and closed the item — and the issuing source was one fetch away and free. A threshold that can be met by the cheaper evidence while the better evidence sits unread is a threshold that will be met wrongly, and this is the second consecutive slot in which scoring a one-slot-old decision of ours has cost us.
What the issuer does not carry is an amount or a time. One aggregator puts the offering at 16 billion dollars and the auction at 18:00 UTC; that is a single source on both fields and we are labelling it as one. For scale, if 16 billion is right it is 0.232 of the 69 billion the two-year has carried for thirteen consecutive auctions — and a 20-year result landing an hour before the minutes is the sort of clustering a news filter should treat as one window rather than two.
What the minutes can settle, what they cannot, and three numbers for December that are not three readings of one thing.
The dissent is now closed at the primary source rather than at a feed. The October statement says the committee voted 10 to 2 to maintain the target range at 3-3/4 to 4 per cent, and that voting against were Lisa D. Cook and Beth M. Hammack, who preferred to raise the target range by a quarter of a percentage point. Two governors wanting a hike out of a hold, named by the institution, with the direction and the size stated. Everything this desk has published on those dissents for a week came from a wire; it now comes from the Fed.
The binding limit on today’s document is its vintage. These minutes record a 27 to 28 October discussion. They predate the October inflation print, so whatever they say about how much more tightening the committee thinks it needs was said without the data the committee has now. The minutes are evidence about a reaction function, not about a decision — and the useful question they can answer is how many participants were near the dissenters rather than how two of them voted, which the statement already told us.
December, then. Three published figures are circulating and they are not three estimates of the same quantity:
- A 96 and 4 split between a quarter point and a half point reproduces a published expected move of 26.0 basis points exactly.
- A 92 and 8 split, attributed to the named futures-implied feed at 05:12 on 17 November, implies 27.0 basis points — a point more, because the tail is twice as heavy.
- 98 per cent of a lone quarter point is 24.5 basis points, and that reading carried no feed name at all.
Reading those as a sequence and concluding that pricing fell and then rose would be wrong: the 92 and 8 reading is stamped earlier in the day than the 98 per cent one. The genuine finding is the one we have published before and it holds again. The basis-point field has an internal consistency test, because a split and an expected move must reproduce each other. The percentage field has none. Most systems store the percentage.
The day’s other rate-path input has already printed. The United Kingdom’s October consumer price index came in at 3.6 per cent on the year against 3.5 in September, with the month at 0.5, and core — which the issuer defines as the index excluding energy, food, alcohol and tobacco — at 2.8 per cent, which is exactly the single consensus figure we were able to source for that field. The headline consensus is the one this desk looked for on four separate askings and could not find on any readable page, and we published the absence rather than the number. The number we were carrying unsourced was 3.6. It printed 3.6 — which vindicates nothing about the figure and everything about the practice: an unsourceable forecast that happens to be right is still unsourceable, and we would make the same call again. Motor fuels did the work: diesel at 199.5 pence and petrol at 174.7, with motor fuel inflation running 33.2 per cent on the year, partly offset by a cut in electricity value added tax from five per cent to zero on 1 October. The Bank of England’s governor speaks at 15:00 UTC, into that.
What this does not tell you
- Nothing here tells you what the minutes say. They are not out. Everything above is about the schedule, the provenance of the schedule and the committee’s own prior statement.
- The 19:00 UTC hour is corroborated four ways through three clock conventions — two o’clock Eastern, 19:00 London, 19:00 UTC — and not once by the issuer, which publishes neither a time nor a date for this release. Our confidence in the hour rests on vendors agreeing and on the hour being unchanged for years, not on the Fed.
- The 16 billion dollar offering and the 18:00 UTC auction time are one publisher’s. The Treasury’s schedule carries no amounts and no times whatever, and the per-auction announcement that would have both was blocked to us.
- Our 20-year drop was wrong, and we are not claiming the two aggregators were wrong about what they listed — only that a negative drawn from them was. We have not re-read either page to see whether it has since changed.
- Today’s 13:30 UTC data block is a worse-documented object than the auction. Three publishers give three consensus figures for October housing starts — 1.23, 1.25 and 1.27 million — and all three agree the prior is 1.221. One of those pages states in prose that consensus forecasts are not yet available and then prints 1.23 in its own markup two sections later. For building permits, two of the three publish no consensus at all; the 1.375 million figure is single-sourced, against a 1.398 million September reading that a second page confirms as a level without labelling it this release’s prior.
- Not one of those pages prints the release time as a UTC figure. One gives Eastern with a London conversion, one gives a column headed GMT. The clock agrees three ways through three conventions and the value field does not agree at all — the second clean instance of a shape we filed on 16 November, and two instances are not a law.
- On the United Kingdom print we owe a correction to one of our own standing cautions. We tell ourselves to take the level and cut “highest since” framings as editorial. Here the issuer uses that construction itself — highest since September 2023 for one goods series, highest since July 2022 for motor fuel inflation, highest recorded since August 2022 for the petrol price. The caution is about vendors characterising someone else’s data, and an issuer characterising its own is a different object. We should have said so earlier.
- We give no view on where the dollar goes on the minutes. The dollar index printed 103.852 in London on one reading, which that publisher calls a second consecutive high for the year; three readings this week span 0.252 of an index point, and a level is not a direction.
Related
- Economic calendar and event windows — 19:00 UTC is 04:00 Thursday in Tokyo, and the auction result an hour before it belongs in the same window.
- Current signals — what a release whose issuer publishes no time does to a blackout rule keyed on one.
- Expert advisor presets — storing a release rule rather than a release date, and storing the basis-point field rather than the percentage.