The United States cash Treasury market does not trade today. Two sources agree on it and one of them is the body that recommends it. What neither of them gives you is a field your system can read: the industry association’s own holiday schedule lists today with no status at all, because in that table a full close is encoded as the absence of an early-close label. Its press release carries the status in prose instead. And the recommendation is not a closure — it is advice each firm decides for itself, covering dollar-denominated fixed income traded in three countries, not one. If your calendar holds today as a boolean called “US holiday”, every one of those four facts is missing from it.
There is no closure today, only a recommendation
The industry association confirmed on 28 October, fourteen days ahead, its previous recommendation for a full market close on Wednesday 11 November for trading of dollar-denominated fixed income securities. Its own release then says these are “recommendations only; each member firm should decide for itself” whether its fixed income departments stay open for trading.
That single sentence changes the object you are modelling. “Closed” is a binary and it is enforceable; a recommendation is a distribution over how many dealers actually went home. You cannot know the fraction, which is exactly why treating it as a closure is the wrong error: a rule that blacks out the session assumes zero liquidity, and a rule that ignores the day assumes normal liquidity, and the truth today is an unknown number between them. The practical form of that is not a directional view. It is a width: your slippage assumption is wrong in an unknown direction, so the only honest adjustment is to size smaller rather than to switch anything off.
And the scope is wider than the name suggests. The recommendation covers that trading in the United States, the United Kingdom and Japan, and the schedule lists the day in all three sections. A filter keyed to American hours blacks out the New York afternoon and leaves the London morning — the hours you are reading this in, the busiest liquidity window of the day — marked normal, when the same dollar instruments are recommended shut in London too.
The status field is an absence, and absences do not parse
Here is the defect in its cleanest form. On the association’s schedule page, the three nearest partial days are labelled explicitly and identically: Early Close (2:00 p.m. Eastern Time) against Friday 27 November, Thursday 24 December and Thursday 31 December. Veterans Day, Wednesday 11 November, carries no label of any kind.
So the table has a status field, it is populated for early closes, and it is blank for full closes. A parser reading that row cannot distinguish “this is a full close” from “we have no information about this date”. The two most common defaults — treat blank as normal trading, or treat any listed date as fully closed — get the full-close days and the early-close days wrong in opposite directions, and the first of those fails toward trading through a shut market. The page’s own metadata, incidentally, carries a publication date of 17 December 2025, 329 days ago, which is the correct way to publish a forward annual calendar and also means nothing on it has been restated since.
Now the part that inverts a rule this desk repeats constantly. We tell you to go to the issuing body rather than an aggregator for anything on a fixed cadence. On this field the aggregator is better. A general-market holiday page, last updated 30 September, carries two separate tables with the status as an actual value in a column: equities Open and bonds Closed for 11 November. It agrees with the association on all three 2 p.m. bond early closes, so it is not a sloppy page — it is a page that populated the field the issuer left blank. The issuer remains authoritative. Its schedule table is simply not where its own status lives, and that is a different claim from the one our rule makes.
One more thing that table gives you free: this is the second such day in 2026, not an exotic one. Columbus Day, 12 October, was equities open and bonds closed as well. If your logs cover October you already have a measurement of how this day behaves and have probably never looked at it.
Equities trade, bonds do not, and one of your inputs goes stale for a day
The asymmetry is the exposure. Foreign exchange trades through, equities trade, and the cash market in the instrument that repriced the Federal Reserve yesterday does not print. Any rule whose input set mixes a Treasury yield with an equity index or an FX rate spends today with one stale input and two live ones, and will not tell you which is which, because a yield that does not move looks exactly like a yield that chose not to.
That matters more than usual today specifically because of what moved yesterday. The December rate distribution repriced hard — our sister piece this morning retracts our own reading of it — and the natural instrument for expressing that repricing has no cash session in which to continue. A correlation estimated on days when all three inputs update is not the correlation that applies today.
We are going to stop short of where the argument wants to go. We did not establish whether Treasury futures keep normal hours today, and we are not going to assert it from the equity rows or from what we think we remember. If your exposure depends on that answer, it is one call to your broker and it is not a question a news desk should be answering from a holiday table.
And one source we read every morning gets this wrong in a way worth naming, because it is the sort of thing that survives translation into a stored field. A Tokyo research note published at 22:51 UTC yesterday tells its readers that US markets are closed for Veterans Day and that American participation will therefore be thin. The thinness is right and the scope is not: equities are open. “US markets are closed” and “the US bond market is closed” are different statements, and only one of them is today.
The slug-as-timestamp rule has a dependency, and it is exactly eight hours
Yesterday morning we published, as a one-off oddity, that a syndicated article whose address encoded 10 November 06:49 carried page metadata saying 9 November 22:49 — eight hours apart, on a publisher whose address slug this desk has treated as a reliable publication minute for years.
We found the second instance this morning, and it is not noise. A mirror of another publisher’s byline has an address ending 202611101838 and machine-readable metadata of 10 November 2026 at 10:38 UTC. Eight hours. Same direction: the address runs ahead of the metadata by exactly +8:00 in both cases.
Two instances with an identical signed offset is a diagnosis rather than an anecdote. A fixed eight-hour forward offset is what you get when a timestamp is rendered in a +8 zone on its way into a filename, and both instances are syndicated mirrors of a byline from elsewhere rather than the original publisher’s own pages. So the rule we have leaned on is narrower than we stated it: a slug timestamp is evidence on the publisher’s own domain and not evidence on a mirror. We could not test the original today — the publisher’s own article page for this morning returned an error on every attempt — so the dependency is identified and the control is still missing, and we say so rather than declaring the rule fixed.
A third slug defect turned up in the same hour and it is a year rather than an hour: the holiday table we relied on above sits at an address containing 2020 while its content is the 2026 schedule, last updated 30 September 2026. Read the date segment of every dated address, and then read the page anyway, because the segment and the content fail independently.
Two things we are closing: a dispute we cannot settle, and a board we were looking for at the wrong hour
We have carried a disagreement about a Dallas Federal Reserve speech time for six slots — 14:35 against 15:50 — and said each time that the way to settle it is a reaction note, something that could only have been written after the event. The event has now happened and no page we have read timestamps the remarks at all. We are formally dropping it. The method stands and the instance is dead: an item you cannot close is a debt, and the honest discharge is to say so in public rather than carry it to a seventh slot.
The option expiry board is the better closure, because we think we have the mechanism. Seven sessions without one, and this morning a Tokyo news index shows an item describing today’s New York expiries with the main strikes and sizes, stamped 12:40 UTC yesterday for yesterday’s cut. We could not open the body. But the stamp is the finding: the one publisher we know of that carries strikes with sizes publishes it around 12:40 UTC, and this slot runs at 06:00 UTC, more than six hours earlier. Today’s board does not exist yet when we go looking for it.
That reframes seven sessions of failure from a sourcing problem into a scheduling one, and it has a concrete consequence: the pre-London slot should stop treating a missing expiry board as a search failure, and either read yesterday’s board as yesterday’s or wait for the pre-US slot, which runs at 11:30 UTC and is the earliest one that can plausibly catch it.
Finally, something wrong on our own surface, which belongs here rather than in a mail to ourselves. The homepage navigation on this site links its expert-advisor page to /en/presets. The footer block the site generates beneath our articles does not: it offers /en/blog on one of yesterday’s pages and, on the other, query-string addresses of the form /en/fxevents?currency=USD. Our own in-body links are correct. If you have been following the footer, you have been following a different map from the one the navigation draws, and that is ours to fix, not yours to work around.
What this does not tell you
- We do not know what fraction of dealers are actually away. That is the whole point of the recommendation-versus-closure distinction and we cannot resolve it from any page. Nobody publishes it.
- We did not establish Treasury futures hours for today, deliberately, rather than inferring them from the equity rows.
- The expiry-board mechanism rests on one index headline whose body would not open. We are confident about the stamp, which is what the argument uses, and not about what the item contains.
- The eight-hour offset is two observations. Two is enough to call it an offset rather than noise and not enough to call it a property of the publisher. We also could not read the control case.
- We cannot tell you where today’s optionality sits, for the seventh session. We now think we know why, which is not the same as having the board.
- One measurement we can offer, because it changed overnight. This site’s own news list has returned a byte-identical page on five consecutive reads across about twenty-four hours, with a frozen top row. This morning it is a different length, both of last night’s articles sit at rows one and two, and its daily marker has advanced. So the freeze clears; it took somewhere between seven and nine hours last night. The thirty-row ceiling against a true count of 281 has not budged.
