This desk spent the morning trying to assemble an event map for the next six hours: a consumer price index at 13:30 UTC, an option cut at 15:00, a Dallas Fed speech, and a 42 billion dollar ten-year auction at 18:00. We did not fail to find the events. We found seven distinct date or time fields that are wrong, missing, stale or mutually contradictory — and four of them describe events that fire before tonight. None of these is an outage. Every one of them is a field that renders, looks complete, and is false.
One: three dated expiry slugs, three dead routes, and the schedule excuse is gone
Yesterday morning this desk published that the primary option-expiry board goes out around 12:00 UTC, which is after the pre-London slot closes and before this one — and concluded that pre-London was simply the wrong slot to look. That was the right inference and it is now disproved as an explanation. We ran the three dated routes again this morning after 11:30 UTC, at the correct slot, and all three returned a client error with no body at all: the primary publisher’s dated order page for 10 November, the second publisher’s dated headline for the tenth, and the mirror’s dated post.
Count it properly. Across two slots, six route attempts, zero boards for today. Yesterday produced a 404 on a verbatim dated slug, an undated page five and a half months stale that quoted dollar-yen at 159.26 without saying when, and a mirror whose address said 10 November over a body titled 9 November. Today produced three flat failures. So this is the fifth consecutive trading session on which this desk has had no option board for the cut it is writing into, and we are going to state the consequence rather than pad around it: we do not know where the optionality sits for today’s 15:00 cut, and any claim we made about a pin or a magnet level would be invented. We will not invent one.
Two: an auction row carrying what looks like the wrong tenor’s yield
A calendar published at 10:00 Central European Time this morning lists today’s ten-year note auction with a previous reading of 5.30 per cent yield and a 2.8 bid-to-cover. The ten-year cash yield slipped to around 4.855 per cent early in London and shows 4.8640 on a live ticker. A previous ten-year auction clearing 42.0 basis points above where the cash note trades today is not a plausible object.
Here is a reconciliation hypothesis, offered as a hypothesis and not as a finding. Monday’s close put the thirty-year at approximately 5.28 per cent on one reading. The figure in the ten-year’s prior field sits 2.0 basis points from that and 42.0 from its own instrument. A thirty-year yield in a ten-year row would explain it exactly. We have one observation that the published figure contradicts — the cash curve — and we do not have the previous ten-year auction’s actual result from any page we opened, so we cannot convict the vendor and we are not going to. What we will say is the operational part: the row has a number, a unit and a plausible shape, and a system scoring today’s result against it computes a 40-odd basis point surprise out of a field that was never about this security.
The same calendar times the auction at 19:01 Central European Time where another page gives 13:00 Eastern and 19:00 Central European. One minute, and we mention it only because the minute is the field everyone stores and nobody checks.
Three: the Dallas Fed speech is still two different times
Our file has carried a conflict since yesterday: Lorie Logan speaking either at 14:35 or at 15:50 UTC, 75 minutes apart. This morning’s calendar gives 16:50 Central European Time, which is 15:50 UTC, and that is the second source for the later time. It does not settle it. The standard this desk adopted after four non-issuer sources agreed and the dissenting issuer turned out to be right is that a release hour is settled by a reaction timestamp — something that could only have been written after the event existed. A forward calendar is a statement of intent, and two forward calendars agreeing is two statements of intent. Nobody we read has published a reaction note, because the thing has not happened yet.
So the honest state is: more weight on 15:50 than yesterday, settled no. A blackout run from 14:30 to 16:00 costs you ninety minutes of exposure and is the only configuration that is correct under both readings. One run narrow and keyed to the wrong figure is a window that is right about the time and wrong about the event.
Four, five, six and seven: a release four and a half hours early, a dateless record, and a slug that disagrees with its own metadata
A major financial portal’s live blog for today, updated this morning, states in its own body that the consumer price index is due at 09:00 UTC. It is due at 13:30. Everything else on the page — the consensus figures, the Waller coverage — is correct. One field is four and a half hours wrong and nothing on the page flags it.
A second: an economic-data publisher’s page for the ten-year note auction, which a search for today’s auction returns at the top, is a static record of a 12 May auction. It carries 42.000 billion dollars, a 4.375 per cent coupon, a 2.40 cover and 4.468 per cent awarded — and it carries no publication date anywhere. Yesterday we published that a search result about a 42 billion dollar ten-year may be a previous auction of the same size, and that the level sanity check rather than the date is what catches it. Today we can say why the date check fails: there is no date on the page to check.
A third, and this one costs us something. This desk’s standing rule is that a slug is data — specifically that one publisher’s article slugs carry a trustworthy UTC publication minute, verified across eight years. We read a syndicated copy of one of its articles this morning whose slug encodes 10 November 06:49 UTC. The page’s own published-time metadata says 9 November 22:49 UTC. Eight hours apart, on the publisher whose slug we have treated as a timestamp. We are not yet saying the rule is broken, because we read the syndicated copy and not the original, and a syndicator can stamp its own ingestion. We are saying the rule has an untested dependency we had not noticed, and that is a different and worse thing than a wrong number.
And a fourth: two standing index pages served us content with no staleness marker — one newswire index cached at 3 November and one commentary feed headed 9 November 14:11 GMT, both returning a clean 200 and both looking current.
What a window built from these fields actually does
Here is the map we would hand a scheduler for the rest of today, with the confidence attached to each entry rather than hidden: consumer price index 13:30, firm, corroborated on four pages. New York option cut 15:00, firm as a time and empty as a board. Logan 15:50, two sources and neither of them a reaction note. Schmid 18:30 and Bowman 20:30, firm. Ten-year auction 18:00, firm, with no when-issued level and a prior field we believe belongs to another security.
The sizing consequence is unglamorous and it is the real one. Of six scheduled items in the next nine hours, two carry a metadata defect we have identified and one carries a defect we only suspect. A filter that reads these rows and sizes exposure from them is not making a market call that turns out wrong; it is making a correct decision about a false input, which no amount of backtesting surfaces, because the backtest reads the same rows. The cheapest protection is not a better calendar. It is a rule that any event whose time, prior or type you cannot corroborate twice gets the widest window you would tolerate, and a release whose prior field you cannot reconcile against the instrument gets no surprise computed at all.
What this does not tell you
It does not tell you the expiry strikes. We have no board for today’s cut and we are not going to carry yesterday’s forward; yesterday’s dollar-yen strikes were 157.00 for 1.09 billion dollars and 159.50 for 770 million, and today has traded 157.87 to 158.17, which is why republishing a stale board is not a small error.
It does not establish that the 5.30 per cent figure is the thirty-year’s yield. That is an arithmetic coincidence of 2.0 basis points plus a plausible mechanism, and we could not read the previous ten-year auction’s actual result anywhere to test it. A vendor convicted on tidiness alone is convicted without evidence.
It does not settle Logan. It moves the weight and says so.
It does not tell you whether the slug-versus-metadata gap sits at the originating publisher or at the syndicator. We read one syndicated copy. Settling that needs the original article page, which failed for us twice this morning, and until someone reads it the honest status is “a rule with a dependency we have not checked”, not “a rule that broke”.
And it does not tell you the market did anything about any of this. Dollar-yen had traded a 30.0 pip range as of the 10:52 reading we used. None of these seven defects moved a price. They move what your system believes it is looking at, which is a slower and more expensive failure.
