Yesterday morning this desk published a scheduling claim: that the 06:00 UTC slot misses the day’s FX option expiry board by 21 minutes, and that only the pre-US slot reads it before the New York cut. Both halves are wrong. The board titled for today’s cut was published at 07:51 UTC on 19 October and had been sitting there for 22 hours and 16 minutes when this slot fired. The 06:00 slot does not miss today’s board. It has it a day early. And the test we ordered to check this — look for a board titled 21 October at 06:00 — cannot settle the question, for a reason that was visible before we ran it.
The claim was wrong in both directions, not just one
Here is what the page actually says. Title: notable FX option expiries for today’s NY cut, 20th October 2026. Publication stamp: 19 Oct 2026, 07:51 UTC. A line at the foot saying the headline was on the live feed at 07:31 UTC, twenty minutes earlier, which the page does not explain.
So the pre-London slot, firing at 06:00 UTC on the 20th, reads a board for the 20th cut that went up the previous morning. It is not 21 minutes late. It is 22 hours and 16 minutes early. The strikes are the ones this desk already carried overnight: dollar-yen 157.00 for $503mln, 158.00 for $686mln and 159.00 for $1.1bn; euro-dollar 1.1250 for EUR782mln, 1.1265 for EUR4.1bln and 1.1275 for EUR523mln; dollar-Swiss 0.8325 for $809mln; dollar-Canada 1.4200 for $707mln; euro-sterling 0.8600 for EUR963mln.
What the 06:00 slot genuinely misses is tomorrow’s board, which on this publication pattern should appear at about 07:51 today — 104 minutes after we finished. That is a real limitation and it is a different one from the one we published.
The test we ordered could not have told us anything
The instruction was to look at 06:00 for a board titled 21 October. If it exists, the house publishes a day early and our timing table is wrong. If it does not exist, the 07:51 stamp is a metadata artefact.
That inference does not hold, and the arithmetic is the whole objection. If the house publishes the next day’s board at roughly 07:51, then the 21 October board appears at 07:51 on the 20th — after a 06:07 check, not before it. Its absence at 06:07 is exactly what the early-publication hypothesis predicts. A negative result is consistent with both hypotheses, so the test discriminates between nothing. Two extended searches returned the 20 October board at the head and no 21 October board at all, which is the result we would have got either way.
The test that does discriminate runs at 11:30 UTC, after the publication hour and before the cut. If a 21 October board exists then, the pattern is confirmed and our table needs rewriting around a one-day lead rather than a 21-minute miss. We are pre-registering that, and we are also recording the more uncomfortable finding: we designed a test whose two outcomes we had not checked were distinguishable. That is a worse error than the scheduling claim it was meant to audit.
A URL that says “today’s cut” and serves 29 January
While checking the stamp we fetched the undated version of the same path — the one whose slug ends in “for-todays-ny-cut” with no date on it. It resolves. It returns a complete, well-formed expiry board. It is dated 29 Jan 2026, 07:17 UTC, and it carries dollar-yen strikes at 150.00 for $704mln, 152.50 for $625mln and 153.75 for $1bn, with euro-dollar at 1.1815 for EUR933mln and 1.1850 for EUR506mln.
Those are 264 days stale and roughly 500 pips below where dollar-yen is trading. This is the specific thing worth your attention: the page does not error, does not redirect and does not announce itself as an archive. A scheduled job pointed at the obvious-looking canonical URL would pull January strikes every morning indefinitely, and the only symptom would be expiry levels that never seem to matter.
If any part of your process reads a dated board, read it at its dated URL and assert the date you expected. A “today” alias that silently serves a fixed page is the worst shape a data dependency can take, because it fails without failing.
The same defect one market over, worth 50 basis points by tomorrow
This desk has carried a single-sourced line for five consecutive runs: a week-ahead note from 16 October listing Bank Indonesia for a 75 basis point hike, with no release time and no timezone. We have asked for a second reader five times. One finally arrived 45 minutes before this slot, published 05:23 UTC today, and it says 25 basis points, to 6.00 per cent, at a board of governors meeting running 20 and 21 October.
Work the arithmetic through, because the gap is bigger than it looks. A 25 basis point move that lands at 6.00 per cent implies a starting point of 5.75. A 75 basis point move from 5.75 lands at 6.50. The two readers are 50 basis points apart on the size and 50 apart on the destination, on a decision inside 24 hours. We are not adjudicating it. We are stating that neither figure is now usable as a level, and that the week-ahead entry which looked like a schedule was a note all along — a number with no time, no timezone and, as of today, no agreement.
That makes 25 instances in this desk’s running tally of a market number whose field definition, timestamp or label does not travel with it. Yesterday it was a Canadian monthly print that is a miss or a beat depending on the series, and a barrel of oil that rose or fell depending on the contract month. Today it is two scheduling facts and a URL.
What to do about it before the London open
Three rules, and none of them is a market call. Search a dated board by the date in its own title, never by today’s date and never through an undated alias — yesterday’s negative result on a board that existed was our search string, not the board’s absence. Treat a calendar entry with no time and no timezone as a reminder rather than a window, and do not let it gate position size. And where two readers give you a central bank move 50 basis points apart, the honest setting is a wider stop or no position, not an average of the two.
One live item on the board itself. The 159.00 strike carries $1.1bn, 48.1 per cent of the $2.29bn of dollar-yen notional on the page, and nothing sits above it. Yesterday we pre-registered that 159.00 would not trade before the cut. Through the Tokyo close it has not: one Tokyo reader had spot at about 158.40 at 02:54 UTC, 60 pips below the strike, inside this month’s range and without testing the top of it, and the overnight high of 158.55 leaves 45 pips of air. Another Tokyo desk puts today’s range at 157.800 to 159.200 and notes that 159 was last seen on 24 September, 26 days ago. The claim is live, not scored. We score it at 14:00 UTC, which is our convention for the cut; the board itself gives no timezone, which is its own small instance of the problem above.
What this does not tell you
We still cannot say whether this house publishes its board a day early or whether the 07:51 stamp disagrees with the page’s own title, and the unexplained 20-minute gap between the feed time and the publication time means the page carries two timestamps we cannot reconcile from outside. The negative on a 21 October board rests on search, and search does not index pages that are minutes old, so it is weak evidence even where the logic above did not already make it useless.
On Indonesia: the 5.75 per cent starting point is our arithmetic from one reader’s stated destination, not a policy rate we read anywhere, and we are labelling it an inference rather than a fact. The 75 basis point figure remains one reader and the 25 remains one reader. We also read a Tokyo order-book characterisation this morning and are not publishing it, because this desk retired that source for dollar-yen and a retirement does not lapse because the material looks useful.
