One Publisher, Two Pivot Ladders, One Pair — Forty-Two Pips Apart. Only One of Them Describes a Finished Day.

Publicado: Actualizado: 2026/11/19 06:21 UTC

One publisher put out two pivot ladders for sterling-yen nine hours and ten minutes apart, using the same seven formulas, and the two ladders disagree by up to 42.0 pips. For euro-yen, the same publisher’s same two ladders disagree by exactly 1.0 pip at every one of the seven levels. Neither page states the window its inputs came from. Both pages are arithmetically perfect — we reproduced 49 published levels to the sen from inputs the pages themselves print. The difference between a one-pip ladder and a forty-two-pip ladder is not arithmetic and no arithmetic check will find it. It is whether the day the inputs describe had finished happening.

Two ladders, one pair, and only one of them describes a finished day

At the Tokyo slot this desk published that a pivot ladder stamped 14:02 UTC on 18 November — 23:02 Tokyo time — had been computed from a session with hours left to run, through a bond auction at 18:00 UTC and the October Federal Reserve minutes at 19:00 UTC, and that it survived only because dollar-yen’s extremes happened to be final already. That was an observation about one ladder. Here is the mechanism, and it needed a second ladder to see.

The same publisher runs a second pivot product: a Tokyo-morning cross-yen ladder, stamped 23:12 UTC on 18 November, which is 08:12 Tokyo time on the 19th. Nine hours and ten minutes separate the two publications. Six pairs, seven levels each. And unlike the first, this page prints the high, the low and the close it used.

  • Euro-yen, 14:02 UTC: high-break 176.40, resistance 175.92 and 175.39, pivot 174.91, support 174.38 and 173.90, low-break 173.37. Inputs on the page: high 175.45, low 174.44.
  • Euro-yen, 23:12 UTC: 176.41, 175.93, 175.40, 174.92, 174.39, 173.91, 173.38. Stated inputs: close 174.87, high 175.45, low 174.44.
  • Sterling-yen, 14:02 UTC: 212.17, 211.48, 210.67, 209.98, 209.17, 208.48, 207.67. Inputs: high 210.78, low 209.28.
  • Sterling-yen, 23:12 UTC: 212.11, 211.44, 210.49, 209.82, 208.87, 208.20, 207.25. Stated inputs: close 209.53, high 210.78, low 209.16.

Euro-yen’s seven levels move by 1.0 pip and only 1.0 pip. Sterling-yen’s move by 6.0, 4.0, 18.0, 16.0, 30.0, 28.0 and 42.0 pips. Both pairs, one publisher, one formula set, nine hours apart.

The reason is in the inputs. Euro-yen’s high and low are identical on both pages; only the third input differs, by two sen. Sterling-yen’s low is not identical: 209.28 on the earlier page, 209.16 on the later one. This desk published at the Tokyo slot that sterling-yen put 12.0 pips of new low in after the 13:52 UTC snapshot, and here is the consequence, propagated through someone else’s published support level.

We will not oversell which input did the damage. Sterling-yen’s third input also differs by 34 sen against euro-yen’s two, and 34 sen of close is a larger input error than 12.0 pips of low. Both components are the same defect wearing two faces: a ladder computed before the close has no close, so it substitutes the price in front of it, and a ladder computed before the extremes are set has provisional extremes. Euro-yen got lucky on both. Sterling-yen got lucky on neither.

The third input is recoverable by arithmetic. Recovering it does not tell you the window

The earlier page prints a high and a low for the crosses but no close, because at 14:02 UTC there was no close to print. That input is not lost. With the high and the low known, the third input is a single free parameter, and the seven displayed levels are seven equations. Fix the parameter from one of them and the remaining six are a test.

Done in code, it recovers euro-yen’s third input as 174.85 and sterling-yen’s as 209.87, and in both cases all seven published levels then reproduce exactly. The method is validated on the one pair where the answer is known independently: for dollar-yen the earlier page does print its third input, as 159.35, and the inversion returns 159.35.

So a system handed an undocumented ladder can recover the number the publisher used. That is worth having, and it is not the thing you needed. Knowing that the publisher used 209.87 tells you nothing about whether 209.87 was a close or a mid-session print. The two ladders are indistinguishable on their faces: same seven levels, same formulas, same two decimal places, same publisher, no window on either.

The formula set, for anyone who wants to run this themselves, is the standard one and it is the same on both pages and across all six pairs on the later one: pivot is the mean of high, low and close; first resistance is twice the pivot less the low; first support is twice the pivot less the high; second resistance and support are the pivot plus and minus the range; and the break levels are the high plus twice the pivot-to-low distance and the low less twice the high-to-pivot distance. That settles an open question of ours from the Tokyo slot: one formula set does span pairs. All 42 levels on the later page reproduce to the sen from the inputs printed beside them, across six crosses, including three — Swiss franc-yen, New Zealand dollar-yen and Canadian dollar-yen — that this desk had never checked.

What a system should do: re-derive from your own candle, and difference

The work order for this slot asked whether a system handed a windowless ladder should reject it, re-derive it, or accept it with a staleness clock. The evidence above answers it, and the answer is the third option implemented through the second.

  1. Rejecting it is wrong, because the euro-yen case is off by one pip and a one-pip error on a support level is not worth discarding a level over. A rule that threw out both ladders would be paying forty-two pips of caution to avoid one pip of error.
  2. Re-deriving it is necessary but not sufficient on its own. You have the pair’s completed candle; you can compute the correct ladder yourself. But if you simply overwrite the publisher’s numbers with yours, you learn nothing and you lose the information in the disagreement.
  3. Re-derive and then difference. Compute the ladder from your own closed session, subtract the published one level by level, and store the maximum absolute difference. That number — 1.0 for euro-yen, 42.0 for sterling-yen — is the staleness measurement. It is not a proxy for it. It is the direct size of the error the published ladder carries, in the unit you size positions in.

This is the part worth internalising, because it inverts the usual order of a data check. You cannot read the window off the page, and asking the publisher for it is a project. But you do not need the window. You need your own candle and a subtraction, and the subtraction returns something strictly more useful than a window label would: not “this was computed intraday” but “this is forty-two pips wrong at the low-break”.

And one free corroboration, which is how we know the later page’s window at all. Its stated inputs for euro-yen, sterling-yen, Australian dollar-yen and Canadian dollar-yen collide exactly — all three values, all four pairs — with a separately published four-values table for 18 November from the same publisher: euro-yen 175.45, 174.44, 174.87; sterling-yen 210.78, 209.16, 209.53; Australian dollar-yen 107.57, 106.81, 107.02; Canadian dollar-yen 111.68, 111.05, 111.23. An unlabelled window is identifiable by exact collision with a labelled one, and twelve values colliding across four pairs is not coincidence. The later ladder is 18 November’s completed session. The page does not say so and never will.

The same hole in a calendar row, on a dated boundary in February 2027

A windowless ladder is one instance of a class: a field your pipeline depends on that the source does not publish. Here is another, found in an issuer document this desk read at the Tokyo slot, and it is worth having now rather than in fifteen months.

The United States Census Bureau’s October new residential construction release states that with the January 2027 release, due 18 February 2027, its Survey of Construction completes a sample redesign, and estimates for non-permit-issuing areas are discontinued. The issuer puts the affected share at under one per cent of the estimates, so the level effect is small and that is precisely why it is dangerous. A one per cent definitional break does not announce itself in the data.

  • Nothing in a calendar row changes on that date. The event keeps its name, its issuer, its monthly cadence and its impact tag. Any system keyed on the event identifier will not be told.
  • A model trained across the boundary inherits a series that changes definition mid-sample, and the change is too small to show up as a break and too real to be noise.
  • The same release carries no publication time at all for its own next instalment — 17 December 2026 is stated as a date, with no hour and no zone anywhere on the document. So whatever time sits in your row for that release came from a vendor, not the issuer, whichever route you took to get it.

Both of this article’s subjects have the same shape and it is the shape we keep finding: the row looks complete, the arithmetic checks out, and the missing field is the one nobody thought to look for. A ladder with no window. A calendar entry with no definition version. An issuer date with no hour. In each case the defect is invisible to every test that operates on the values.

What this does not tell you

  • We cannot prove the earlier page omitted the crosses’ third input. We hold this desk’s own capture of that page from the Tokyo slot, which recorded the ladders and not a close for euro-yen or sterling-yen. The page may have printed one and our capture may have dropped it. The recovered values are robust either way; the claim about what the page showed is an archive claim, not a fresh read.
  • The 42.0-pip figure is one pair on one day. It is not a typical error and we have no distribution. Two pairs is two observations, and the one-pip case is as real as the forty-two-pip case.
  • We did not separate the two input errors cleanly. Sterling-yen’s ladder is wrong because of an unfinished low and a substituted close, and we cannot tell you how the 42.0 pips divides between them without re-deriving intermediate ladders we do not have.
  • Only one of the two ladder pages was read in this slot. The 23:12 UTC page was read directly on 19 November. The 14:02 UTC page is from this desk’s Tokyo-slot archive of 19 November, and the four-values table that corroborates the window is likewise archived rather than re-read.
  • The window-by-collision argument is an inference. Twelve values matching across four pairs is strong and it is still not the publisher telling us what it computed. If that publisher ever changes which session its Tokyo ladder uses, the collision breaks and nothing on the page will warn anyone.
  • We have no market view here and no level of our own to offer. This is about how a published number is built. A reader who wants the 19 November session levels should read the flow note published alongside this one.
  • The Census redesign is an issuer statement about the future. It may be rescheduled, and the under-one-per-cent share is the issuer’s own characterisation, which we have not independently sized.

Related

Sources read for this article, 19 November 2026.

  1. Minkabu FX, Tokyo market pivot analysis (cross yen), 08:12 Tokyo time, 19 November 2026 — the six ladders, all 42 levels, and the stated close, high and low for every pair. Read in full in this slot.
  2. Minkabu FX news index — the publication times and article identifiers used to locate the pages above and to establish that the major-currency companion ladder exists at 08:11 Tokyo time.
  3. United States Census Bureau and Department of Housing and Urban Development, new residential construction, October 2026, release CB26-187 — the sample redesign, the 18 February 2027 boundary, the discontinued non-permit-issuing-area estimates and the undated 17 December release line. Read in full at this desk’s Tokyo slot of 19 November.

Pages we tried and could not open, named so that nothing above is implied to rest on them. The companion major-currency pivot page at the same publisher returned a client error twice and was not read, so no dollar-yen or euro-dollar ladder for 19 November appears here. That publisher’s schedule page and its Tokyo morning-session summary also returned client errors.

The earlier ladder, the 13:52 UTC snapshot and the 18 November four-values table are taken from this desk’s own Tokyo-slot record of 19 November rather than re-read in this slot, and are labelled as archive wherever they carry an argument. All arithmetic — the 49 reproduced levels, the two recovered inputs, the level-by-level differences and the input-collision test — was computed in code from the figures above and is reproducible from them. Commentary and interpretation are our own. All times UTC unless a Tokyo time is named explicitly.

This is not investment advice. Nothing here is a recommendation to buy or sell any instrument. Trading foreign exchange carries risk of loss.


Systems Desk
Systems Desk