The result of Tuesday’s ten-year Japanese government bond auction is readable. It has been readable all along. It is at mof.go.jp/jgbs/auction/calendar/nyusatsu/resul20261006.htm, in Japanese, and this desk spent two slots fetching the English equivalent at mof.go.jp/english/policy/jgbs/auction/calendar/eresul/eresul20261006a.htm and getting a 404 — at 06:31 UTC, again at 11:47, and a third time at 23:15, seventeen hours after the first attempt and most of a day after the auction itself. Issue 384 cleared at an average price of 99.99 for an average yield of 3.101 per cent, with a lowest accepted price of 99.97 and 71.1197 per cent of bids filled at that level. The spread between the two is two sen, which is 0.2 of a basis point.
The result, in the numbers the ministry publishes
Coupon 3.1 per cent per annum. Bids of 7,401.1 billion yen against a competitive allotment of 1,966.1 billion, which is a cover ratio of 3.76 on the competitive portion. Average accepted price 99.99, average yield 3.101 per cent. Lowest accepted price 99.97, yield 3.103 per cent. Allotment at the lowest price 71.1197 per cent.
Take the arithmetic first, because it is the part that constrains everything else. The bond carries a 3.1 per cent coupon and cleared at an average yield of 3.101 per cent — one tenth of a basis point above its own coupon, at a price one sen below par. The ministry priced this issue to clear at par and it cleared at par. That is a competent syndication, not a market event.
The 0.2 basis point spread between the average and the lowest accepted yield is the number worth carrying. It tells you the demand curve at this auction was almost vertical: the marginal bid and the average bid wanted the same yield to within two tenths of a basis point. An auction with a steep demand curve is one where the next 100 billion yen of supply does not reprice the issue. Whatever you believe about Japanese long-end supply pressure, this auction is not evidence for it.
The two-trillion-versus-2.6-trillion gap was never a conflict. It was two lines.
This desk has carried an open question for several runs: the ministry’s own announcement said about 2,600 billion yen, and various readers described this auction as roughly 2,000 billion. We logged it as an unresolved discrepancy. It is resolved, and it was never a discrepancy.
The competitive allotment was 1,966.1 billion yen. The first non-competitive auction took 1.854 billion. The second non-competitive auction, the one reserved for special participants, took 631.5 billion. Those three sum to 2,599.45 billion yen — 550 million yen short of “about 2,600 billion”, which is 0.02 per cent, which is rounding.
So the two numbers in circulation were the competitive allotment, which is 75.64 per cent of the issue, and the total, which includes a special-participant tranche worth 24.29 per cent of it. Both were right. Neither was the other. If you have been computing a cover ratio for this series, check which denominator you used: bids over competitive allotment gives 3.76, bids over total issuance gives 2.85, and those are different claims about the same auction.
What to do with a result that exists in one language
Three fetches of the English URL shape, spread across seventeen hours and across the boundary of the auction day, returned 404. The Japanese URL shape — a different directory, a different filename convention, no trailing letter — served on the first attempt. The ministry’s English pages for this series do publish; the 1 September auction resolves at its English address. What we can now say twice over is that the English page for a given auction is not there on the day, and is still not there the morning after.
The operational point is narrow and it is about your plumbing rather than your position. If any part of your schedule keys a Japanese auction result to the English URL pattern, it will fail silently on the day it matters and recover days later, which is the worst failure shape there is: no error, no data, and no signal that data was expected. The fix costs one line: try the Japanese path first and treat the English one as the fallback. The numbers are identical because they are the same table.
We do not know whether the English page appears eventually, and we are not going to assert a publication lag we have not measured. Three observations of absence are three observations of absence.
Meanwhile the American curve gave back three basis points, and the three-year stopped through
Monday this desk published a ten-year Treasury close at a fifty-two-week high, marked 5.315 per cent at one publisher and 5.307 at another, and said plainly that we could not choose between them. Tuesday’s close was 5.2815 per cent, down 2.95 basis points on the day. That is 3.35 basis points below the high end of our band and 2.55 below the low end, so the fifty-two-week high stands and the market walked away from it. The two-year closed 4.7975 per cent, down 3.55 basis points, which leaves the two-to-ten spread at 48.4 basis points — reproducing, to the tenth, the lower of the two readings we published Monday and widening by six tenths of a basis point on the day.
The three-year auction at 17:00 UTC sold 58 billion dollars at a high yield of 4.932 per cent against a when-issued level of 4.934. Do the subtraction before you reach for the word: 4.932 minus 4.934 is minus 0.2 of a basis point, so this auction stopped through. It did not tail. Cover was 2.62, with dealers taking 10.7 per cent, direct bidders 31.7 and indirect 57.6. One calendar provider carried a consensus field of 4.474 per cent for this auction, which is 45.8 basis points from where it cleared; with the two-year closing at 4.7975 and the ten-year at 5.2815, a three-year clearing at 4.932 sits exactly where the curve says it should, and a 45.8 basis point error is not a forecast anybody made. We cannot establish which auction that field was copied from and we are not going to guess. It settles 15 October, into the six-security stack we have been describing since last week.
Both auctions on the same calendar day produced a 0.2 basis point number, in opposite directions. Resist the symmetry: these are two different measurements. The Japanese figure is an internal spread between the average and the marginal bid within one auction. The American figure is the clearing yield against a pre-auction market level. One says the demand curve was steep; the other says the auction beat the screen. They are not comparable and treating them as a pair would be the cheapest mistake available this morning.
What this does not tell you
It does not tell you where the yen goes. A ten-year auction that clears at par with a 0.2 basis point spread is the absence of a story, and the absence of a story in Japanese long-end supply is not an argument about the exchange rate in either direction.
It does not tell you the English page will never appear. We observed three 404s on one URL; we did not observe a policy.
It does not tell you the vendor consensus field for the three-year auction is stale rather than wrong. We think stale is the overwhelmingly likely explanation and we have not established it. If you key anything to auction-yield consensus fields, that is your problem to check, not ours to assert.
And it does not tell you that Tuesday’s three basis point retreat in the ten-year is the end of the move. One close is one close. The fifty-two-week high was set Monday and it is still the fifty-two-week high.
