Tokyo Core Printed 2.6 and 2.4. The Issuer Publishes Both, and the Word Is Doing Four Jobs.

What the issuer actually printed

From the Tokyo Ku-area preliminary release for October 2026, published 30 October, index base 2025 = 100. October figure first, then the September figure as the issuer states it:

• Headline, total index: 103.5, up 2.8 per cent year-on-year, from 2.7 per cent. Month-on-month seasonally adjusted +0.5, from +0.6.
• Excluding fresh food — the series most desks mean by core: index 103.3, up 2.6 per cent, from 2.7. Month-on-month +0.3, from +0.9.
• Excluding fresh food and energy: index 103.6, up 2.6 per cent, from 2.9. Month-on-month +0.2, from +0.7. Weight 9162.
• Excluding food other than alcohol, and energy: index 103.0, up 2.4 per cent, from 2.7. Month-on-month +0.6. Weight 6959.

There are the two disputed numbers, eight lines apart in one PDF. The reader quoting 2.6 with a 2.9 prior was reading the 9162-weight basket. The reader quoting 2.4 with a 2.7 prior was reading the 6959-weight basket. Neither was wrong, neither mislabelled anything, and neither was reading the other’s series. They were reading two different exclusion baskets, and the gap between them — twenty basis points on the level, twenty on the prior — is the price of excluding processed food and alcohol as well as fresh food and energy.

We should be honest about one step: the issuer’s text names the 9162 basket explicitly and we matched the 6959 row by its weight, which is unique in the document and appears beside that label. Call the attribution of the 2.4 to that specific basket an inference from the weight rather than a printed statement.

Both were right about the month, too

This is the part that would have looked most like an error. One publisher ran a headline saying Tokyo core ticked higher in October, on a month-on-month figure of +0.6 against +0.4 in September. The other gave month-on-month core-core of +0.2 against +0.7. One of those says underlying prices accelerated and the other says they decelerated sharply, about the same month, in the same city.

The issuer gives +0.2 for the 9162 basket and +0.6 for the 6959 basket. Both month-on-month figures are in the document, attached to the two baskets their publishers were already reading consistently. The September comparison is where it stops being clean: the issuer’s September figure for the 9162 basket is +0.7, which matches the second reader exactly, and we could not find a September month-on-month figure for the 6959 basket in the document at all. So the +0.4 is the one half of this that remains single-sourced, and the acceleration headline rests on it.

In September this desk published a split verdict when two vendors on Canadian inflation turned out to be each right about a different field, and called the question “which vendor has the right consensus” malformed. Today is the harder version of the same lesson. Nobody was half-right. Both were entirely right, and the malformed question was ours: there is no fact of the matter about what Tokyo core did in October until you say which basket you mean, and the one-word field in your database cannot hold the answer.

The consensus field is where it genuinely broke

The real disagreement is not in the prints. It is in what the prints were measured against. On the headline, which the issuer puts at 2.8 per cent:

• One reader gives an expected 2.7, making 2.8 a beat of one tenth, and that is the reading its write-up takes.
• One reader gives an expected 3.3, making 2.8 a miss of half a point, and its own commentary calls it a clear miss versus consensus.

Same print, one tenth above consensus or fifty basis points below it, depending on which page you had open. A fifty-basis-point gap in a consensus field for a release that has existed for decades is not a difference of opinion among forecasters. One of those two numbers is not a forecast.

Here is a reading, offered as one desk’s inference and not as a finding: the 2.7 expected figure is identical to the September headline print that the issuer confirms at 2.7, and the same vendor’s table carries 2.7 in both the September column and the expectation column. A prior sitting in a forecast field is a failure mode we have logged repeatedly on this channel, and this has its signature. But a consensus genuinely can equal the prior, which is why this is an inference. What we can say without inference is the practical rule: a release that prints within a tenth of its prior and within a tenth of one vendor’s consensus is not a tradeable surprise under either reading, and the 3.3 would have told you it was a large one.

Scoring the forecasts we published seven hours ago

Last night, before the print, we published the forecast set we had: one desk’s medians of 2.7 headline, 2.7 core and 2.8 core-core, with ranges of 2.4–2.9, 2.4–2.8 and 2.5–3.0, alongside an aggregator’s single 2.6 consensus and a house forecast of 2.5. We also noted that the forecasts contained each other, which was the reason to think the dispute was mild. Marked against the issuer:

• Headline: printed 2.8, median said 2.7. Missed by one tenth.
• Core, ex fresh food: printed 2.6, median said 2.7. Missed by one tenth. The aggregator’s lone 2.6 was exact.
• Core-core: printed 2.6 on the 9162 basket, median said 2.8. Missed by two tenths, and the printed figure sat at the very bottom of the 2.5–3.0 range.
• The house 2.5 missed core by one tenth in the other direction.

So a three-number median set with published ranges was beaten, on the series most people trade, by one aggregator’s single figure. That is now the fourth consecutive occasion on which this desk has resolved a consensus dispute and then watched the resolved consensus lose to the print or to a simpler number. The standing instruction we wrote for ourselves after the third — prefer a house’s conditional threshold to anyone’s point estimate — held again, in the weak sense that nobody published one today and so nobody was wrong in an interesting way.

The headline rose because of vegetables

The one thing worth knowing about the composition, and it is the issuer’s own arithmetic rather than anyone’s interpretation: the headline accelerated from 2.7 to 2.8 while every underlying measure decelerated. Fresh food, weight 379, went from +1.7 per cent year-on-year to +8.6, lifting its contribution from 0.06 of a point to 0.32, which the issuer says widened the headline gap by 0.26 of a point. Energy, weight 460, went from +2.2 per cent to −1.9, its contribution from +0.09 to −0.08.

A central bank that moved on this morning’s headline would be moving on vegetable prices against a falling energy bill. It did not: the Bank of Japan held at 1.25 per cent a few hours later and raised its fiscal 2027 inflation path, which is covered in today’s other note. The transmission you care about runs through the 2.6 and the 2.4, both of which fell.

One live catch for your scraper while we are here. A well-regarded aggregator carries a page headed “Tokyo Core Inflation Tops Expectations in October” which gives core at 2.8 per cent against a forecast of 2.6. It is October 2025. Its figures are internally consistent and its framing is the opposite of this year’s, and the 2.8 is numerically identical to this year’s headline, so a date-blind fetch lands on a plausible, wrong, confidently-worded page. Check the year on everything you reach through search, including on sources that are usually right.

What this does not tell you

It does not tell you which basket the Bank of Japan weights most heavily in its own judgement, and the Bank’s own documents were not reachable for this note. Our reading of the 2 per cent target language published this morning is secondary.

It does not settle the +0.4 September month-on-month figure behind the acceleration headline, which we could not find in the issuer’s release and which therefore stays single-sourced. It does not give you a seasonally-adjusted flag on every month-on-month figure here: the issuer marks the ones in its summary text and its seasonal-adjustment table as adjusted, and does not say for every detailed table.

Our figures come from a text extraction of the issuer’s PDF rather than from the rendered document, which is a real if small risk on a table-heavy release — check any single figure you are about to size against. And the broader claim this article does not make: that either publisher is unreliable. The opposite happened today. Both read the issuer correctly and the incompatibility was created downstream, by an English word that does not distinguish four baskets, and by us for assuming a single value existed.

Related

• Economic calendar and event windows — German flash Q3 GDP at 09:00 UTC, euro-area flash Q3 GDP at 10:00, German flash October CPI at 13:00, US employment cost index and Canadian GDP at 12:30, Chicago PMI at 13:45.
• Signals
• EA presets