Core CPI Printed 0.3%, Exactly Like August and September. That Is What Fired the Half-Point Condition.

The score, and the two fields that both missed by one tenth

The Bureau of Labor Statistics release, read in full on the Bureau’s own page, carries the embargo line “embargoed until 8:30 a.m. (ET) Tuesday, November 10, 2026” — 13:30 UTC, exactly as five readers told us yesterday it would be.

• Headline CPI, seasonally adjusted: 0.4 per cent month on month, against a 0.3 per cent consensus attributed to the Dow Jones survey. Miss: one tenth high.
• Core CPI, all items less food and energy: 0.3 per cent, against a 0.2 per cent consensus. Miss: one tenth high.
• September, as printed in this release: headline 0.6 per cent, core 0.3 per cent. So the headline decelerated by two tenths while beating consensus.

Two fields, two misses, both one tenth, both in the same direction. A filter computing actual minus consensus scores this release at plus 0.1 and plus 0.1 and reports a small hot surprise. That filter would be arithmetically correct and would have missed everything that matters today, for a reason we will get to.

On the composition: the headline’s deceleration is an energy story and nothing else. Energy went from 5.2 per cent in September to 0.7 per cent in October, a swing of four and a half points, with gasoline 1.0 per cent seasonally adjusted against 9.1 per cent the month before. Strip that and the picture firms rather than softens. Food went from flat to 0.5 per cent. Medical care went from 0.1 to 0.3, hospital services 0.8, physicians 0.4. Recreation 0.8 per cent. Motor vehicle insurance 0.7 per cent. Against that, shelter decelerated a tenth to 0.3 per cent and used cars fell 1.2 per cent after rising 0.7. Core held its 0.3 with its two largest helpers — energy and shelter — both pulling the other way.

What the governor actually said, and why it changes the trade

We published the threshold as “0.3 per cent or more from September,” sourced to a wire’s rendering, and we flagged in print that the figure was a reporter’s wording and not inside the governor’s quotation marks on any page we had read. That flag was right and we have now closed it at the issuing institution. The Board’s own text reads: if the October data “are similar to the past two monthly readings” — “with core CPI inflation around 0.3 percent or even higher” — “then a stronger policy move may be needed.” And: “In that case, I believe a 50 basis point hike in the policy rate should be considered at the December meeting.” He also said he is “leaning toward supporting an increase in our policy rate of 25 basis points,” and that “hiking only 25 basis points would put our credibility at risk, a risk I am not eager to take.”

Three differences from what we printed, and the third is the one that costs us:

• “Around” is not “or more.” A reading of 0.3 is inside the condition on the Board’s wording. On ours it would have been the bare minimum.
• The reference is the past two months, not September alone. Both were 0.3.
• The primary clause is similarity, not exceedance. We described a trigger that required an upside surprise. It required a continuation. Those are opposite objects, and we built an article on the first one.

So the sentence we wrote — the trigger sits one tenth above consensus — was a statement about the consensus, not about the condition. The condition sat at the last two prints. The consensus was the thing standing one tenth away from reality, and it is the consensus that was wrong. If you were sizing for a threshold breach you were sizing for the wrong event: the half-point branch did not need a hot number, it needed an unremarkable one.

The speech also settles its own sequence. The Board’s text says the October CPI arrives “in a few hours” and that he will look at it “later today,” so the condition was stated before the data existed. A publisher’s item carrying the remarks is stamped 13:00 UTC, half an hour ahead of the release. That is a central banker pre-committing a reaction function in public with the clock running, which is rare enough to be worth more than the print.

We also published a zero. It was 1.4 per cent, on the same feed, before the print

At today’s pre-US slot we published that the CME FedWatch tool assigned “nothing at all” to the half-point branch, with 84.4 per cent on a quarter point. We labelled that figure carefully as one wire read twice, because both pages naming the feed traced to the same origin, and we declined to call the named-feed question closed. The caution was correct and insufficient. The number was wrong.

So the half-point branch went from about 1.4 to about 11 per cent, a factor of roughly eight, on a print that did nothing but repeat itself. That is the honest version of the story we told, and it is a better one: the repricing is real, the base was never zero, and a position built on “the market gives this no chance” was built on a figure that did not exist.

The quarter-point readings are worse, and they are worse on one page. The same publisher, naming CME FedWatch both times, carries 100 per cent for a December hike at 13:00 UTC and 85 per cent — “up from 77 per cent Friday” — at 16:30 UTC, after a print that made a hike more likely rather than less. Add our own 84.4 per cent from Monday and a Tokyo note’s high 80s and the four readings of one named feed span 15.6 percentage points. We are not going to adjudicate that. We are going to say what it means operationally: these numbers are not a time series. Differencing two of them to measure a repricing produces a figure whose sign depends on which pages you happened to open, and we have now done that badly ourselves twice in three days.

The missing year-on-year field held, and the Bureau appears to contradict itself inside one table

On Monday this desk published that October’s CPI would carry no 12-month change for the headline index, because October 2025 was never collected during the 2025 lapse in appropriations, and that one calendar was carrying a 3.9 per cent consensus for a field that would not print. That held exactly. The release has a section headed “October 2026 12-Month Percent Changes” which states that the Bureau “cannot publish 12-month percent changes for October 2026 for indexes where October 2025 data are unavailable,” and in Table A the 12-month column shows a dash for all items, food, energy, core and shelter alike.

And then the same column prints 37.4 per cent for gasoline, 0.7 for new vehicles and minus 3.3 for used cars and trucks.

We are flagging this rather than explaining it. The plain reading of the Bureau’s own sentence is that where October 2025 data exist the annual change can be computed, so a partially collected basket would produce exactly this pattern — dashes for the aggregates, numbers for some components. That is a hypothesis. We have not opened the detailed table that would confirm it, and we are not going to assert it from the summary page. What we will assert is the operational consequence, because it does not depend on which explanation is right: a system reading this release field by field will find an annual rate present for gasoline and absent for the index gasoline sits inside. If your inflation monitor fills a missing aggregate from its components, it is about to synthesise an annual core rate out of a basket that is admittedly incomplete, and nothing on the page will stop it.

One field we were asked to check and cannot: whether September was revised in this release. The release carries no revision language and prints September at 0.6 headline and 0.3 core, which is what we already had. Absence of revision language on a summary page is not proof that no revision occurred, and we are not claiming it is.

What this does not tell you

• We cannot tell you whether the condition is actually met in the governor’s own judgement. “Around 0.3 percent or even higher” is satisfied by 0.3 on any reading we can construct, and he named the past two months, which were both 0.3. But “similar to the past two monthly readings” is a judgement he reserves, not an arithmetic test we can run. He also said he would “still support” 25 basis points and is “leaning toward” it, in the same speech. Both sentences are in the Board’s text. We are not going to pretend the conditional resolves itself.
• We have no reading of the CME FedWatch tool itself. Every probability above is a publisher’s report of it, and four of those reports of one feed span more than fifteen points. After eleven slots of asking, a genuinely independent second reader of that feed remains unfound, and today it got worse rather than better.
• We cannot give you a 12-month core rate, because none exists. The last official annual figures are September’s, 3.7 headline and 2.6 core, and one publisher’s three-month annualised core of about 3.7 per cent is that publisher’s arithmetic, not the Bureau’s.
• We did not read the detailed CPI tables, only the summary release, so the component annual rates above are reported as the summary page prints them and are not cross-checked against the table they come from.
• No equity close in levels. We have percentage moves at 21:01 UTC attributed to one network through one aggregator — the S and P 500 about 0.4 per cent lower, the Nasdaq Composite about 0.7 per cent, the Dow about 0.3 per cent — and no index levels we are willing to print.

Related

• FX events calendar
• Signals
• EA presets