Regime
research.

The committee raised rates on 16 September without a single dissent, deleted the energy language from its statement and put one more rise in the dots. The desk registered four tests on that decision and pre-committed in print to scoring them on the following Thursday's settle. It did not. This note scores them seven sessions late, says so in the first row of its own table, and then scores a level and a claim the desk had to defend and did not hold: a thirty-year range that broke on 24 September and has widened since, and a buyback queue the desk is retiring rather than arguing with. What survived is the mechanism, and it is the part worth reading. Measured from the close before the August inflation print, the ten-year is twenty-nine basis points higher and the market's own inflation forecast is six lower. The whole move is real, and then some.

On 18 August this desk wrote that late July's seven hawkish dissents across three central banks were the high-water mark of rate-rise pricing, and named the sequence that would make that mark a tide line: the shock returns, the August inflation print lands hot into the September decision and the rise comes back.

The first enlarged buyback was disappointing before it ran. Treasury capped Thursday's ten to twenty year operation at $6bn, above the $4bn floor it had guided and below the $10bn some had come to expect, and the curve sold the number into a $100 barrel: the ten-year to its highest in nearly three years, the thirty-year into the upper half of the range the desk redrew in August, the two-year backing up toward the Fed's own week. On Thursday the desk got both hands on one tape, a $22bn thirty-year auction at 18:00 London and the operation itself an hour later. The read was written down before the prints and is scored on the settles below.

Payrolls printed 162,000 against a consensus near 56,000, and the revision mattered more: July's minus 23,000, the hinge of the desk's registered growth test, was revised to plus 21,000. The first negative print no longer exists, so the second one can no longer be the test. The desk's internal pre-print branch map carried this outcome at fifteen percent, and the desk scores that first, before the argument. What survives the print: the wage line stayed quiet, the services survey still says contraction, and everything now compresses onto Friday's CPI, the reweight the desk pre-committed a month ago.

The strait went from warning to exchange and Monday's settles decided: crude up two dollars and change on both benchmarks, the thirty-year up close to four basis points, above its 5.20 pivot, the two-year flat, gold flat, credit quiet and the desk's refining margin holding just under the record it set on Friday. On Friday the committee wrote the front end while the long end stood still. On Monday the shock wrote the long end while the committee waited. The desk scores its registered discriminators here.

The chair's first Jackson Hole address names the two conditions the desk registered as its risk branch: inflation progress he cannot call meaningfully improved, financial conditions he is hard pressed to call restrictive. The front end reprices September from roughly a third to a coin flip. The thirty-year moves one and a half basis points. Five tests were published before the events; the desk scores them here, the miss first.

On Friday morning the new Fed chair gives his first Jackson Hole address, and at the same hour the statisticians publish the preliminary payrolls benchmark revision, an exercise whose previous preliminary estimate subtracted 911,000 jobs. Between now and then: the Fed's preferred inflation gauge, a Bank of Korea decision that could deliver the global tightening cycle's newest hike, and a Treasury reported, twice in one Monday session, to be willing to do whatever it takes to defend its own long end. Three hands now reach for the same pen: a committee that declines to write, a market that has been writing all month, and an issuer that wants to steady the hand. The desk's practice before weeks like this is to publish its tests first and score them after. Here are the tests.

On Wednesday the US Treasury said it will at least double its buyback operations in the two longest nominal sectors, citing consistent strong sponsorship rather than strain. The verdict took four minutes: the dollar fell, global bonds rallied and gold rose a per cent. Then the twenty-year auction tailed anyway. The operation is worth basis points and the announcement is worth more, because all year the market has offered this door around ten dollars of long bonds for every one accepted, and Tuesday, the quietest such day of 2026, was no exception. The issuer widened the exit the morning after the shortest queue of the year, in a week that had already carried thirty-year yields to their highest since 2007.

On 3 August this desk pre-committed: if a September rise slid below half and stayed there, the seven hawkish dissents were the high-water mark, and the desk would write that note.

Inside seventy-two hours the Federal Reserve held nine to three, the Bank of England held six to three and the Bank of Japan held eight to one. Every dissent on every committee was a vote for higher rates; not one member, anywhere, voted for a cut. By Friday morning a September rise in the US was priced near two in three, above where the decision left it, through a growth miss and a soft inflation month. The desk scores the week it published in advance, its own error included.

The committee left rates at 3.50 to 3.75 on a nine-to-three vote, with three members dissenting in favour of a rise, and the statement named energy supply shocks as a driver of elevated inflation. The market read it the desk's way: by Wednesday evening a September rise was majority-priced, above anything ever assigned to July. A hold that makes the next rise more likely is not a pause. It is the ratchet, confirmed in the committee's own record.

Crude round-tripped ten dollars in four sessions and the odds of a rate rise gave back only a fraction of their tripling.