Skip to content
A Long Game Project lab
Prussian-blue graded 1857 etching of the Bastille lock on the canal Saint-Martin, two keepers at the gate, with a water mark rising across the plate from left to right

The Treasury’s Tell

Is the US Treasury pushing long-term interest rates down? Nobody has said it may. Who makes it admit the aim, and what does admitting it cost?

Report 005 · Dr Dan Epstein · version 1.5 · 19 October 2026
A Strategy Soup Scenario What is an exercised scenario?

The question

The US Treasury is the government’s finance department. It borrows what the government spends by selling bonds, and the interest a bond pays is its yield. In August 2026 the Treasury doubled the size of its buybacks, where it buys back long-term bonds it sold earlier. Many people suspect the real aim is to push long-term interest rates down.

A bond’s price and its yield move in opposite directions, so holding a yield down means holding a price up. The Fed, the US central bank, sets short-term rates, and its officials agree the mix of short-dated and long-dated debt is the Treasury’s business. Nobody has said whether the Treasury may use that mix to push long-term rates down.

On 4 November the Treasury publishes its quarterly refunding statement: how much debt it will sell over the next three months, and why. If it sells fewer long-term bonds, it must give a reason. So who, if anyone, forces the Treasury to admit its aim before 4 November, and what does the reason it gives cost it?

Our answer, after 45 runs: nobody forces the Treasury by argument alone. In the 30 runs with no extra spending, the question came into the open in 4. In the plan where the Treasury spent an extra $40bn, the question came into the open in 10 of its 15 runs, in the turn of the spending or the one after. Every opening followed the spending, but other players moved in the same turns, so the runs do not prove the spending caused it.

Our rules fix three things. Only Plan 3 spends the extra cash. Traders can bet against the Treasury only after that spending, or once the question is open. And we chose every charge to trust, our score for how far the market trusts the Treasury’s plan: 15 points for stating the yield level, or 20 if chance makes the move fail, and 6 or 12 for citing structural demand.

So the order of the reasons by cost comes from our charges. The runs decided the rest: how often speeches and chance opened the question in eight weeks without the cash, how often it opened once the cash was spent, and when.

Real-world facts stop on 3 September 2026. Every later event is the simulation’s story, not a forecast, and some of its dates have now passed.

The claim, 45 runs, five turnsFig. 1

Without extra spending, the question opened in 4 of 30 runs. After the $40bn, it opened in 10 of 15. What each reason costs is a charge we chose.

How openly the aim is discussed, 0 to 100Each line: the middle runopen at 6015no extra spending4 of 30 runs opened$40bn drawn10 of 15 runs openedopened: reached 60before turn 5Trust in the Treasury, 0 to 100Each line: the middle of 12 runs7063 kept wording53 cited demand36 admitted aimstart12345speech$40bnreason
The middle run at the start and at the end of each turn, on two scales of 0 to 100.
LineStartTurn 1Turn 2Turn 3Turn 4Turn 5
How openly the aim is discussed, 0 to 100: no extra spending1515263542
How openly the aim is discussed, 0 to 100: $40bn drawn1515265467
Trust in the Treasury, 0 to 100: kept wording706866646263
Trust in the Treasury, 0 to 100: cited demand706866646253
Trust in the Treasury, 0 to 100: admitted aim706866605436
Scales we wrote, 0 to 100. Cyan: Plan 3, the only plan that drew the extra $40bn. Simulation record.

How the simulation works

Seven roles, each following a written policy. The simulation runs five turns of one to two weeks each, 9 September to 4 November. Each turn every role tries a move, and chance decides whether it works, at odds the simulation sets.

The TreasuryPicks what to sell on 4 November, and the reason it gives.
The Fed BoardThe chair and governors in Washington. They set short-term rates and say nothing about long-term ones until the question is open.
The Fed presidentsThree regional Federal Reserve presidents who want a rate rise.
Hedge funds and dealersTraders in long-term bonds who bet against the Treasury if they think it is defending a price.
Pensions and insurersBig, slow buyers of long-term bonds.
Money funds and stablecoinsHold only very short-term paper. A stablecoin is a digital token meant to be worth one US dollar.
Foreign governmentsMainly Japan, which would rather protect its own currency than sell US bonds.

We gave the Treasury three plans for 4 November.

Plan 1Keep its current wording.
Plan 2Sell fewer long-term bonds and cite “structural demand”, the Treasury’s own published test: a lasting shift in who wants to buy which bonds, which nobody can check inside a quarter.
Plan 3Sell fewer long-term bonds and admit it wants a lower yield. In October it also draws $40bn from its cash account at the Fed, an amount we chose, to make its buybacks bigger than the $4bn each it had announced.

Each plan ran 15 times, and each run had the same chances under all three plans, so the plans compare move for move. We saved every move of three runs per plan, the runs with saved moves. The other twelve, the further runs, saved only their scores.

Every run was then played a second time as a control, with one rule switched off: the rule that lets a role react to the reason another role gives, as well as to the money it moves. With the rule off the question never opens, yet 44 of the 45 runs kept their ending, meaning whether the Treasury cut and the reason it gave.

Three scales of our own, each 0 to 100 and none a measurement:

OpennessHow openly the Treasury’s aim is discussed. Starts at 15, and at 60 the question counts as open.
TrustHow far the market trusts the Treasury to stick to its published plan. Starts at 70.
Chair’s authorityThe Fed chair’s sway over his own committee. Starts at 55.

The Fed and the Treasury agree on who owns the debt mix

On the public record as at 3 September 2026, a former regional Fed president says that once the Fed has set the overnight rate, everything else about the Treasury market belongs to the Treasury. Stephen Miran, a Fed governor, backs a smaller Fed balance sheet, the stock of bonds the Fed owns, because it keeps the debt mix a matter for the Treasury. There is no clash while the Fed’s own bond holdings stay flat, and they were flat in early September.

Scott Bessent, the Treasury Secretary, told a Treasury market conference in November 2025 that his job is to be the nation’s top bond salesman and that Treasury yields are a strong barometer of his success. The day after the August increase, CNBC reported, he said the level of yields did not factor into that decision.

Speeches moved the debate a little and never enough

A Fed president’s speech landed in 24 of the 30 runs of Plans 1 and 2 and moved the openness scale, yet the question came into the open in only 4 of them, and no speech did it alone. The endings came out the same whether it landed or missed. The October dissent count (Fed officials voting against the majority) did not: it ran from 0 to 5 where a speech landed and from 1 to 3 where none did.

The only players willing to say in public that the Treasury is steering prices are the three who want a rate rise, so the question arrives sounding like a bid for something else. In the simulation the Fed’s September vote split its committee whatever the Treasury’s plan, so those dissent counts describe only the simulation.

The question opened mostly after the Treasury’s own spending

In Plan 3 the Treasury drew the $40bn and changed not one word of what it said. We assume its cash account moves that much on an ordinary day, so the draw itself is hard to spot. What the market saw was each buyback, published with its result and bigger than the $4bn announced.

The question then came into the open in 10 of those 15 runs, and no speech did it alone. In 7 of the 10, two or more players moved in the same turn: a routine buyback read as a defence of a price, a Fed vote that went badly, and hedge funds and dealers betting against the level, the yield they believe the Treasury is defending.

Those bets succeeded before 4 November in 11 of the 15 Plan 3 runs and in 2 of the other 30, first in the fourth turn every time. Our rules open the bet only after the cash or once the question is open, so in Plans 1 and 2 it paid only in the two runs where the question opened first.

We also played each plan once as a live check, a run in which every player chose its own moves and gave its reasons. Told to keep its wording or to cite structural demand, the Treasury in the live check still drew on its cash, a move our rules gave only to Plan 3, and the question opened in all three plans. One run per plan shows no spread.

Admitting the aim cost the most, at the charges we chose

In the middle of the 12 further runs, Plan 1 lost 7 points, Plan 2 lost 17 and Plan 3 lost 34. Plan 1 lost 7 even though it changed nothing, because our rules charge each routine buyback a little.

Before any reason was given, Plan 3 had lost 16, against 8 for the other plans. The extra 8 came with the cash draw and the bets against the level that it opened. In the last turn, when the reason was given, Plan 3 lost 18 and Plan 2 lost 9. Part of Plan 3’s 18 came from more bets against the level in that turn, in 7 of its 12 further runs.

The market believed “structural demand” in 10 of the 12 further runs, and trust still fell, because our rules charge that reason 6 points even when it is believed. In two of the others the market named the yield as the real aim anyway, and trust ended at 50 and 48, against 53 where the reason was believed.

A reason nobody can check works only while people believe it.

The Fed chair’s best move, saying in public that long-term rates belong to the market and the debt mix to the Treasury, becomes available only once the question is open. In the two runs with saved moves where he said it, he ended six points of authority above the same run in the control.

What it means

Our reading: an institution scored on a number it only partly controls is judged on the reason it gives for changing course. Three questions follow, our judgement beyond what the simulation tested.

The cheap reasonWhich of your justifications is a “structural demand”: permitted by your own stated policy, impossible to disprove inside a quarter, and a total loss if it stops being believed?
The visible moveWhat is the first move of yours big enough for the market to notice, as distinct from your loudest one?
The answer kept backWho on the other side of the table has an answer they cannot give until you have spoken first?

What this does not prove

In the control, no Fed vote or dissent count changed, and the 30-year yield path in the runs with saved moves matched in 7 of 9. Trust matched in 38 of the 45 runs, because each reason’s cost is our fixed charge on the Treasury’s own move. In the other seven it ended 4 to 8 points lower with the rule on, and the chair’s authority moved in 12 of the 45. So the runs cannot show that the reason changes any decision inside the eight weeks, because the reason is given last.

The openness scale, the speech odds and the $40bn are ours. The starting cash balance, about $950bn, comes from press reporting and was not checked against the Treasury’s daily statement. We chose the small drift in the 30-year yield, so nothing here is scored on the yield, and Plan 3 did not lower it in a single run with saved moves.

The White House, the largest player in the story, enters only through outside events we scripted, and made no move in the eight weeks because it has made none on the public record. If it acts inside the window, we re-run the simulation.

What to watch

Run two, from the record of 17 September, gave 9 of 15 and 2 of 30. That record includes the Fed’s 16 September decision: a quarter-point rise, 12 votes to none, with no dissent. This page describes the first run.

Nine signals to watch, and where each is published.

4 NovemberThe reason in the refunding statement, and whether “at least the next several quarters” survives in it. That phrase is its promise to keep bond sale sizes unchanged, so if it survives there is no cut.
16 Sep, 28 OctThe dissent count at the Fed’s two rate meetings, and whether any governor joins the three presidents.
DailyThe Treasury’s cash account in its daily statement.
Each buybackBuyback results, and any operation above $4bn.
WeeklyThe Fed’s total assets in its weekly release, flat or not.
DailyThe 30-year, 10-year and 2-year yields, as context and never as score.
MonthlyJapan’s holdings in the Treasury’s monthly foreign-holdings report.
Any dayAny on-record remark by a Fed president naming Treasury operations.
UnresolvedThe size of the stablecoin market. We have not yet found a primary tracker.

The record

Every other chart from the 45 runs, in the order the report uses them. The counts are the simulation’s own record. The openness scale is one we wrote.

Every run, by planFig. 2

Without the extra spending the question opened in 4 of 30 runs. With it, in 10 of 15.

Without the extra spending4 of 30 openedWith the extra $40bn10 of 15 openedthe question came into the opena speech raised it, it stayed shutno speech landed, it stayed shut
One dot per run, 45 runs, 15 on each of the Treasury’s three plans. Cyan: the question came into the open before 4 November. Grey: a Fed president raised it in a speech, and it stayed shut. Hollow: no speech landed, and it stayed shut. The simulation’s record.
Openness, the plan that keeps the wordingFig. 3

Across Plan 1’s 15 runs, the middle run rose from 15 to 44. Two runs reached 60 before the refunding turn.

Sep15 OctNovOpenness scale, Plan 1, middle 80% of runsAt the end: middle 44, middle 80% from 27 to 93
Our openness scale, 0 to 100, across the 15 runs of Plan 1, where the Treasury keeps its wording. The line is the middle run, ending at 44. The top of the band reaches 72 by 28 October. The highest single run reached 84 before the refunding turn, past the 60 that counts as open. Plan 2 runs the same until the last turn.
Openness, the plan that spends the $40bnFig. 4

The $40bn lands in the third turn, and the middle run jumps from 26 to 54 inside it.

Sep15 OctNovOpenness scale, Plan 3, middle 80% of runsAt the end: middle 100, middle 80% from 99 to 100
The same scale across the 15 runs of Plan 3, which draws the cash. The middle run reads 15, 26, 54, 67 and 100 across the five turns, against 15, 26, 35, 42 and 44 under Plan 1. The 100 is 4 November itself: admitting the aim puts the question fully in the open by our definition. The speeches were the same in both plans. The money was not.
The chain in Plan 3Fig. 5

In Plan 3 the money came before every opening, and no speech opened the question alone.

How the question came into the openA Fed presidentgives a speechthe openness scalemoves a little, neverto 60The Treasurydraws $40bn ofcashbuybacks lifted abovethe $4bn announcedThe market readsa routinebuyback as adefence of apricethe spending made itvisibleHedge funds anddealers betagainst thelevel11 of 15 Plan 3 runs,open only after thecashThe question isout in the open10 of 15 runs, usuallywith several playersreactingOn 4 Novemberthe Treasurygives its reasonand the market gradesit
The usual order of events in Plan 3. In 6 of its 10 openings the question crossed in the turn of the draw, before any bet. Before 4 November, hedge funds and dealers bet successfully against the level in 11 of the 15 runs, first in the turn after the cash every time. Our rules allow that bet only after the cash or once the question is open, and Plans 1 and 2 reached either in only two runs each.

Want the evidence - the dated record, the sources, and our own list of weak points?

Read deep mode