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 Silence in the Doctrine

Who makes the US Treasury admit it is steering long-term interest rates? A simulated eight weeks, 9 September to 4 November 2026.

Report 005 · Dr Dan Epstein, with Claude · version 1.1 · 9 September 2026
A Strategy Soup Scenario What is an exercised scenario?

What you’re reading

An exercised scenario: one question played out as a game, then reported with the record attached. This is v1.1, 9 September 2026. Seven roles, each a computer program following rules we wrote, played five turns from 9 September to 4 November 2026. The Treasury had three plans for 4 November 2026, and we played 15 games per plan, 45 in all. A control run replayed every game with reactions to reasons switched off. Every number quoted from the games was reproduced from the run record. Dr Dan Epstein set the question and signed the brief. Claude, the AI, drafted the essay from the run record. Dan signs every report before it goes out.

Each marked number cites the evidence pack (55 checked claims) or a public source. This is a game played under rules and written up, not a forecast. How reports get made

Perpetual game

This report is the first run of a perpetual game. A perpetual game is one we play again on a fixed public calendar, and check against what really happened each time. The eight weeks to 4 November are its first cycle.

The claims and their dates

The report makes nine claims about what will and will not happen by 4 November. We wrote them on 4 September 2026. They resolve on 4 November, when the Treasury publishes its refunding statement, the announcement of how much debt it will sell and why. We play the next run after the Fed’s meeting on 15 and 16 September. Every number the game starts from, the bond yields and the Fed’s rate among them, is reset to what the public record shows on 17 September.

Ninety machine-played games stand behind the nine claims. Each claim sits on the record page, a page beside this report that lists every claim with the date we wrote it, the date it resolves, and how many games it came from. One thing to know before you read them. Three of the nine follow from the Treasury plan we chose, not from the play, and the record page marks which three.

When the cycle closes we publish here what resolved which way, what happened in the real world that no player in the game could have done, and what we changed in the game because of what it got wrong.

TLDR

1The US Treasury, the government’s finance department, is buying back its own long-term bonds.

2A bond is a loan to the government. A buyback is the Treasury buying its bonds back early.

3Many people think the real aim is lower long-term interest rates. Nobody has agreed the Treasury may do that.

4On 4 November, a real date, it must say what it will sell, and why. We played those eight weeks as a game, 45 times.

5No speech forced the question open. The Treasury’s own extra spending did, in 9 of 15 games.

6Admitting the aim cost it far more credibility than selling fewer bonds. Credibility is the market’s trust in its plan.

How it was made

Seven roles, each a computer program following rules we wrote. Not people, and not a chatbot.

Five turns, 9 September to 4 November. Each turn a role tries a move, a referee program sets the odds, and dice decide.

The Treasury had three plans for 4 November. We played each plan 15 times. Three games used matched dice, so the plans compare move for move. Twelve more used fresh dice.

Plan 1Keep its current wording.

Plan 2Sell fewer long-term bonds and cite “structural demand”, a lasting change in who buys bonds.

Plan 3Sell fewer and admit it wants lower rates. In October it also spends an extra $40bn on buybacks, an amount we chose.

Everything before 3 September 2026 is real. Everything after is the game.

The players

The TreasuryBorrows for the government. Picks what to sell on 4 November, and gives the reason.

The Fed presidentsThree regional heads of the Federal Reserve (the Fed), the US central bank. They ask the question out loud.

Hedge funds and dealersProfessional bond traders. If they think the Treasury is holding rates down on purpose, they bet against it.

The Fed BoardThe Fed’s chair and governors. They set short-term rates and say nothing about long-term ones.

Pensions and insurersBig, slow buyers of long-term bonds.

Money funds and stablecoinsHold only very short-term government debt. 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.

The numbers

The 45 gamesFig. 1

No speech opened the question in 30 games. The Treasury’s own spending opened it in 9 of 15.

One dot per game, filled where the Treasury’s real aim came into the open before 4 November.

The Treasury’s credibilityFig. 2

Admitting the aim cost the Treasury half its credibility. Selling fewer bonds cost little.

Credibility runs 0 to 100 on a scale we wrote. Each bar is the middle value at the end of the 12 fresh-dice games on that plan. The last two plans sell the same bonds and differ only in the reason.

The counts come from the games’ own record.

What it means

The reason cost more than the decision. Citing “structural demand” cost the Treasury about a fifth of its credibility. Admitting it wanted lower rates cost half.

The cheap reason is a bet. “Structural demand” was believed in 11 of the 12 fresh-dice games on that plan. In the twelfth, the market named the real aim for the Treasury.

If you are judged on a number you only partly control, look hard at your reason for changing course. A reason nobody can check is a bet that nobody ever will.

The line

No speech made the Treasury explain itself. Its own spending did.
And the Treasury was the one player that most wanted the question left alone.

Got it?

READ NORMAL MODE