A government bond is a promise, and the yield is what the promise costs. When that cost rises there are only three explanations available: the borrower will need more money than expected, the money will be worth less than expected, or the promise is believed slightly less than it was.

The first two are ordinary and both are measurable. The third is harder to see directly, which is why it tends to be asserted not demonstrated — and why most claims about the dollar's decline are worth very little.

What follows is an attempt to demonstrate it, using instruments that separate the three explanations: an inflation protected auction that isolates what a bondholder is paid for, a flow record that shows which buyers left and which arrived, two independent surveys measuring the same property from outside finance, and a reserve composition that records what central banks did rather than what they said.

Nothing here shows capital leaving the United States; it did not, in the month under examination, by any aggregate measure. Nothing here shows a successor to the dollar; there isn't one, and the arithmetic of depth says there can't quickly be. What the evidence supports is smaller and more specific — that the particular benefit of borrowing long, cheaply, from buyers who don't haggle is being withdrawn, and that the tools reached for to arrest it are losing their grip faster each time they're used.

That claim is wrong if an official intervention regains durable purchase — if long yields fall after the buyback operations begin on 9 September and stay down beyond a week. It is wrong if the coming quarter shows foreign official institutions returning to US government paper at the long end. And it is wrong if the trust measurements stabilise while the events that moved them continue.

None of those has happened. If any does, it will be said here.

The 30Y Treasury yield closed at 5.31% on 17 August, its highest since June 2007.

It got there against a floor that has risen every year without exception. The lowest close of 2022 was 2.01%; of 2023, 3.54%; of 2024, 3.94%; of 2025, 4.41%; of 2026 so far, 4.64%. Above that rising floor sat a ceiling — the 5.11% reached in October 2023 — which held through two subsequent years. Neither 2024 nor 2025 got back to it. This month it broke, and the market is now pricing 30 year US government duration at a level last seen before the financial crisis.

Two days after that close, the Treasury moved.

At 08:30 Eastern on 19 August, pre-market, it announced it would increase by at least double the size of its liquidity support buyback operations for longer dated nominal coupons — the 10 to 20 and 20 to 30 year sectors — from a maximum of $2bn to at least $4bn per operation, effective 9 September and expiring 4 November, with future sizes to be addressed at that date's quarterly refunding.

The framing that circulated within the hour was unprecedented escalation. The measure is quarter bounded: 7 weeks, with an expiry written into the announcement rather than an open ended commitment. It is per operation and sector specific, not a doubling of the programme. And $4bn is not a new number. On Treasury's published tentative schedule for the current quarter, a $4bn maximum already applied to liquidity support operations in the 1 month to 2 year, 2 to 3 year, 3 to 5 year and 7 to 10 year sectors, while both long buckets — 10 to 20 and 20 to 30 year — sat at $2bn. What was announced brings the long end up to a size already in use elsewhere on the curve.

What is structurally new is the tenor.

Three official mechanisms had already converged on the same part of the curve: the Treasury committing incremental supply to bills, the Federal Reserve buying bills across two maturity buckets, and the Secretary seeking an expanded standing facility. All three sat at the front end. This one is explicitly 10 to 30 years — the tenor where June's flows showed foreign official demand going negative, and where auction concessions have been running largest. Official support is now applied at both ends of the curve. The June flows were published on 17 August; the announcement came pre-market on the 19th.

The price response was immediate. The 10Y closed 6bp lower at 4.65%, the 20Y 11bp lower at 5.17%, and the 30Y 9bp lower at 5.19%. The dollar fell and equity futures rose.

The arithmetic is what makes that a signal and not flow. An additional $2bn per operation, in 2 sectors, across 7 weeks, sits against a quarterly borrowing requirement measured in hundreds of billions. It is a rounding error. A 9bp move at the long end on a posture change with no material purchasing behind it is the market repricing what it believes the issuer will do, not the issuer doing it.

The constraint was never appetite for the operations. On 18 August, the session before the announcement, dealers offered Treasury $19.87bn of 20 to 30 year bonds. Treasury took $2.00bn — the full cap, and roughly a tenth of what was in front of it. Cover ran 10.7x in late June, 15.3x in mid July and 11.0x at the end of it. The long end has been queuing to hand duration back for two months, and the cap held at $2bn throughout.

The reading taken on the day was that this represented a repriced reaction function — that the market now believed the Treasury would defend the long end and had marked duration accordingly. The following session tested it, and the answer depends on which price you read.

On intraday prints the 30Y traded as high as 5.27%, its level immediately before the announcement, paring to around 5.25% after comments from the Secretary; one outlet has it at 5.256% and the 10Y at 4.704%, describing both as back where they started. On the constant maturity closes the 30Y finished at 5.23% and the 10Y at 4.69% — recovering roughly half the announcement move at the long end and two thirds at the 10Y, but still below where they sat on the 18th. The two are not the same instrument: constant maturity is an interpolated par curve read off bid side quotes in the afternoon, not a traded print. There is no reconciling them.

What is not ambiguous is that a 9bp move bought with no flow gave back most of itself inside a single session, and that the tenor structure of the giveback is intact on either basis. The 2Y closed at 4.19% on the 18th, the 19th and the 20th — unchanged to the basis point across the entire episode. This was long end specific rather than a parallel shift, which leaves the front end picture untouched and confirms that the two ends of the curve are answering different questions.

The announcement itself, the intraday move, the flow arithmetic and the tenor pattern all hold. What doesn't hold is the reading that the price response marked a genuine change in belief about how the Treasury will defend the long end. A squeeze unwinds within days; a repriced belief does not.

One thing complicates that. The Treasury sold $8bn of 30 year inflation protected paper on the afternoon of 20 August, the same session as the reversal, and a long end auction is itself a supply event that can move long yields for reasons that have nothing to do with the buyback. What an auction effect cannot explain is the 2Y sitting flat while the 20Y and 30Y gave back most of their move — that shape is specific to the buyback unwinding, not to new supply landing on the curve. On that basis the squeeze reading holds.

The official response to the reversal was verbal, not operational. The Secretary said the operations could exceed the announced $4bn, that they would increase by at least double, and that he has a big toolkit. Against the ladder that matters — announced, signalled, executed — the $4bn is announced and "could exceed" is merely signalled: no figure, no date, no operational commitment. The first actual operation is not due until 9 September. Signalling is being substituted for flow, and the market has now demonstrated it will price the signal for less than a single session.

The floor that has risen every year since 2022 did not move during any of this. Nobody defends a rounding error against noise.

On 20 August the Treasury sold $8bn of 29 year 6 month inflation protected paper, CUSIP 912810US5, at a clearing real yield of 2.973%.

The same security was originally issued on 19 February this year, at 2.473%.

That is not a comparison across a tenor or a vintage; it is the same paper, 6 months apart, 50bp cheaper. And on the full auction record back to 1998, a long dated TIPS has not cleared above 2.973% since October 2001. The intervening period includes 2008, 2011 and the 2022–23 tightening cycle. Nothing in it comes close: the previous high in that span was 2.650%, at last August's reopening of the preceding security.

The gap between the two auctions is not a repricing — a 50bp move between infrequent auctions can hide almost any path. So here it is. The constant maturity 30Y real yield ran 2.43–2.63% through February, 2.47–2.78% in March, held 2.62–2.84% from April to June, then broke to 2.78–3.03% in July before setting 3.06% on 17 August — the highest reading in that series' available history, which begins in February 2010. The prior peak was 2.75%, in May 2025. The 50bp accumulated across 6 months with its steepest leg in July, and it is still accumulating.

What the instrument does that a nominal yield cannot is separate the two things a bondholder is being paid for. Decomposed across the same window: from 19 February to 20 August the nominal 30Y rose 53bp, of which 48bp was real yield and 5bp was breakeven inflation. 91% of the move is the real component.

The implied 30Y breakeven sat at 2.23% in February and 2.28% on 20 August, having traded a range of 2.21–2.28% across the whole of this month. The 10Y breakeven closed at 2.34% on Friday — inside its own 2025 range of 2.17–2.46%, and below its 2026 high of 2.50%.

A term structure being repriced by inflation fear looks nothing like this. Breakevens widen; the compensation for holding nominal paper rises faster than the compensation for holding indexed paper. Here the indexed paper is doing all of the work while the inflation component sits flat within its prior year band, at a 30 year horizon, in the middle of an energy shock and a war. The long end is not charging for prices, but for duration.

One limit on that. A real yield is not a term premium; it decomposes further into expected real short rates and the real premium demanded on top of them. A market pricing higher real policy rates for a generation would produce the same chart. What argues against that reading is the front end, which is not moving with it — the 2Y closed at 4.19% on each of 18, 19 and 20 August while the 30Y round tripped, leaving 2s30s at 104bp on Thursday's closes against 112bp on the 17th. The steepening is being done from the long end. If the market expected a permanently higher real policy rate it would say so at the 2Y first, and it has not.

There is a genuine counter datum in the auction itself. The 20 August auction was well bid. Bid to cover came in at 2.82, above the 2.75 of February's original issue and the 2.78 of last August's reopening — the strongest coverage of the three. Indirect bidders, the category that includes foreign official accounts, took 84.4% of accepted competitive bids, up from 78.3% in February and 70.4% a year ago. Primary dealers were left with 2.1%, down from 2.5% and 4.5%.

That is not a failed auction. Demand for 30 year US government paper is present, and the foreign share of it rose. What changed is the price at which that demand appears. The buyers are still in the room; they are bidding 50bp lower than they were in February for the identical bond.

The intervention of the previous session leaves one more fingerprint here. On 19 August, when the buyback announcement moved the long end 9bp, the 30Y real yield fell 9bp — from 3.03% to 2.94% — while the implied breakeven did not move at all, holding at 2.25%. Official support at the long end was priced entirely as a change in the compensation demanded for duration risk. Whatever the market thinks the Treasury can influence, inflation is not on the list.

In June, foreign investors were net sellers of US Treasury paper across both tenors to the tune of $22.2bn, while buying $181.4bn of US equities in the same month.

Both figures are exact, and need taking apart.

The Treasury number is the sum of two things moving in opposite directions. Foreign buyers took a net $6.8bn of bonds and notes — positive, not negative — while selling a net $29.0bn of bills. So the headline is a bills story, and anyone reading it as evidence that the world has stopped buying American duration has the sign wrong at the tenor that matters most.

The split that carries the finding is not by maturity but by who is doing the buying. Private foreign investors bought $16.6bn of bonds and notes and $6.6bn of bills in June. Foreign official institutions sold $9.8bn of bonds and notes and $35.6bn of bills. Officials were net sellers at both tenors; private accounts absorbed the coupons.

That is the absorption chain doing exactly what it has been doing all year, and it is the mechanism that connects to the auction. A foreign central bank buying Treasuries to manage a currency peg or park a reserve is close to price insensitive; it buys because it must, at whatever the market gives it. A private asset manager is not. When the second replaces the first as the marginal holder, the paper still clears and does so at a yield that satisfies someone who had a choice. The 50bp gap between February's TIPS auction and August's is what that substitution costs, expressed as a number.

Now the counter evidence.

Official selling of long term Treasuries has slowed sharply. Across the 12 months to June 2025, foreign official institutions were net sellers of $91.1bn of Treasury bonds and notes. Across the 12 months to June 2026, that figure was $34.9bn. The pace of official disposal has fallen by roughly 60% year on year. That is direct evidence against any accelerating flight reading.

Nor is capital leaving the United States. June's total net TIC inflow was $133.5bn. Foreign residents bought a net $207.1bn of long-term US securities. Foreign official institutions were net buyers of long-term US securities overall, at $37.3bn, and the official flow across all instruments was a $48.4bn inflow. On every aggregate measure, June was a month in which the world sent America money.

Set the two 12 month windows side by side by asset class. Treasury bonds and notes went from an official net sale of $91.1bn to a net sale of $34.9bn. Agency bonds went from a net sale of $53.0bn to a net sale of $30.5bn. Corporate bonds went from a net purchase of $33.6bn to a net purchase of $57.3bn. Equities went from a net purchase of $1.7bn to a net purchase of $114.3bn.

The official sector has not withdrawn from dollar assets. It has moved along the risk spectrum inside them — out of government paper, into corporate credit and, overwhelmingly, into equities, where a $1.7bn annual flow became a $114.3bn one inside a single year. Reserve managers who once held claims on the US government now hold claims on US companies.

Those are not the same thing, and the difference is the whole argument. Holding dollar assets is reserve status. Holding the government's own liabilities at a yield the government can afford is the funding privilege — the specific benefit that lets a sovereign run large deficits cheaply because a class of buyers takes its paper without haggling. Reserve status is intact and moves on a slow clock. The funding privilege is what these numbers are measuring, and it is not intact.

One date, without further comment. The June figures were published on 17 August and the Treasury announced the long end buyback expansion on the morning of 19 August.

On 20 January, Brand Finance published the Global Soft Power Index for 2026, built on a survey of more than 150,000 respondents across over 100 countries, scoring all 193 UN member states across 55 metrics. The United States recorded the steepest decline of any nation brand in the world — down 4.6 points to 74.9 — while holding first place. China rose 0.7 to 73.5, the only top 10 nation brand to gain. The gap between them narrowed from 6.7 points to 1.4. On the reputation pillar, China rose 9 places to 18th and overtook the United States for the first time; it now scores higher on 19 of the 35 attributes measured. The US declined on every metric except familiarity, with the losses concentrated in trustworthiness, international relations, governance and perceived reliability.

On 15 July, Pew published its Spring 2026 Global Attitudes Survey — 42,151 respondents across 36 countries, fieldwork from 8 February to 13 May. Across the 20 countries surveyed annually since 2023, 46% held a favourable view of China against 36% for the United States. It is the first time in roughly two decades of the series that China has been the more favourably viewed of the two. The US is now seen more positively than China in 6 countries: India, Japan, the Philippines, South Korea, Israel and Poland. In Canada, US favourability fell from 57% in 2023 to 33%, while China's rose from 14% to 44%.

Two methodologies, different sampling frames, different fieldwork windows, same direction.

This however has to be viewed as Western decline rather than a Chinese ascent, and both instruments say so directly. The UK recorded the second steepest fall in the index, down 3.2 points to its lowest position on record and overtaken by Japan; Germany and France also weakened; Brand Finance's own summary has Western nations dropping by more than the global average. The Council on Foreign Relations read the Pew result as a relative repositioning rather than a transfer of affection — China has won the comparison without winning hearts and minds. Anyone reading these numbers as capital preparing to relocate to Beijing has misread them. They measure the erosion of a premium, not the construction of a replacement.

Which raises the obvious objection: reserve managers do not answer surveys, and a foreign ministry's reputation score does not clear an auction.

The connection is narrower than a general claim about sentiment and it runs through one property. What both instruments measure — most explicitly in the attributes where the US fell hardest, trustworthiness and reliability — is whether a state is believed to honour commitments when honouring them becomes inconvenient. That is not adjacent to sovereign creditworthiness. It is the same question, asked by pollsters instead of by creditors. A government bond is a promise; the price of a promise depends on belief in the promiser; and a 30 year bond is a promise about 2056.

The balance sheet version of the same question is more precise.

On the European Central Bank's report of 2 June, gold reached 27% of total official foreign reserves at end 2025, against 22% for US Treasuries and 15% for the euro, with foreign exchange and gold both valued at market prices. A year earlier those figures were 20%, 25% and 15%. Measured that way, gold is now the largest single reserve asset in the world.

The ECB then takes its own headline apart, and the dismantling is the more useful number. Gold rose roughly 60% in 2025 after roughly 30% in 2024, which mechanically lifts its share of anything it is measured against. Correcting for that by holding the gold price at its end 2023 level, gold sits at 16%, the euro at 16%, and US Treasuries markedly higher at 26%. On a constant price basis the crossing has not occurred. What crossed is a valuation.

So the share is a headline and the tonnage is the signal.

Central banks bought around 850 tonnes in 2025 — down from over 1,000 tonnes annually across 2022 to 2024, and still, in the ECB's own characterisation, well above historical norms despite record prices. Poland was the largest single purchaser last year at roughly 100 tonnes, ahead of Kazakhstan, Brazil, China and Türkiye. Since the invasion of Ukraine the cumulative order runs China above 350 tonnes, Poland 320, Türkiye 220 and India 130, with the ECB noting Chinese purchases may be understated. Official holdings now exceed 36,000 tonnes, against roughly 38,000 in the Bretton Woods era, when the dollar was pegged to the metal.

None of that is a price effect. It is a sustained transfer of reserve assets into an instrument that pays nothing, and the ECB's attribution of motive is explicit: persistent geopolitical tension, with central banks making the larger purchases tending to sit in regions of higher external conflict risk. The acceleration dates to 2022.

That is what makes gold the mechanism, and what answers the strongest objection to any erosion reading. The standing rebuttal has always been that nothing has the depth to absorb a reallocation out of Treasuries — the euro has displaced nothing in a quarter century and its share sits where it has sat for years, the renminbi holds near 2%, and the dollar's share of allocated FX reserves is unchanged at around 57%, with dollar assets in aggregate still the largest reserve block at 42%. That objection is a currency argument. It asks where the money goes instead, and assumes the answer has to be another sovereign's paper. Gold is not a competing reserve currency; it is the opt out from holding one — no depth requirement, no market maker, no capital account, no sovereign counterparty.

The counters are substantial. The Turkish central bank sold or loaned around 130 tonnes after the outbreak of the Middle East war — one of the largest reserve drawdowns in recent years — to defend the lira and absorb soaring energy import costs, having accumulated 220 tonnes since 2022. Russia is selling too, reportedly to fund its war. Official gold is not a one way accumulation; a reserve manager under funding stress liquidates it like anything else, and the ECB itself notes gold's limitations as a reserve asset. An instrument paying no coupon is a poor answer to a funding problem — which is why buying it says something about what the buyer fears more than a funding problem.

The custody question is where the abstraction becomes concrete. The Bank of Korea, buying for the first time in 13 years, is reported to be weighing domestic vaulting. And 31 tonnes of Venezuelan sovereign gold held at the Bank of England has been reported as set for routing to a US Treasury account. That transfer has not been executed, and it does not need to be — every reserve manager holding metal in someone else's vault has already read it.

Surveys record what states say about one another. Reserve composition records what their central banks did with the money, and the tonnage records how much of that was deliberate. All three point the same way.

Four official actions in 6 weeks, each intended to move a price. Set them in order and the interesting variable is not whether they worked but how long they lasted.

The currency, 30–31 July. Japan's Ministry of Finance bought yen on both days in coordination with the US Treasury, under the joint statement the two finance ministries issued in September 2025. Goldman Sachs estimates roughly $60bn on the Thursday and $25bn on the Friday, with perhaps a further $20bn on Monday 3 August — up to $85bn across 2 days, against an average daily market volume of around $30bn. It was the largest two day intervention on record outside October 2011, and the biggest coordinated operation in 15 years. Tokyo said it would not hesitate to act again. The effect held for roughly a week.

The chokepoint levy, 13 July. A demanded 20% charge on all cargo transiting Hormuz, withdrawn before it took effect. It was never priced, because it never became operative.

The long end, 19 August. The buyback expansion, 9bp at the 30Y, fully described in the opening section. Reversed the following session.

The verbal escalation, 20 August. With the reversal underway, the Secretary said the operations could exceed the announced $4bn, that they would increase by at least double, and that he has a big toolkit for liquidity problems in the government debt market. Analysts described the market impact of that appearance as minimal. It was priced for less than a single session.

A week, then a day, then part of a day.

The limit on that observation should be stated before anyone builds on it. These are 4 different instruments, in 4 different markets, at wildly different magnitudes, and 4 points do not make a rate of decay. A currency intervention and a buyback announcement are not comparable operations, and the intervals between them are not measurements of the same quantity. What can be said is narrower and still uncomfortable: every one of them was an official action taken to move a price, every one was reversed, withdrawn or ignored, and the reversals came faster in each successive case.

The mechanism that makes the sequence matter — rather than just a run of bad luck — is that the failures are public.

An intervention that works quietly tells the market nothing it did not already believe. An intervention that fails in front of everyone is new information, and the information is not about the price. It is about the issuer. Each of these episodes was observed in full by the same population of institutions that has to decide, at each auction, whether to hold 30 year paper. They watched an $85bn operation buy a week. They watched a doubled buyback buy a day. They watched the Treasury Secretary describe a large toolkit to a market that moved less than a basis point at the front end while the long end rose.

The pattern is not confined to markets. On 13 August the Secretary previewed sanctions measures “never seen in the history of economic isolation on a country,” telling an interviewer to watch for announcements the following week. What that week produced on the Iran line was the OFAC action of 20 August: 3 individuals resident in Turkey, designated through links to the IRGC Qods Force, inside an action whose bulk was an Ecuadorian cocaine network and a Hizballah cash smuggling ring. The same release loosened a Russia related licence. Announcement substituting for instrument, in a second domain, in the same week.

Which brings the sequence to Friday.

Before boarding Air Force One on 21 August, the president was asked whether he had directed the Treasury Secretary to intervene in the bond market. He said he had not. Bessent, he said, is a very capable man who wanted to do it, is very good at it, and has a good natural touch for bonds and interest.

The reporter noted that yields had come back up since the announcement, and asked whether he had spoken with Bessent about another type of intervention.

"We have many types of intervention. That's one," he said. "The ultimate intervention is our military. And if we have to use that, we will."

What he meant by it is not knowable from the transcript and is not the useful question. What is on the record is the position of the sentence in a sequence: an operational measure that failed inside a session, a verbal escalation that failed inside a day, the president publicly disowning authorship of the measure, and then the largest category of state power named as the next rung — 19 days before the first buyback operation is due to take place.

There is a structural reason that rung does not connect to this problem. The assumption underneath it is that power is fungible — that dominance in one instrument converts into compliance in another. The 30 year yield is the price at which institutions, most of them in allied and partner states, agree to fund the United States government until 2056. A threat directed at the holders of that paper is not an argument for holding more of it. It is a reason to demand more compensation for the risk of holding any, which is the term premium, which is the number the whole exercise was meant to bring down.

An instrument that raises the price it was reached for to lower has stopped being one.