The long bond was bought, and gold noticed

The 30-year fell to 5.20% and the dollar fell with it — EUR/USD +0.86%, USD/JPY -0.80%. Gold took +2.82% and silver +2.79% on the same tape, while the VIX dropped 6.00% to 14.89 and all three US indices closed green. Duration, havens and risk rallied together. Something paid for that.

The long end did not heal yesterday; it was bought. The US Treasury ramped up its buyback operations and the 30-year fell to 5.20% — but the dollar fell with it and gold took the largest metals move of the week. That combination is the tell. A genuine repair of the term premium lifts the currency and leaves gold alone; getting the opposite means the buyer of last resort has arrived and the market has repriced the money rather than the credit. The same indifference to events shows up in our prediction feed, where the odds of a near-term Hormuz settlement have now decayed for eight straight sessions while crude added +0.66% — the strait has stopped setting the oil price. Two unrelated markets, one message: the denominator is doing the pricing now, and the assets that cannot be issued are the ones keeping score.

The evidence that settles which story this is comes from the metals, and it is close to decisive. Gold rose +2.82% and silver +2.79% — a gap of 0.03 percentage points between an asset that is a haven and an asset that is mostly a high-beta claim on the same monetary thesis. A geopolitical premium does not do that. It buys gold, it sells industrial risk, and it bids volatility. Copper did not move (+0.05%), the VIX fell 6.00% to 14.89, and all three US indices closed higher. Whatever bid the metals yesterday was not fear about a strait; it was a view about the unit of account. The Treasury's own operations are the named mechanism , and the alarm around government-bond yields that framed the week — with federal debt at $40 trillion — is the context that makes the operation necessary in the first place.

The proprietary layer says something the tape cannot. Our probability feed has split cleanly in two on Hormuz: the diplomatic bands have decayed for eight straight sessions, from 0.685 to 0.205 for an August agreement and from 0.320 to 0.075 for the 22 August window, while the physical closure band has round-tripped roughly ninety percent twice in three sessions. Conviction sits in the diplomacy and nowhere else. That matters because the band which has actually trended all week is neither — it is the toll, up from 0.355 to 0.440, the one outcome that needs no signature and no closure. Crude, which added +0.66% and is holding the level it struck on 17 August, is the wrong instrument to express any of it. This desk should say plainly that it learned this the expensive way: on 18 August it led on a single print of the closure band and was wrong within a day. The correction is in the footnote and stays there.

Step back and the day is about what a price is for. A long yield is supposed to be a vote — on inflation, on solvency, on what term risk costs. When the issuer becomes a large and discretionary buyer of its own long bond, that yield stops being a vote and becomes a policy setting, and the information the market used to put there has to go somewhere else. Yesterday it went into gold and the dollar. This is the same structural drift this desk flagged on 19 August as a de-linked regime — the long end and the metals no longer agree — and the disagreement has now been resolved in favour of the metals on a day when the sovereign was buying. That call is paid. The falsification is clean and dated: if the long bond holds below 5.10% through the next long-dated auction WITHOUT further buyback support, and gold gives back the 2026-08-19 move, then this was a genuine term-premium repair and the flow reading was wrong. Conviction on the monetary read: high. Conviction on the physical Hormuz path: low, and deliberately so — see the footnote.

**Methodology footnote.** Cash figures are the last SETTLED session (19 August 2026); this note is written before the 20:00 UTC US settle and no unsettled print is quoted. Positioning is the CFTC report dated 2026-08-11 and is therefore nine days stale — it is cited as a reporting category, never as a census of buyers. Probability bands are live, unsettled market-implied odds unless labelled settled. **Correction of method:** on 18 August this desk led its Signal on a single print of the Hormuz zero-transit band (0.494); the band halved the next session and the call was scored wrong on 19 August. Today it prints 0.426. We now treat that series as noise at daily frequency and read only its level, and we place conviction in the monotone deal series instead. Stated once, not revisited.