Two islands were in the news overnight. Only one of them was attacked.

US Central Command said its aircraft struck two Iranian rocket launchers on Larak Island, inside the Strait of Hormuz, after observing crews preparing to fire sea mines into the shipping channel. It was the first American strike on Iranian territory since 29 July, ending a month in which Washington had traded bombing for sanctions. Iran's state broadcaster IRIB reported that the strike killed and wounded fighters and civilians.

The Islamic Revolutionary Guard Corps replied within hours. It said it had fired ballistic missiles at the King Hussein and Al-Azraq air bases in Jordan, hitting what Iranian state television described as technical infrastructure, maintenance facilities and fighter-jet positions, and claiming heavy damage. Tehran also said it struck American personnel at Al Minhad in the United Arab Emirates and downed a drone over the strait. The UAE denied that a base housing US forces had been targeted. The two governments are describing the same night in incompatible terms, which is now routine, and we are printing both descriptions rather than picking one.

Then came the third event, and it did not happen.

Late Sunday the president posted a ten-second video to Truth Social showing an oil storage tank and a tanker consumed by explosions, with an aircraft overhead. The caption read: *"Kharg Island being blown to smithereens!!! President DJT."* Kharg is in the northern Persian Gulf, roughly 660 kilometres from Larak. There is no report of any strike on it. The White House and the Defence Department did not respond to Reuters outside business hours. Detection tools and wire services identified the footage as synthetic.

Here is the part worth keeping. Kharg Island handled close to 90 percent of Iran's crude exports before the war. According to the ship-tracking firm Kpler and the consultancy Energy Aspects, its loading berths have been inactive since 31 July. The video destroyed a terminal that had already gone dark a month earlier. The image was not a preview of a strike. It was an announcement about a facility that had already been taken out of the market by other means, dressed as an air campaign.

The market did not read it that way, and the market was not entirely wrong to ignore it. Brent rose above $90 a barrel on the Larak strike, closing around $90.28, up about 2.5 percent, after an overnight high of $90.69; West Texas Intermediate traded near $85. That is a war premium going straight back into a contract that had shed more than 4 percent last week on hopes that shipping through Hormuz might resume. One island, two launchers, eight dollars of round trip.

The underlying traffic numbers explain the sensitivity. Roughly five commodity vessels a day crossed the Strait of Hormuz over the weekend. Before the war the figure was near one hundred. Kpler puts direct crude exports through the strait at about 2.2 million barrels a day, down from roughly 17 million barrels a day of Gulf crude exports in 2025 to around 9 million now. Note, however, that other reporting circulating on Monday put total flows through Hormuz at 6 to 8 million barrels a day. Those two figures are not reconcilable, they are both being quoted by serious people, and the difference between them is larger than the entire production of several OPEC members. We do not know which is right. We are telling you the dispute exists rather than quietly choosing the number that makes the sentence read better.

What is not in dispute: the war crossed its six-month mark last week; a ceasefire was agreed in April and a memorandum toward a peace deal signed in June, and neither has produced an agreement; and Washington's stated strategy since late July has been economic pressure rather than airpower. Sunday broke that pattern in the narrowest possible way — two launchers, one island, inside the channel itself.

The strait is where the war now lives. Everything else is set dressing, including the dressing the president made himself.

Tell us where this is wrong. If you work in Gulf shipping, marine insurance, or tanker chartering and you can reconcile 2.2 million barrels a day with 6 to 8 million, we will print your explanation with your name on it, or without it if you prefer.

Nepal's disaster authority put the death toll from Wednesday's glacier collapse at 903 on Monday, with the combined Nepal–China figure at 908 and more than 3,000 people still missing. Rescue teams reached the upper section of a mud-filled tunnel at the Trishuli 3A project over the weekend, drilled through, pumped in air and lowered a camera toward roughly a hundred workers believed to be alive inside. Specialist tunnel teams from India and China are on site. Nepal's mountain rescuers, the ones who normally work above 6,000 metres, are working underground.

Those are the numbers everyone is running. Here is the one that has barely been printed outside Kathmandu.

The flood took 431.1 megawatts off Nepal's national grid — twelve generating facilities, eleven hydropower plants and one 25 MW solar farm, according to the Nepal Electricity Authority and the energy ministry. Nepal's installed capacity is roughly 4,300 MW. A single wave removed about a tenth of the country's electricity in the time it takes to boil rice. Another 470 MW under construction across fifteen projects was damaged: next year's power, gone before it arrived. The Trishuli 3B 220 kV hub substation in Nuwakot was destroyed outright.

Read the list of casualties as an engineering document rather than an obituary: Rasuwagadhi 111 MW, Sanjen 78 MW, Upper Trishuli 3A 60 MW, Sanjen 42.5 MW, Chilime 22 MW, Langtang 20 MW, Trishuli 24 MW, Devighat 14 MW. Every one of them sits on the Bhote Koshi–Trishuli corridor. They are there because that is the best stretch of falling water in the country, which is another way of saying they are there because it is steep, glacier-fed and above a border where a lake has just formed and begun overflowing.

The missing list has the same geography. Of the people unaccounted for, 933 were employed at eleven hydropower sites and tunnels in Rasuwa and Nuwakot, per the Independent Power Producers' Association. Nepal did not merely lose plants and separately lose workers. It concentrated the generating capacity and the workforce in the same valley, and the valley was the hazard.

This is not an argument against hydropower in Nepal, and nobody serious in Kathmandu is making one. Water is the one energy resource the country has in surplus, and it spent a decade climbing out of load-shedding that once ran eighteen hours a day. The World Bank-adjacent reviews and the NEA's own figures point to a wider import gap from India this dry season rather than a return to the blackout years. The argument is against serial siting: six facilities on one river turns a local glacial event into a national infrastructure event, and turns a labour force into a single point of failure.

Finance Minister Swarnim Wagle has put losses above $4bn. The European Union has pledged $2.3m. The IFRC has allocated just over $1m and appealed for $31m. China has sent 2,100 rescue workers and about $32.7m. The relief total is currently around one percent of the loss estimate, and the second phase of relief — humanitarian supply and road reopening — has not started, because the roads are the other thing the river took.

Fresh flood warnings remain in force. Chinese authorities alerted Nepal to surging flow upstream on the Bhote Koshi over the weekend. The same corridor, the same alert chain, the same week.

Three weeks from Sunday, Mecklenburg-Western Pomerania votes. The current polling average has the AfD at 35.9 percent against the SPD's 28.9, Die Linke at 11.2 and the CDU at 9.8 — a projected 30 seats to the SPD's 24, with Die Linke on nine and the CDU on eight. Saxony-Anhalt votes this month too, and there the polls suggest the AfD could come close to an absolute majority. Berlin also goes to the ballot.

What makes the northern race worth watching is not the headline number but the seat maths underneath it. Thirty of seventy-one is not a government. The AfD needs either a coalition partner or its own majority, and neither is on offer: every other party in the Landtag rules out a coalition. The firewall holds in Schwerin not because anyone reaffirmed it this week but because 35.9 percent is not 50.

The AfD's lead candidate, Leif-Erik Holm, has structured his campaign around that gap in an unusually revealing way. He is a sitting Bundestag member and he declined a place on the state list. He is running only for Schwerin's direct mandate, against Minister-President Manuela Schwesig. If the AfD leads a government he becomes minister-president; if it does not, he keeps his federal seat and loses nothing. The party's 237 delegates in Grimmen unanimously adopted a ninety-page document titled a *Regierungsprogramm* — a governing programme — promising a *blaue Wende* across migration, security, education and culture, much of which a single state government could not enact even with a majority. It is a programme written to be quoted rather than executed.

The more consequential fight is in Karlsruhe, and it has not started. Calls to refer the AfD to the Federal Constitutional Court as unconstitutional have revived ahead of the September votes. Either the federal executive or the Bundestag can file; from the legislature it takes only a simple majority. The obstacle is arithmetic again: Chancellor Friedrich Merz's CDU opposes a ban, and without the CDU there is no majority to file.

So the German centre is holding two positions simultaneously. The AfD is dangerous enough to be excluded from every coalition in every state parliament, and not dangerous enough to be referred to the court that could rule on whether it is dangerous. Those positions are compatible in law. They are increasingly hard to hold in public, especially in a state where the excluded party is polling nine points clear of the governing one.

One historical caution against reading the polls as a result. Mecklenburg-Western Pomerania's incumbents tend to recover late. In 2021 the SPD trailed the CDU through the summer and won the autumn by twenty-six points. Schwesig personally polls at 46 percent as preferred minister-president against Holm's 25. The AfD's problem in the north has never been its ceiling of support. It is that the party is more popular than its candidate.

Brent settled Friday having lost more than 4 percent on the week, its first weekly decline in three, on a story that went roughly: mines are being cleared, Gulf flows may improve, higher US rates will dent demand. By 05:24 GMT Monday it was quoted as high as $90.68, up a reported 5.29 percent from the prior session on one exchange feed and about 2.5 percent to $90.28 on another. WTI traded near $85. The whole of last week's move reversed on two rocket launchers.

That is what a supply curve looks like when the elasticity has been removed from it. Consider the traffic. Five commodity vessels a day crossed the Strait of Hormuz over the weekend. Pre-war the figure was around a hundred — a decline of roughly 95 percent sustained since mid-July. Direct crude exports through the strait average about 2.2 million barrels a day on Kpler's numbers, against Gulf crude exports of roughly 9 million a day now versus 17 million in 2025. Analysts estimate five to seven million barrels a day of Gulf oil is being disrupted. Much of what still moves does so under naval escort or with transponders switched off.

When ninety-five percent of the traffic is already gone, the remaining five percent carries the entire price signal. Every barrel in that channel is a marginal barrel. Two launchers on Larak are therefore worth more to the front-month contract than a serious change in OPEC policy would be, because OPEC policy cannot get through the strait either.

Three things follow, and readers positioned in this market should weigh all of them.

One: the reopening trade keeps failing. Each flare-up pushes the expected restoration of Hormuz shipping further out and makes a return to US–Iran negotiations before the end of the third quarter less likely. Traders keep buying the peace and then selling it back within days. Note, though, the counter-signal: an Iranian official said Friday that resuming diplomacy "isn't impossible" after talks with Qatar, the standing mediator.

Two: the sanctions channel is now permanent. Treasury has promised fresh secondary sanctions on a weekly cadence. That is a policy designed to produce a headline every seven days in a market that reprices on headlines.

Three: Venezuela does not fix this week. The reserves Washington announced control over are real and large — 65 billion barrels — and they are geologically and logistically incapable of moving a cargo in the next month. Heavy sour crude in the Orinoco is not a substitute for Gulf medium sour loading at Ras Tanura, and the infrastructure to lift it does not exist yet. Anyone pricing Venezuelan barrels into a Q4 balance is pricing a press conference.

Despite Monday's jump, both benchmarks are heading toward a modest decline for the month of August. The volatility is enormous and the trend is flat, which is the signature of a market that has stopped discovering a price and started reacting to a wire feed. Inventories built before the war have largely been consumed. If nothing structural changes, the next six months carry more risk than the last six did, and with far less cushion.

On Tuesday, Shein begins trading in Hong Kong under 0625.HK. It priced at HK$48.56, raising about $1.73bn at an equity value near $26.5bn. That is 73 percent below the $98.2bn attached to a 2022 private round and about 59 percent below the $64bn Series D-plus in 2023. The book was covered. Retail demand was not enthusiastic: roughly 4.66 times in broker margin financing.

On the same day, in the same city, Mech-Mind debuts. It is an industrial machine-vision and robotics company. In the identical subscription window it was oversubscribed 3,842 times in margin financing.

That ratio — call it eight hundred to one in appetite — is the most honest sentence anyone has written about the state of capital allocation in Asia this year.

The context is a listing boom that has already outgrown last year. Hong Kong and Shanghai IPOs and secondary listings have raised more than $54bn in 2026 against $46bn for all of 2025, roughly 21 percent of the global total and second only to Nasdaq's 55 percent, which was itself inflated by SpaceX's $75bn float in June. In July, CXMT, China's largest memory chipmaker, raised more than $8.6bn on Shanghai's STAR market and rose 466 percent on its first day. Its revenue grew more than 700 percent year on year to 50.8bn yuan in the first quarter, on AI chip demand. AGIBOT and Deep Robotics are queued behind. "The current IPO boom is powered by investor appetite for AI and robotics," S&P Global Market Intelligence's Ruiying Zhao said, and the subscription data does not leave much room to argue.

Shein's discount is not entirely a verdict on Shein. Some of it is policy: the US and EU rollback of de minimis exemptions on small parcels removed a structural cost advantage that the private valuation had capitalised as permanent. Some of it is competitive. And some of it, as WPIC's Jacob Cooke put it, is that "the AI investment cycle is absorbing much of the risk appetite that would have otherwise flowed to a company like Shein."

But read the terms rather than the narrative. Because the IPO priced below thresholds set in earlier rounds, Shein may distribute as much as $3.5bn in cash and additional shares to preferred investors whose downside protections were triggered. Against $1.73bn raised, that is a company going public partly to settle with the people who funded it privately. New public shareholders are buying into a cap table that owes money to the old one.

The use of proceeds says the company knows what the market wants: roughly 40 percent to technology infrastructure, 40 percent to global expansion and brand, with investment flagged for Guangdong manufacturing hubs — a direct answer to the supply-chain scrutiny that killed the New York and London listings. Boyu Capital, Tiger Global, Tencent and General Atlantic stayed on as cornerstones at the reset price.

One analyst's framing is worth keeping: unlike PDD, Shein has no anchor in the Chinese domestic market; unlike Alibaba, it has no cloud or AI business to grow into. What it has is logistics and taste, both of which are now cheaper to buy than a robot that can pick a part out of a bin.

Two things are happening at once in American AI policy, and they only make sense read together.

The first. OpenAI has a model family called Astra. It confirmed the name on 1 August with a report claiming an internal version solved ten open problems in mathematics and theoretical computer science, several of them untouched for decades, with Lean certificates published to GitHub — machine-checkable proofs, at a reported token cost around $2,000. On 7 August the company flagged internally that Astra may meet the "Critical" cyber-capability threshold under its own Preparedness Framework: capable of independently discovering zero-day vulnerabilities and running complex attacks against hardened systems. OpenAI says the finding is preliminary and that Astra has not demonstrated every Critical capability under production conditions. It also paused its largest planned frontier reinforcement-learning run, published a document on 18 August titled *Pacing model development in an era of cyber-critical capabilities*, extended monitoring to all Astra tool-use inference and sandboxed execution. Outputs from a checkpoint under the codename mozaik-alpha-fdm began circulating on 29 August. There is no model card, no pricing, no model ID, and no decision on whether it ships as GPT-6 or as a point release.

The second. On 2 June the president signed an executive order, *Promoting Advanced Artificial Intelligence Innovation and Security*, without a ceremony or a livestream. It gave agencies sixty days to build a classified process for designating a "covered frontier model" and a voluntary framework letting companies submit such models for government review up to 30 days before release. The deadline was 1 August. The White House says it met it. The programme is called TRAINS, runs out of NIST, and involves more than ten federal agencies and ten national labs. OpenAI, Anthropic, Google, Microsoft and xAI have signed. Meta has not.

The framework itself has not been published. Neither has the benchmark that decides which models are covered, which is classified and led by the NSA, nor the threshold that separates a covered frontier model from an ordinary release. The framework document is *unclassified* and still withheld. A White House official's explanation, on the record: "Just because things are unclassified that doesn't mean we are going to broadcast them to everyone."

The order forbids using the framework as "a mandatory governmental licensing, preclearance, or permitting requirement." That is the legal position. The practical position is that former officials told Defense One participation is the gateway to federal funding eligibility, against a Defense Department FY2027 request seeking more than $54bn for AI. Roughly a hundred organisations reportedly hold some form of access already, under eligibility criteria that have also not been published. The signed order is a substantial retreat from the draft the president rejected on 21 May, which had a ninety-day mandatory review and formal federal evaluation authority.

Put the two halves together. The most consequential model of the next year is one its own developer says might cross a Critical cybersecurity line. It is expected to be among the first through a federal review process whose passing grade is a state secret. Open-weight models are explicitly excluded from the covered-model designation, so the entire structure applies only to closed systems — and, as one analyst noted of the 30-day window, it is hard to give the government early access to weights you intend to publish anyway.

We are not arguing here that the benchmark should be public. There is a real case that publishing a cyber-capability test is publishing a study guide. But a secret standard is not accountable to anyone outside the room, and "voluntary" is doing an extraordinary amount of work in a sentence that also contains fifty-four billion dollars.

If you were in the room at the White House industry briefings in early August, we would like to hear what the framework actually asks for. We protect sources.

The Australian Recording Industry Association announced on 25 August that recordings it classifies as AI-generated will no longer qualify for the ARIA Charts. The rule takes effect with the chart dated 31 August — today's. Music made with generative tools can chart only if it is "substantially human made" and "raises no stream or chart manipulation concerns." Wholly AI tracks are also barred from the ARIA Awards.

The proximate cause was a cover of Madonna's *Like a Prayer* that charted for weeks and reached No. 4 on the Top 20 Australian Singles chart. Nobody stopped it because nobody was quite sure they were allowed to, and because a decent number of people liked it.

ARIA is following IFPI, which issued similar guidance last month for the charts it runs across Latin America, the Middle East, Africa and Southeast Asia: mostly human-made, no manipulation, and produced on a properly authorised and lawful AI service. In the US, Universal, Warner and Sony have demanded the same framework, asking for "a clear boundary between human-led creativity and works that are purely synthetic or generated by unauthorized AI models." Synthetic tracks have already surfaced on several Billboard charts, including Hot Gospel and Hot R&B.

What interests me is not whether the rule is right. It is what a chart is now for.

A chart was a measurement. It counted what people bought and played, and its authority came from indifference — it did not care whether a record was good, only whether it moved. What ARIA has done is add a qualifying condition that has nothing to do with consumption. The AI *Like a Prayer* had genuine listeners. Real people streamed it and, presumably, enjoyed it. The rule removes it anyway. The chart has stopped being a thermometer and become a licensing body: it certifies provenance, not popularity.

There is a coherent argument for that, and it is not the sentimental one. It is an antitrust argument. Synthetic supply is effectively infinite and costs nothing; human supply is finite and expensive. Put them in the same ranking and the ranking degenerates into whoever can generate the most. Spotify removed 75 million spammy tracks in one sweep last September and found AI songs uploaded to dead musicians' profiles. That is not a taste problem. It is a flooding problem, and flooding is what charts are structurally defenceless against.

The cost is that verification becomes the industry's new job, and everyone is bad at it. A study last year found 97 percent of people cannot distinguish AI from human music. Meanwhile audiences have overcorrected into universal suspicion: the reveal trailer for *Humankind 2* was denounced within hours as "AI slop," and Amplitude Studios had to insist it used "real actors, practical sets, and custom-made props." That may well be true. It does not matter. In 2026 an image is guilty until proven human, and studios now cite their production methods the way restaurants cite their farms.

The Academy went the other direction for the 98th Oscars: AI use neither helps nor harms a nomination, and voters may weigh how far a human remained at the centre of authorship. That is vaguer and probably more honest, since a film today may contain AI-assisted colour, de-aging, crowd generation, dubbing and object removal without the filmmaker having gone anywhere.

A chart can say a record was substantially human made. It cannot say the record was worth making. We have spent thirty years pretending those were the same question, and the machines have finally made us tell them apart.

Argument invited. If you think ARIA has it right, or you make records and can describe where "substantially human made" actually breaks down in a session, write in.

Every deepfake threat model I have read since 2019 has the same protagonist: a hostile intelligence service fabricating a video of a president announcing something he did not announce. Provenance standards, cryptographic signing, detection tools, rapid-response teams to say *that was not him*. All of it assumed the fake would be unauthorised.

Late Sunday, the President of the United States posted a ten-second AI video of Iran's Kharg Island exploding, with the caption "Kharg Island being blown to smithereens!!!" There was no strike on Kharg. The actual American operation that day hit two rocket launchers on Larak, six hundred and sixty kilometres away. Every element of the detection stack works perfectly here. The clip is trivially identifiable as synthetic; wire services said so within hours. Provenance is not in question, because the account is verified and the poster is the head of state. The signature is valid. The content is false. The system performed exactly as designed and prevented nothing.

This is the failure the field did not plan for: not counterfeit authority, but authentic authority issuing counterfeit events. There is no technical remedy for a genuine signature on a fabricated fact.

It is worth being precise about why the post works. Kharg handled roughly ninety percent of Iran's crude exports before the war. Its berths have been inactive since 31 July. The terminal was already out of the market. So the video does not claim an attack that could be immediately disproven by a tanker sailing away from it; it claims, in the register of spectacle, something that is arguably already true in the register of economics. Kharg *is* out of action. The video simply supplies a cause more satisfying than sanctions and insurance withdrawal.

That is the innovation. Not a lie about the world, but a dramatisation of a fact the audience has no independent way to picture. The economic strangulation of an oil terminal is invisible. Fire is not. Generative video closes the gap between what is true and what can be felt, and it closes it in the direction of force.

Consider the incentives from a strategist's desk. A real strike on Kharg would cost aircraft, risk aircrew, kill people, invite retaliation on Gulf allies, and spike crude past $120 as it did in April. A rendered strike costs nothing, kills nobody, and delivers most of the coercive signal — Tehran cannot be certain the next one is fake, Gulf capitals must plan as if it were real, and the domestic audience gets a victory. If you believe deterrence runs partly on the perception of willingness, synthetic force is deterrence at a ninety-nine percent discount. It would be strange if states did not use it.

The cost is not borne by the person posting. It is borne by every subsequent true claim. Iran's Revolutionary Guards said Monday they had inflicted heavy damage on two Jordanian air bases. The UAE denied that an American-occupied base had been hit. Somebody is wrong. Ordinarily the American government's account would be the load-bearing one for most Western readers. It is now the account of an institution whose principal published a fabricated battle damage assessment nine hours earlier. That is not a partisan observation; it is an epistemic one, and it applies to whoever holds the office next.

Journalism does not have an answer yet. The old rule was *verify before you amplify*. It is insufficient: when the fabrication is official, the post has done its work in the ninety minutes before anyone checks, and printing the correction reprints the image.

My tentative rule: describe, never display. We will not embed the video. We will tell you what it purports to show, that it is synthetic, and what the actual military record says, in that order, and we will name the real island every time we name the fake one. It is a small discipline and it may be futile. Tell us if you have a better one.

On 21 August the president announced that up to 300,000 metric tons of ground beef could enter the United States for ninety days without triggering the out-of-quota tariff, with importers committing to sell it at 25 percent below market. Ground beef is running near seven dollars a pound. The midterms are nine weeks out. The ninety days expire comfortably after the votes are counted.

Cattle markets fell the same morning.

What makes this worth a column is not that a politician did something short-term about a price. That is the job. It is that the intervention is aimed at a shortage the administration's own policies helped manufacture, and it is opposed most loudly by the people who voted for him.

The American cattle herd is at its smallest since the 1950s — a seventy-five-year low — after years of drought, high feed costs, herd liquidation and now New World screwworm. Rebuilding a herd is a biological process measured in years: you hold back heifers instead of selling them, you eat the revenue loss now, and you get more calves in twenty-four to thirty-six months. The only thing that makes a rancher do that is a confident expectation of high prices. The single most efficient way to destroy that expectation is to announce, without warning, that the government will arrange below-market foreign supply whenever the retail number gets politically uncomfortable.

That is why the National Cattlemen's Beef Association, representing 180,000 producers across 44 states, called it flooding the market with government-subsidised, below-market beef. It is why the US Cattlemen's Association noted that 300,000 tonnes is roughly half of total US beef export volume this year and said ranchers are "being used as pawns in a 90-day political timeline." It is why Republican senators from ranching states broke publicly with the president over it. These are not opposition talking points. They are the president's own coalition explaining, in trade-association prose, that the policy does the opposite of what it says.

And the consumer benefit is close to a rounding error. Analysts put the likely retail effect under one percent. Beef trimmings are an input; the path from a tariff waiver to the shelf price of eighty-five-fifteen at a supermarket runs through packers, grinders and retailers, all of whom have margin to absorb before any of it reaches a customer. Meanwhile the same ranchers are paying war prices for fertiliser and diesel, because the Strait of Hormuz has been running at five vessels a day since March and fertiliser is one of the shortages that produced. The administration is subsidising imported protein to offset an input-cost shock generated by its own military campaign, and calling the subsidy affordability policy.

There is a real counterargument. Seven-dollar ground beef is a genuine hardship, and a president who ignores grocery prices for the sake of long-run supply elasticity deserves whatever the electorate does to him. The Yale Budget Lab estimate that the full tariff programme costs the average household nearly $1,100 a year is the kind of number that should force emergency measures.

But notice what an honest version of this would look like. It would be permanent tariff relief on inputs rather than a ninety-day window on outputs. It would not require exporters to make a price commitment the government cannot audit. It would not conveniently expire after an election. The design tells you the objective, and the objective is the announcement.

Beef will be marginally cheaper this autumn and meaningfully more expensive in 2029, because the herd that would have been rebuilt this year will not be. That is the trade. It was made without telling anyone it was a trade.

Ranchers: write to us. If you were holding back heifers in August and changed your mind after the 21st, we want to publish that, with numbers.