At some point before dawn on Wednesday, United States Central Command finished what it called a wave of strikes against Iranian air defence sites, radar systems, maritime assets, mine-laying capabilities and communications sites. Iran answered into Jordan, Iraq and Bahrain. Kuwait said its air defences engaged missiles and drones. Axios reported that American forces had also struck two Iranian government tankers, the first time the tankers have been hit as retaliation rather than as blockade enforcement. Asked afterward whether the strikes were designed to push Tehran back to the table, the President said they were not.

That is the sixth month of a war that began on 28 February, and it is the most serious exchange since July.

In a war, the question that actually tells you something is not what will you do next. It is what would make you stop. Put that question to every institution setting the price of this thing and you get, with one exception, silence.

Central Command published a target list on Wednesday. A target list is not a threshold. The President removed negotiation as a terminus by saying the strikes were not intended to produce one. At Jackson Hole last Friday, Federal Reserve Chair Kevin Warsh declined to give forward guidance and offered instead a standard: "We must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed. Otherwise, we have work to do." Confidence is a state of mind. It has no number, which is the point of saying it that way. The House of Representatives voted 370 to 48 on Tuesday to fund the government to 11 December, a date chosen because it falls after the election. And the Pentagon, which Reuters reported in August has used "virtually all" of its Army Tactical Missile Systems and Precision Strike Missiles, has not published the inventory level at which strike tempo changes. The Center for Strategic and International Studies estimates more than half of at least four critical precision stockpiles are gone, over 1,400 Patriot interceptors fired, and three to five years to rebuild the sophisticated ones.

Now the exception.

On 28 February, the Lloyd's Joint War Committee listed the entire Persian Gulf as a high-risk area. It has said what it will require before it delists: sustained incident-free passage, independently verified demining, and a settled geopolitical picture. Those are exit criteria. They are public, they are testable by a third party, and they are enforced by people who lose their own money if they get them wrong.

The consequences of that one written-down paragraph are now larger than the consequences of most of the week's diplomacy. Before the war, hull war-risk cover for a strait transit ran about 0.25 percent of vessel value. Marsh puts the current range at 3 to 10 percent. On a $100 million tanker that is a per-transit insurance bill of up to $10 million, against roughly $250,000 in January. TotalEnergies chief executive Patrick Pouyanné said last week it costs about $20 million to move barrels through Hormuz. IMF PortWatch recorded one commercial transit on 16 August, the most recent published day, against a pre-crisis median near 73 vessels a day.

That is what an enforced stopping condition looks like from the outside. It looks like a closed strait.

And it prices everything else. Brent sat near $95 on Tuesday after a rise of about 5 percent; US crude cleared $90. Euro area inflation came in at 3.3 percent for August, up from 2.9 percent, driven almost entirely by energy, which rose 14.3 percent year over year. The European Central Bank is expected to lift its deposit rate to 2.50 percent on 10 September. Markets put a Federal Reserve hike on 16 September at roughly two-thirds. The ten-year Treasury yield touched about 4.79 percent, its highest since January 2025. Japan's ten-year hit 3 percent, a thirty-year high. Thirty-year gilts reached levels last seen in 1998.

Every one of those numbers is downstream of a waterway that underwriters will not clear until three conditions they wrote down are met. The belligerents have not written anything down. Neither has the central bank whose rate decision the war now dominates. Neither has the legislature that funds the war and voted overwhelmingly this week to avoid discussing it until December.

There is a defence of vagueness, and Warsh made the best version of it at Jackson Hole: publish a trigger and the market trades the trigger instead of the economy, a hall of mirrors. Fine. But the alternative he chose is not clarity, it is discretion, and discretion has a cost that nobody books. It means a single payroll print on Friday morning now carries a policy decision that ought to rest on a quarter of data. It means the twentieth month of a five-month war arrives without anyone having to explain the difference.

Underwriters are held to a standard the people running the war are not. They must say, in advance and in writing, what would change their minds.

So: what is your institution's stopping condition, and is it written down anywhere a stranger could check? Send it. We print letters on the front page, not the back.

By late morning in Tehran there were four accounts of the same house.

The Iranian Red Crescent said five people were killed and 50 wounded when a residence hosting a wedding in Kuhestak, in the coastal county of Sirik on the Strait of Hormuz, was struck early Wednesday. The Mehr news agency reported that a four-year-old was among the dead. IRNA, citing a provincial official, put the wounded at 68. Esmaeil Baghaei, spokesman for Iran's foreign ministry, called the attack "brutal" and said more than 50 people, including women and children, were killed and wounded, a formulation that merges the two categories and cannot be reconciled with either agency figure. Press TV said at least five were killed.

United States officials had not commented on the incident as of publication. Central Command's statement on Wednesday's operation listed air defence sites, radar systems, maritime assets, mine-laying capabilities and communications sites. It did not mention a residence.

The mirror image ran the same hour in the other direction. The Islamic Revolutionary Guard Corps said it had killed a large number of American forces at bases in Jordan. Two US officials told Reuters that initial assessments found no casualties. Jordan's armed forces issued their own statement. Kuwait confirmed its air defences were engaging missile and drone threats. Qatar condemned Iran's attacks on its Gulf neighbours. Iranian media said the IRGC had downed an American MQ-9; that claim remains unconfirmed.

Baghaei placed Kuhestak in a sequence, saying the strike "cannot be separated from the chain of atrocities that preceded it" and pointing to the first day of the war, when Iranian authorities said American forces hit an elementary school in Minab, killing more than 150, and a sports hall in Lamerd, killing at least 21. Those figures are Iranian claims. This paper has not verified them and does not present them as established.

Here is what the arithmetic of month six looks like in practice. The first release from either side is a number produced by an interested party within hours, before anyone has counted. The second release, the one that revises the first, is usually not published at all. Between them sits the only thing readers can actually use, which is the spread.

We are going to keep the spread. When the Red Crescent figure and the foreign ministry figure and the provincial official's figure disagree by a factor of ten, we will print all three and name who said each. When one of them changes, we will say which one changed and when, on the front page.

The Tasnim agency reported explosions Wednesday in Konarak, Bandar Abbas, Jask, Asaluyeh and Ahvaz, and on Qeshm Island. Those are population centres along the same coastline as the strait the war is nominally about. No casualty figures for any of them had been released by any party at the time of writing, which is its own kind of number.

The Secretary of the Army resigned this week, in the sixth month of a war in which thousands of Army soldiers are deployed, and the stated reason was that the Army is not ready.

Dan Driscoll submitted his resignation to the White House after months of friction with Defense Secretary Pete Hegseth, which the White House confirmed on Monday. A source familiar with the matter told Reuters that Driscoll "brought up concerns with the administration surrounding Army transformation and readiness and Hegseth's blocking those efforts specifically by firing the generals who were responsible." Deputy Army Undersecretary Dave Fitzgerald is expected to step down this week as well.

The generals in question are not abstractions. Hegseth pushed out General Randy George as Army chief of staff earlier this year; George and Driscoll had jointly rolled out the Army Transformation Initiative, whose central bet was unmanned systems. General Chris Donahue, commander of US Army Europe and Africa, filed for retirement. A newly formed Army drone battalion, created specifically to harvest lessons from Ukraine's drone war, is being phased out under acting chief of staff General Christopher LaNeve. Navy Secretary John Phelan left earlier this year. Admiral Alvin Holsey retired from Southern Command.

Set that against the inventory. Reuters reported on 4 August that the Army had expended "virtually all" of its global stockpile of ATACMS and Precision Strike Missiles during the war, a fact that circulated inside the government during arguments about how much longer the United States could keep striking Iran without limiting its ability to respond elsewhere. CSIS estimates the US has expended more than half of at least four of its most critical precision stockpiles, with more than 1,400 Patriot interceptors fired and fewer than 1,000 held in reserve, and puts the rebuild for the most sophisticated munitions at three to five years.

The industrial answer exists and it is real. Framework agreements with Lockheed Martin run up to seven years and contemplate raising PAC-3 MSE output 233 percent, from roughly 600 to 2,000 interceptors a year by 2030, THAAD interceptor capacity 317 percent, from 96 to 400 units annually, and Precision Strike Missile capacity fourfold to around 550. RTX plans more than 1,000 Tomahawks a year.

By 2030. The war is six months old.

That gap is the entire content of the resignation. A modernisation programme is a bet that the next war looks different from the last one; a munitions ramp is a bet that it looks the same but longer. Driscoll and George were placing the first bet. The people who remain are, by default, placing the second, and they are placing it while spending down the stockpile that funds either.

Nobody has published the number. There is no stated inventory floor at which the strike tempo changes, no threshold at which the Pentagon says the readiness cost of another wave exceeds the target value of another wave. Officials told Reuters those conversations happen. They happen in a room.

An Army secretary quitting is the loudest available substitute for a published threshold, and it is a poor one, because it conveys that something is wrong without conveying what would make it right. Driscoll is close to Vice President JD Vance, which will invite a reading of this as positioning. Maybe. The readiness numbers do not care who is positioning.

Kevin Warsh has been Fed chair long enough to give one Jackson Hole speech, and its most consequential sentence contained no numbers at all.

"We must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed," he said on 28 August. "Otherwise, we have work to do." The market heard "at sufficient speed" and repriced. Odds of a 25 basis point hike at the 15-16 September meeting went from roughly 36 percent before the speech to about 66 percent on Monday and 65 to 68 percent on Tuesday, per CME FedWatch. Warsh also noted that inflation has run above the 2 percent target for 65 months, and described short-term rates as the Fed's "predominant tool," reclaiming a job the long end had been doing for him.

What he did not do was give forward guidance, on principle. He has called the alternative a hall of mirrors: publish the trigger and you get traders forecasting your reaction function instead of the economy. It is a defensible position, and it has a specific consequence this week. With no threshold on the record, the entire decision now rests on the last data before the communications blackout begins Saturday. That is the August employment report on Friday morning.

The labour data is not cooperating with the hawkish case. July payrolls fell 23,000, and May and June were revised down by a combined 103,000. Unemployment is 4.1 percent. ISM services showed employment contracting at 47.4 while prices paid ran at 70.3, which is the signature of a cost shock rather than a wage-price spiral. David Kelly of JPMorgan Asset Management wrote this week that "the economy doesn't have quite as much momentum as Kevin Warsh suggested," and that "there is little in the labor market to suggest inflationary trouble ahead."

Meanwhile the strength is real too: second-quarter corporate profits rose 22.8 percent, and capital expenditure is expanding on AI build-out. This is a genuinely split committee facing a genuinely split economy, which is exactly the situation in which one monthly print should not be decisive and now is.

The deeper problem is what the inflation is made of. Euro area prices rose 3.3 percent in August, and energy contributed 14.3 percent. American pump prices are downstream of Brent near $95 and of a freight market where a Gulf-to-China supertanker fetched $647,000 a day in late August. A 25 basis point increase in the federal funds rate does not demine the Strait of Hormuz, does not lower a war-risk premium running at 3 to 10 percent of hull value, and does not build a refinery. It compresses domestic demand in order to offset a supply shock nobody at the Fed controls, which is a legitimate policy choice and an expensive one.

The bond market has already made its own decision. The ten-year Treasury reached about 4.79 percent Tuesday, the highest since January 2025, and started 2026 at 4.20. The thirty-year sits just under 5.3 percent. The two-year is at 4.37, up from roughly 3.50 in January. Japan's ten-year hit 3 percent, a thirty-year high; thirty-year gilts hit levels last seen in 1998. The S&P 500 closed down 0.7 percent at 7,631.47 and the Nasdaq down 1.0 percent at 26,099.77, a third straight losing session.

If Friday's payrolls repeat July's decline, the hawkish case collapses and hike odds go with it. If they rebound, the September dot plot becomes the story, and the question moves from one hike to two. Either way, a central bank that refuses to publish a threshold has handed the threshold to whoever prints the jobs number, and nobody voted for them either.

The most profitable position of the Iran war belongs to a Korean shipowner who bought dozens of supertankers before the shooting started.

Sinokor Group, led by Ga-Hyun Chung and now the largest supertanker player in the world, made what was described in the market as the biggest oil tanker bet ever in early 2026, acquiring vessels ahead of the 28 February strikes. It is now chartering them into a market where the Baltic Exchange benchmark for Saudi Arabia to China reached about $647,000 a day on 27 August and was reported near $656,000 by the end of last week, more than ten times the rate a year earlier.

The reason the number is that large is that it is not one number. Moving a barrel out of the Gulf now carries two shipping costs: a lump sum to get a hull through the strait itself, and then, after the cargo is transferred to a second tanker outside Hormuz, a lower rate for the onward leg. That second leg, Oman to China, runs about $220,000 a day, up from $131,000 a month ago. Patrick Pouyanné of TotalEnergies put the all-in cost of moving barrels through Hormuz at roughly $20 million, and two market participants said it had risen further within the same week. Lars Barstad of Frontline told analysts this is the strongest market since 2004, and roughly twice that.

Embedded in the lump sum is insurance, which is where the structural damage sits. Hull war-risk cover for a strait transit was about 0.25 percent of vessel value before the war. Marsh's range is now 3 to 10 percent; S&P Global reported 7.5 to 10 percent in late July. Ten percent on a $150 million tanker is a $15 million bill for a single passage, against about $375,000 in January.

The underwriters are not making money on this. Howden Re data puts market-wide claims across the war, terror and political violence segment at $2 billion to $3 billion, against estimated annual global premium volume of $1.5 billion to $2 billion for the whole segment. The line has paid out more than a year of income on one waterway. Howden Re called Hormuz 2026 a rare multi-line event testing the global reinsurance market simultaneously, and said the repricing of the marine war-risk baseline is likely permanent.

Permanent is the word to sit with. The Red Sea offers the precedent: Houthi attacks on shipping fell off sharply in late 2025, and premiums in that corridor stayed substantially elevated for months afterward. The Lloyd's Joint War Committee will not lift its Persian Gulf listing until it sees sustained incident-free passage, independently verified demining and a settled geopolitical picture, and each new military event resets the underwriter's incident clock. Tuesday's exchange reset it again. So did Wednesday's.

Which means the cost structure now facing every Asian refiner, every petrochemical buyer and every airline hedging jet fuel does not unwind on the day a ceasefire is announced. It unwinds on the day a third party certifies that nothing has happened for a while, and the war has not gone four consecutive weeks without an incident since February.

Two classes of participant have priced that correctly. One of them bought hulls in January. The other one writes the cover and has already lost the argument about whether the premium was high enough.

The most informative document OpenAI published in the past month was a job listing.

The company is hiring a Power Trading Lead inside its Power & Land team, a role focused on managing exposure to electricity and natural gas, including strategies to hedge against energy price volatility. Roughly two weeks later it posted for a Clean Energy and New Technology Lead inside its Industrial Compute organisation, the group that handles site selection and construction, to evaluate clean power, storage, grid flexibility and low-carbon backup.

Read separately, two hires. Read together, a company that has stopped treating electricity as procurement and started treating it as a position.

You hire a trading desk when you face a price you cannot forecast and cannot avoid. That is a more honest statement about the AI build-out than any of the lease announcements, and the lease announcements have been enormous. Anthropic signed roughly $35 billion with Lambda for about 350 megawatts at a Hut 8 campus in Nueces County, Texas, on which Nvidia holds the lease, days after committing about $45 billion to Nscale for roughly 460 megawatts in West Virginia. Eighty billion dollars of contracted capacity inside a week. Hut 8 expects to connect Nueces County to the grid in the first quarter of 2027, with the first data hall live about six months after that.

A lease is a contract. Electricity is a market, and this year it is a market with a war in it. Brent near $95, European energy inflation at 14.3 percent year over year, and American natural gas exposed to the same LNG dislocations are the reason an AI lab now wants someone on staff who can run a gas book. The compute bill and the tanker bill are the same bill with different intermediaries.

What makes OpenAI's version interesting is that it has already written down a condition, in a place that is enforceable. Under its arrangement for the 3.2 gigawatt Project Camellia in Effingham County, Georgia, the company has committed to provide Georgia Power with up to 1,000 megawatts of flexible demand response over 25 years. That is a published stopping point: when the grid needs the power more than the model does, the model stops. It is in a contract, a regulator can look at it, and a utility can call it.

Compare that with what the same company does when the conditions do not work. OpenAI paused Stargate UK outright, citing the cost of energy and an unresolved regulatory environment. Microsoft subsequently took over capacity at Stargate Norway after OpenAI failed to reach an agreement. Those are also stopping conditions, exercised rather than announced, and they carry a message to every jurisdiction currently offering land: the constraint is not chips and it is not capital, it is whether a governor can tell you the price of a megawatt-hour in 2031.

Texas has paused elements of its interconnection process in ways that could affect how quickly Nueces County reaches full output. Georgia has a 25-year demand-response commitment on paper. One of those is a number. The other is a queue.

The frontier lab that hires the best power trader will win a year of margin that no amount of model quality can produce, and nobody will write about it, because it will show up as an absence: the quarter in which the electricity bill did not spike.

Spotify's AI Persona badge starts appearing on profiles in mid-September, and it is worth being precise about what it does, because the company was precise and almost every summary has not been.

The badge does not tell you whether a song was made with a machine. It tells you whether the artist is a person. It flags a public identity — the name, the face in the banner, the biography — that "may be AI-generated and does not represent a real person." It will show up on the profile, in the About section, in search, and on track rows inside playlists. Artists can self-disclose through Spotify for Artists; Spotify will also apply a "Likely AI Persona" badge on its own review, starting with profiles above defined audience thresholds, and the artist can appeal. Listeners will be able to report suspected personas in the coming months. Separately, an AI Credits tool lets musicians disclose how they used generative tools in the actual work, and tens of thousands of those are submitted daily.

So: a fraud rule, not an aesthetics rule. Correct call, and I say that as someone who does not much care whether a snare was recorded or synthesised. The injury of The Velvet Sundown, the fictional band that pulled millions of streams last summer before anyone noticed it did not exist, was never the arrangement. It was the invented biography, the fake photographs, the sense that you had discovered somebody. Discovery is the entire product. Faking the person is fraud on the listener in a way that using a plugin is not, and Spotify has drawn the line at exactly the place where a listener's actual grievance sits.

The real news, though, is one clause down the page. Music from an AI Persona will be excluded from editorial and algorithmic recommendations by default, unless a listener explicitly follows the act. That is not a label. That is a demotion, and it is the first time Spotify has publicly admitted that the recommendation engine is a curatorial instrument with a taste it can be asked to defend. The company's own phrasing — that it tunes algorithms toward "authentic human artists" and against "spam, slop, or low-effort content" — is an aesthetic judgment written in operations language, and I would rather have it stated than smuggled.

The scale it is being applied to is the part that should make you sit up. The Verge reports that AI-generated tracks now account for about half of Spotify's daily uploads. Half. Spotify removed more than 75 million spam tracks in 2025. Luminate has found listener interest in AI music declining, notably among Gen Z and Gen Alpha, which is a lovely rebuke to anyone who assumed the young would be indifferent to provenance.

Now the loophole, because there is always one, and this one is going to be fun. The rule attaches to identity, not to production. The compliant move for anyone running a synthetic act at scale is therefore to hire a human being: a real face, a real name, a real Instagram, who signs a release and does not write the music. This is not hypothetical. It is the oldest business model in pop, and it is now the cheapest way to stay in the recommendation feed.

Give it a year. The badge will be honest, the profiles will be human, and the songs will be whatever they were going to be anyway. Spotify has solved the lie it could verify. The one it cannot verify is a job posting.

The House passed the continuing resolution 370 to 48. Nineteen Republicans and twenty-nine Democrats voted no. The Senate had already cleared it last month by a similarly crushing margin, and it now goes to the President's desk. Every write-up you will read calls this bipartisanship.

It is not bipartisanship. It is a scheduling agreement, and the schedule is the policy.

Look at the date. Funding runs to 11 December. Not to 30 September, when the fiscal year turns and appropriators would have to defend line items. Not to 31 January, which would put the fight in the lap of whichever party wins the House. Eleven December: five weeks after the polls close, two weeks before the holidays, in the exact window when the country is least able to pay attention and the lame-duck arithmetic is most favourable to whoever needs a favour. Both parties looked at the calendar and agreed, overwhelmingly, that the one thing neither could survive was a vote with consequences before 3 November.

The bill's contents are not nothing, and I want to be fair about that. Democrats extracted language they say restricts further funding transfers to Border Patrol without new limits on federal immigration agents, and language halting an OMB rule that would let political appointees decide grants. Debbie Wasserman Schultz is right to call those real. The measure blocks the White House's $1 billion request to begin building "Trump-class" battleships, which is the funniest line item in American government and also, on the merits, correctly killed. It carries anomalies for WIC, disaster relief and wildland fire, extends surface transportation programmes, and renews economic pacts with sub-Saharan African countries and Haiti for two years.

All good things. All of them expiring, by design, into a month when nobody is watching.

Here is the comparison that should embarrass everyone involved. This Congress found 370 votes to postpone a domestic budget argument. In six months it has not held a single vote on the war that has closed the Strait of Hormuz, put Brent near $95, added 14.3 percent to European energy prices, driven a Fed hike to two-thirds odds, and consumed, according to Reuters, virtually the entire American inventory of long-range precision missiles. The Army Secretary quit this week over readiness. The legislature that would have to authorise any of this has spent its floor time buying itself a quiet autumn.

The defence, offered privately by members of both parties, is that a shutdown would have been worse. Probably true. It is also the argument that has worked every single time it has been made, which is what makes it a mechanism rather than a judgment. A body that can always be talked out of a fight on the grounds that fights are costly has, functionally, delegated its powers to whoever is willing to have one.

There is a version of this vote I would applaud without reservation. It looks identical, 370 to 48, same text, same anomalies, and it is accompanied by a floor debate and a recorded vote on the war. That would be a Congress choosing its battles. This is a Congress choosing its calendar, and the two are only distinguishable by what happens next.

Watch 11 December. Not for what passes. For who is in the room.

We spent the front page this morning demanding that institutions publish the conditions under which they would change course. It would be cheap to do that without publishing our own. So here are this paper's, in the plainest language we can manage, and you should hold us to them.

We print the spread, and we say who said what. When four bodies give four casualty figures for the same house, as happened at Kuhestak on Wednesday, we print all four with the name of the body attached to each. We do not average them. We do not pick the one that suits the sentence. If we ever run a single unattributed casualty number in a headline, we have broken this rule and you should say so.

Corrections run on the front page, at the prominence of the error, in the next issue. Not a footnote, not a silent edit, not a note at the bottom of a page nobody scrolls to. If we get a number wrong in a lead, the correction goes where the lead was. This is the rule most papers say they follow and most papers do not.

We kill a frame when it starts choosing the facts. Every newspaper develops a way of seeing, and a way of seeing eventually becomes a filter. If this paper runs three issues in a row that bend the day's events toward the same metaphor, the metaphor is doing the reporting, and the metaphor goes. We will tell you the day we retire one.

We stop leading with a war when the marginal day adds no decision. Month six of the Iran war generated a real escalation on Wednesday and it earned the front page. Some Wednesday it will not. On that day the lead will be something else, and we will say plainly that nothing changed rather than dressing motion up as news. A war that gets the lead automatically is not being covered, it is being logged.

We do not run a claim we could not defend to a hostile reader using only the sources in the piece. If the argument requires you to trust us, it is not finished. Every reported and analytical article in this issue carries its sources. Click them. Some of them will contradict each other; where they do, we have said so in the text.

We publish our own numbers, including the unflattering ones. Readership, corrections, the days our judgment was wrong. A paper that audits institutions and not itself is running a protection racket.

Two things we will not do. We will not lead on a casualty figure sourced to a single anonymous official from a belligerent government, on either side, no matter how well it reads. And we will not manufacture a controversy to be forwarded. The temptation in this format is enormous and permanent, and the tell is easy to spot from the outside: a paper that is picking fights it did not have to pick is a paper that has run out of things it actually found out.

That is the list. It is short enough to be checked and specific enough to be violated, which is the only test that matters for a document like this.

Now your turn. What is the stopping condition where you work — the written, checkable statement of what would make your employer, your agency, your fund or your union change course? Most people, asked this, discover there isn't one. Some of you will find that there is, and that it is quietly excellent, and we would very much like to read it.

Letters go on the front page. Corrections go there too.