PANAMA CITY — Asked on Thursday how long the United States can strangle Iran's ports, Defense Secretary Pete Hegseth gave the answer that costs nothing to say.

“Indefinitely the United States Navy can maintain a blockade like that because we'll rotate ships in and out, as we have, and we'll continue to,” he told reporters. Treasury Secretary Scott Bessent added the financial half, promising economic measures “that have never been seen on Iran.”

The word *indefinitely* is a claim about hulls. Hulls are the easy part. The hard part is the roughly 5,000 people aboard the USS Abraham Lincoln, which has now been deployed more than 250 days on a cruise that was supposed to end in May.

Here is what has come out of that ship in the past week. A sailor went overboard earlier this month, three U.S. officials told CNN — an incident that had not been previously reported. Families and active-duty crew told Stars and Stripes that at least one other crew member was physically stopped from going over the side. Representative Mike Levin, writing to the Pentagon, itemized the conditions: “moldy showers, broken toilets, laundry down for weeks, long stretches with no hot water, a meal that came down to half a cup of rice and two tortillas. No soap, deodorant, or toothpaste.” A separate congressional letter cited water contamination, deck safety, and care packages lost in transit for months.

The Navy's rebuttal is the most revealing document in the file. A Navy official told the BBC that “traditional supply hubs” had been “disrupted by combat actions,” and described the resulting triage in order: “First food, then hygiene items, then mail.”

Read that sentence next to Hegseth's. A naval blockade is not a moral position; it is a logistics operation. It works by making the other side's supply lines unaffordable. The Navy has just explained that its own supply line to the ship enforcing the blockade has been degraded by the same war, badly enough that the service had to rank food above toothpaste above letters from home. The theory of victory and the reported failure are the same theory, pointed in two directions.

Navy leadership disputes the framing. Officials say the Lincoln has adequate food and clean water and functioning air conditioning, that fresh produce is hard to deliver because it spoils en route, and that logistical flow is being actively restored. In April, when photographs of small and unappetizing meals first circulated, Hegseth called the reporting “fake news.” On Thursday he said the latest accounts were “completely misrepresented.” The Navy has not confirmed any increase in suicidal ideation aboard.

Relief is finally moving. The USS George Washington strike group departed Vietnam on August 12 and is reported to be heading to replace the Lincoln. That rotation is the physical proof of Hegseth's word, and it also demonstrates its limit: the Navy can sustain this posture indefinitely only in the sense that it can keep spending carriers on it. There are a finite number of them, each one arrives with a maintenance debt, and the last one came home with a mental-health problem the service is still describing in the passive voice.

The clock is not only mechanical. The war began February 28. A June memorandum of understanding to reopen the Strait of Hormuz has collapsed. The International Energy Agency now expects global oil supply to fall 4.3 million barrels a day this year, about four percent, and revised that estimate downward by 600,000 barrels in a single month. The national average for regular gasoline is $4.07 a gallon, the highest August on record. Midterm elections are in November.

An indefinite blockade is a policy that requires either an adversary who folds or a public that stops noticing. Iran has not folded; it is charging tolls. The public has noticed; 81 percent disapproved of the administration's handling of fuel prices in May polling, including 94 percent of independents.

Which leaves the third party to the arrangement, the one that cannot vote on it and cannot leave. Asked what the deployment feels like, a Lincoln sailor told Stars and Stripes: “There's no Monday, Tuesday, Wednesday, it's just a day.”

That is what indefinitely means, operationally. It means a calendar with no marks on it. The secretary used it as a boast. Somebody on that ship has been living inside it since February.

There is a cheap way to settle the argument over who controls the Strait of Hormuz, and it does not require reading a single statement from either government. You look at where the ships actually go.

In August, according to ship-tracking data from Kpler cited by The Wall Street Journal, there were 166 confirmed transits of the strait. Two of them used the coastal route through Omani waters that the U.S. Navy protects. Two.

About half of the remaining traffic used the route administered by Iran. The other half switched off their transponders, which is its own answer: those captains preferred to be invisible to everyone rather than visible to the wrong party. On a single day this week, 14 vessels transited, 11 of them by Iran's lane. Before the war, the strait carried more than 130 ships a day.

On Wednesday the president wrote that “The U.S.A. has total control over the Strait of Hormuz. I THINK WE WILL KEEP IT! Our Naval Blockade is being called, by everyone, ‘A WALL OF STEEL,’ and there is nothing Iran can do about it.”

There is a specific thing Iran can do, and it is doing it. Control of a waterway does not require defeating the Sixth Fleet. It requires making commercial operators afraid enough that the protected route becomes the one nobody uses. Iran is not contesting the U.S. Navy. It is contesting the underwriters, and the underwriters concede.

The price of that concession is public. War-risk premiums for a Hormuz transit now run 7.5 to 10 percent of a vessel's hull value, per Marsh and S&P Global figures. Before February 28 they were about a quarter of one percent. For a $100 million tanker that is a jump from roughly $250,000 per crossing to $7.5 million or more — a fortyfold repricing, quoted in seven-day increments and re-marked every 24 to 48 hours. Ships with American, British, or Israeli ties pay multiples of the base rate. Some brokers have eliminated no-claims discounts entirely.

That number is the actual measurement of control, and it is not being paid to Washington.

The attacks continue to make the case for the underwriters. The UAE foreign ministry condemned what it called a hostile Iranian attack on two Abu Dhabi National Oil Company vessels transiting the strait on Thursday evening, days after a similar strike on an ADNOC tanker on Saturday. No injuries in either. ADNOC says 15 of its vessels have been attacked in the strait since the war began — a striking figure for a country that is not a belligerent and is formally aligned with the United States.

What the UAE is absorbing, in other words, is the transaction cost of an American security guarantee that a majority of ship operators no longer route themselves through.

Meanwhile the arithmetic hardens. The IEA now projects global oil supply falling 4.3 million barrels per day in 2026, with a 1.27 million bpd deficit against demand, and describes the strait as effectively closed since the June deal collapsed. Prediction markets that priced a 91 percent chance of normal Hormuz operations resuming by year-end in June had it at 47 percent by August 11.

Iran's stated conditions for reopening have not moved: sanctions relief, release of frozen assets, an end to U.S. military operations. It is negotiating with Oman over the strait's future management, which is to say it is negotiating over an asset it is behaving as though it owns.

Two ships out of 166 took the American route. That is not a wall of steel. That is a lane nobody is using.

Ukraine hit Novorossiysk on Wednesday night with a mixed salvo of long-range drones and missiles. Three people were killed, including an eight-year-old, and 24 were injured. President Volodymyr Zelenskyy said the frigates Admiral Essen and Admiral Makarov, a large landing ship, and a corvette were struck. He did not mention grain terminals. The grain terminals were struck too — all three of them, closed.

Add the earlier strike on Taman and the subsequent suspensions, and roughly 95 percent of Russia's Black Sea grain export capacity is currently offline. Russia is the world's largest wheat exporter. Its seaborne grain exports were already down 20.3 percent year over year in July. The head of the Russian Grain Union told Reuters that August shipments may struggle to reach two million tonnes, against 5.7 million tonnes a year ago. One analyst puts 2026/27 exports at 21.7 million tonnes, or 17.5 million if the security situation worsens, against more than 60 million last season.

Chicago wheat closed the week around $6.71 a bushel, up 0.2 percent.

That is the whole story, and it deserves to be stated without decoration: a quarter to a third of the world's exportable wheat supply is sitting behind damaged infrastructure in an active war zone, and the price moved about two-thirds of one cent.

Wheat has risen roughly 25 percent from January, to two-year highs, so the market is not asleep. But the recent tape says something more specific than complacency. It says traders have priced a resolution: that the terminals get repaired, the cargo gets rerouted through the Baltic or Caspian or over land as Russia's agriculture ministry has proposed, and the tonnage eventually reaches the same buyers a few weeks late.

That is a forecast about politics, not about grain. And the political precondition is missing. Ukraine has reportedly transmitted an offer through a third party for both sides to stop attacking civilian targets in the Black Sea. As of Thursday it had received no answer.

The supply picture underneath is not a cushion either. The USDA now estimates that combined production among the top seven wheat exporters will fall 11 percent for 2026/27, with their exports down 7 percent, on drought across the Northern Hemisphere. Fertilizer availability is constrained, partly because of the Persian Gulf. Russia struck Ukrainian ports on the Danube in reply to Novorossiysk. The head of the Kazakhstan Grain Union estimates the two countries together could fall 30 to 40 million tonnes short of world market needs — enough, he says, to push global food prices 15 to 20 percent higher.

There is no clean way to hedge that at $6.71. Which is roughly what a very good analyst note said out loud this week: we cannot choke off close to a third of the world's wheat export trade and expect the price to sit still forever. The longer the disruption runs, the larger the correction it has deferred.

The buyers who will settle this argument are not in Chicago. Wheat importers across the Middle East and North Africa are sitting on decent domestic harvests and have, if anything, slowed purchases. That patience is the load-bearing element of the current price. When one large state buyer decides to build strategic stocks instead of buying spot — and that decision will be made by an agriculture minister who reads the same shipping data everyone else does — the calm ends in an afternoon.

Markets are usually right about grain and usually wrong about wars. This is a bet that the second thing resolves in time to protect the first.

The S&P 500 closed Thursday at 7,798.99, a record, up 0.65 percent, after crossing 7,800 intraday for the first time. The Nasdaq gained 0.81 percent to 26,803.03. The Dow added 0.13 percent, its gains capped by Cisco, down 9 percent.

The stated reason: July producer prices were unchanged month over month against expectations of a 0.2 percent rise, with core PPI up 0.2 percent. That followed a July CPI that rose 0.1 percent, in line. The 10-year Treasury yield eased to about 4.65 percent.

Now the part that gets skipped in the recap. Traders bought a record high because the CME FedWatch tool moved to roughly a 63 percent probability that the Federal Reserve holds rates steady in September — down from a 55 percent chance of a *hike* one week ago, to about 35 percent on Thursday.

The bull case is the absence of a rate increase. Headline PPI is running 4.7 percent year over year, core 4.2 percent. Equities are at all-time highs in an economy where the live monetary question is whether policy needs to get tighter.

That inversion is worth sitting with, because it tells you what the market thinks it owns. It is not pricing disinflation. It is pricing the belief that this inflation is imported through a strait, and therefore not the Fed's problem to solve with the funds rate. As one chief investment officer put it plainly on Thursday, the key to the inflation picture is a resolution in the Middle East or pipelines that bypass Hormuz, and the Fed has no influence over either.

He is right about the mechanism and that is precisely the risk. If your rally rests on the premise that the central bank will look through supply-side inflation, you are long a judgment call by a committee, made under political pressure, about a war with no ceasefire. The Fed's own economists will tell you the distinction between transitory supply shocks and embedded expectations is one you only get to make in retrospect.

There is also a hole in the calendar. August CPI and PPI both land before the September decision. Thursday's data was not the final word; it was the second-to-last data point traders will get to be wrong about. Meanwhile crude settled back near $83 a barrel and the AAA national average for gasoline ticked up to $4.07 — after a week in which the market's own oil narrative was falling demand and a 17.4 million barrel build in U.S. crude inventories.

A sober reading of the week: consumer prices decelerated at the margin, wholesale prices flattened, the labor market softened enough in July to take the hawks off the table, and equities converted all three into a record. It is a coherent trade. It is also a trade whose main assumption is being negotiated, badly, by governments in Muscat and Washington.

Earnings season is nearly done. Jackson Hole is at month's end. Between now and September there is nothing to hold the index up except the story it just told itself.

Jose Chajon-Raxon was processed into Delaney Hall, a privately run immigration jail in Newark, on July 18. He had a seizure on July 19. The Department of Homeland Security says he was taken to the hospital and “released from ICE custody on July 22.” He later died.

DHS did not disclose the death and, this week, explained why it did not have to. “Per ICE policy, when an individual is no longer in ICE custody then ICE will no longer be responsible for monitoring or reviewing deaths that may occur,” the department said. “This is common sense.”

It is not common sense. It is a definition, adopted earlier this year, that ends the practice of reporting deaths that occur within 30 days of release. And it is a definition that does specific work: an agency that can release a person into an ambulance can decide, at the moment of transfer, whether the outcome will appear in its own statistics. DHS has not said when Chajon-Raxon died, or whether he ever left the hospital.

The number that rule now protects is 51 — the deaths DHS has reported in detention since the start of the second Trump administration. Researchers writing in JAMA in April concluded that ICE custody is deadlier than at any point in two decades, with suicides at an unprecedented level. That is the count before the accounting change takes effect on the next twelve months of data.

What sits underneath the count is harder to see, and a KFF Health News and Associated Press investigation published this month is the most systematic attempt so far. Reporters analyzed dockets in roughly 33,000 habeas corpus cases filed by detainees between January 2025 and March 2026. Only about 4,400 included the original petition, because federal rules keep those filings off public court websites; the rest had to be collected by volunteers or read in person at courthouses. Within that narrow window, reporters identified more than 300 cases containing specific sworn allegations of delayed, denied, or deficient care, across at least 33 states — medications withheld for high blood pressure, diabetes, epilepsy, HIV, Parkinson's; infections left to fester; cancers untreated; seizures.

The reporters excluded dozens of vaguer claims to be conservative. They note the sample is neither random nor representative and the allegations were not independently verified. Both caveats cut the same direction: 300 is the floor of what is visible in the sliver of the record the public is allowed to read.

Courts have started to say the same thing in the language that binds. On February 10, a federal judge in California ordered ICE to provide constitutionally adequate health care at the California City Immigration Processing Center, a CoreCivic-operated facility in Kern County, and appointed an outside monitor. The monitor, Dr. Muthusamy Anandkumar, found 13 areas of deficiency, including inadequate emergency care and chronic disease management and untimely access to prescribed medication, and reported in a filing made public July 27 that the agency is in noncompliance with the order.

The money has its own paper trail. ICE told Congress it spent more than $390 million on detainee health care in fiscal 2023; its then-acting director said in May the agency had already spent nearly half a billion this year, with more than 65,000 people in ICE and CBP custody. When the Department of Veterans Affairs stopped processing ICE's offsite medical claims in the fall of 2025, the agency awarded two no-bid contracts worth over $110 million to replace it. Records obtained by American Oversight indicate outside providers have not been paid for care delivered on or after October 1, 2025, even though one contractor has received roughly $45 million.

Hospitals that do not get paid see fewer patients from the facility that does not pay. That is not a scandal; it is a supply curve. It is also the mechanism by which a policy about billing becomes a policy about who receives care, and then a policy about who dies within 30 days of a release that was itself a discharge from the ledger.

An agency is allowed to define its own custody. It is not entitled to have the rest of us adopt the definition.

Two frontier labs shipped on Thursday, and the interesting thing about both announcements is what they did not say.

Google released Gemini 3.7 Flash, generally available in more than 160 countries, three weeks after Gemini 3.6 Flash. A million-token context window, 64,000 tokens out, text and images and video and audio and PDFs, tool use, computer control. The headline number is not a benchmark. It is $0.75 per million input tokens and $3.75 per million output — an introductory price that is half what 3.6 Flash cost, running through year-end before rising to $1.50 and $7.50 on January 1.

OpenAI previewed Ultrafast, a service tier that runs GPT-5.6 Sol at up to 750 output tokens per second, roughly fourteen times its standard speed, powered by Cerebras hardware. Not a new model. The same model, on a different track. On GDP-Val, a benchmark of economically valuable knowledge work, the company reports a 5.6x end-to-end speedup with no quality degradation. It is invite-only, expanding as capacity grows.

The framing to notice: one company competed on unit economics and one competed on latency, and neither one led with being smarter. That is what a market looks like when the product has stopped differentiating and the buyers have started reading invoices.

The reason is agents. A chatbot answers a question and a human waits three seconds and nobody dies. An agent runs a hundred model calls to finish one multi-step job, and at that point latency compounds and cost compounds and both of them show up in a monthly bill that a procurement officer has to defend. Google is explicitly pitching 3.7 Flash as the cheap brain for autonomous systems and putting it behind Spark, its always-on personal agent. OpenAI's named use cases — incident response, customer support, financial market analysis, e-commerce — are all places where the model is inside a loop with a clock on it.

There is a structural detail in the OpenAI announcement that deserves more attention than the speed figure. The fastest way to run OpenAI's best model is currently on chips OpenAI did not make, from a vendor whose whole business is being faster than the GPU incumbent. The frontier labs spent three years arguing that scale was the moat. The thing they are now selling is inference throughput, which is a hardware property, and it is available for rent.

Anthropic offers a fast mode that does not reach these speeds. Everyone will ship one.

What none of this tells you is whether the models got better this week, and the silence is the news. Gemini 3.7 Flash is described as the most intelligent workhorse model, which is a category with one member. Ultrafast is described as the same intelligence, sooner. Halving a price and multiplying a speed are both real engineering and both are the moves of an industry that has run out of a bigger claim to make and has started fighting over margin.

For buyers this is good news and it is temporary. Note the date on Google's price: the introductory rate expires January 1, and doubles.

Since August 2, every response from a new Claude model has carried a statistical watermark: small, deliberate patterns in word choice that a human reader cannot see and a detector can. Generated files get digitally signed provenance metadata. Anthropic says this applies globally, not only in Europe, and that a text detection API is coming.

The legal driver is Article 50 of the EU AI Act, which requires providers to mark AI-generated or manipulated output. Models released after August 2 must comply now; older models have until December. OpenAI has outlined its own approach, so far focused on images and audio rather than text. This is not one company's philosophy. It is the beginning of a compliance regime, and it arrives with a design flaw the company itself has published.

The flaw: a mark applied at the model level cannot distinguish between writing an essay and fixing a comma. Anthropic's own documentation says it plainly — people use Claude to proofread, translate, summarize, and convert files, and the output can carry a Claude mark even when the underlying ideas and text came from the person. Axios ran the corporate version of the scenario: a communications team that pastes a human-drafted press release into Claude to clean up formatting has now stamped it.

Which means the watermark answers a question nobody was asking. It does not tell you *who wrote this*. It tells you *this passed through Claude*, a fact roughly as diagnostic as knowing a document was opened in Word.

Anthropic is careful about this in its documentation and it does not matter, because the mark will not be read by careful people. It will be read by a plagiarism dashboard at a university, an HR contractor, a freelance client who wants a discount. Those systems do not have a field for *processed, not generated*. They have a boolean.

The reaction has been split in a way I find more interesting than either side. On Reddit, one long post called the watermark a digital tattoo on the user's forehead. The most-upvoted replies were variations of “take a deep breath.” Both camps are correct about different things. The panicking user is right that a mark with no gradations will be used as if it had none. The unbothered majority is right that the detector is fragile: run Claude's draft through another model, rewrite it heavily, mix it with your own paragraphs, keep it short, and the signal degrades. Anthropic concedes all of this, and adds the corollary that nobody wants to hear — the absence of a mark proves nothing either.

So we get the worst topology for a trust system: a signal that is unreliable in both directions and authoritative in tone.

The cultural consequence is not that AI writing becomes detectable. It is that a defensive literacy becomes standard practice. The professional class that has spent three years quietly using these tools now has an incentive to launder outputs through a second model, or to retype rather than paste, or to keep a document's revision history as an alibi. We are going to develop the habits of people who assume their prose is being fingerprinted, because it is.

Meanwhile the platforms are converging from the other side. LinkedIn is testing a button that lets readers flag something as “seems like AI slop.” Substack has embedded a third-party detection suite. Snap has stopped promoting AI-generated video in its main feed. None of these need the watermark to work. They are the audience building its own instruments.

That is the part that will outlast the compliance deadline. Regulators wanted labels. What readers actually want is a way to know whether anyone was home when the sentence was made. A statistical pattern in token selection cannot tell them. It can only tell them which vendor was in the room.

The numbers are genuinely astonishing and I want to give them their due before I ruin them.

*Spider-Man: Brand New Day* crossed $700 million domestic on Wednesday, thirteen days after release. Endgame and The Force Awakens each took sixteen. It opened to $360 million, the biggest domestic opening ever. Its $1.108 billion international haul makes it Sony's largest overseas release, passing *No Way Home*. Worldwide it sits at $1.81 billion, tenth all time and climbing, with trackers projecting a $910 to $960 million domestic finish — which would put *The Force Awakens* and its $936.6 million genuinely in range.

Destin Daniel Cretton made a superhero movie about loneliness and audiences went four times. Tom Holland and Sadie Sink are being praised for acting, in a Marvel film, sincerely. Good. I liked it.

Now the ruin. The summer box office crossed $4 billion this week for only the second time since the pandemic, the other being Barbenheimer summer. That is being reported as health. Look at what is inside the number: one Spider-Man, one *Odyssey*, one *Toy Story 5*. Take *Brand New Day* out of the summer and the season is ordinary. A $4 billion box office assembled from three tentpoles is not an industry with a wide base; it is an industry with three legs, and Sony is currently standing on one of them and describing the view.

This is what I mean by lifeboat. The theatrical business has not recovered the middle — the $40 million adult drama, the mid-budget comedy, the thriller that plays for six weeks because people tell each other about it. It has instead perfected the event: a film so large that not seeing it becomes a social position. *Brand New Day*'s second Wednesday did $13.5 million. Its second weekend was the third biggest ever. That is not a movie people are seeing; that is a movie people are *attending*, repeatedly, because for a few weeks it is the only thing everyone else has also seen.

Monoculture is not a dirty word to me. I miss it. The problem is that we now manufacture it deliberately, three times a year, and call the intervals a slump.

What interests me more than the receipts is where the next franchise is sourcing its identity. Paramount has named Tems executive music producer and soundtrack curator for *Children of Blood and Bone*, giving a Nigerian artist creative authority over a global property's musical architecture — not a needle drop, the architecture. Studios have worked out that a soundtrack featuring Tems and Burna Boy and Ayra Starr arrives with an audience already assembled across streaming and diaspora networks, before a frame is shown.

That is a smarter idea than another record. It says the growth is in reaching audiences Hollywood has spent a century treating as a secondary market, and letting them shape the thing rather than receive it. The Spider-Man record will be broken. Some of those records get broken every few years by whoever owns the biggest character.

A business that can only survive on its biggest character is a business that has agreed to stop looking for the next one.

Opening this weekend against the juggernaut: *The End of Oak Street*, with Anne Hathaway and Ewan McGregor, and *PAW Patrol: The Dino Movie*. One of those is the mid-budget movie everyone claims to want. Go see it, or stop complaining.

Every failing policy eventually reaches its adverbial phase. The nouns run out — no ceasefire, no reopened strait, no capitulation — and the government begins governing with modifiers. *Indefinitely.* *Never seen before.* *Total control.* You can date the end of a strategy fairly precisely by the moment its defenders start describing duration instead of destination.

That moment was Thursday. The Secretary of War said the Navy can maintain the blockade indefinitely. The Treasury Secretary promised economic isolation like the world has never seen. Neither man said what either instrument is supposed to *produce*, or when, or how anyone would know. Six months into a war that began February 28, the American position has been distilled to an assurance that it can be held.

It can be. That is not in dispute. The question no one in the administration is willing to put in a declarative sentence is what holding it costs, so let me itemize, using the government's own numbers and this week's prices.

Gasoline: $4.07 a gallon, the highest August average on record, against $3.16 a year ago. Diesel: $5.26, up more than $1.50 in twelve months — a freight tax that arrives in every grocery aisle about six weeks later. Global oil supply: down a projected 4.3 million barrels a day this year, per the IEA, a forecast the agency worsened by 600,000 barrels in a single month. Exxon and Chevron: $26.5 billion in combined second-quarter net income, Chevron's a record. Approval of the president's handling of fuel prices: 19 percent, with 6 percent among independents. Overall approval: 34 to 35 percent in recent polling.

And the strategic return on all of it: two out of 166 transits of the Strait of Hormuz this month took the route the U.S. Navy protects.

A blockade is a bet that your opponent's pain arrives before your own. Iran's pain is severe and real — its main source of hard currency is cut off, its energy infrastructure damaged. But Iran's government does not face voters in November, and the material it needs to keep going is patience, which is domestically produced. The United States needs gasoline under $3, a functioning majority in Congress, and about five thousand people on a carrier who do not break. Two of those three are on a deadline the other side can read in the newspaper.

This is the specific incompetence of the current policy: it has made American endurance the variable and Iranian endurance the constant. Every additional week of blockade transfers a little more leverage to the party that is not counting weeks.

I would take the hawkish case seriously if it were being made. There is a coherent argument that reopening Hormuz on Iran's terms — sanctions relief, unfrozen assets, tolls on passage — would ratify a precedent worse than $4 gasoline: that a middle power can convert a chokepoint into an annuity. Fine. Say that. Say the price and say what we are buying and let voters weigh a maritime precedent against their fuel bill. That would be a strategy, and an honest one.

Instead we get *indefinitely*, which is an argument that requires no defense because it makes no claim.

The administration is not choosing to prolong this war. It has simply not found the sentence that would end it, and duration is what accumulates in the absence of a decision. Somebody will eventually write down what the last six months bought. My guess is that the honest ledger has one entry on the credit side — a precedent not conceded — and that the debit column runs several pages, in dollars, in barrels, in a carrier crew that lost track of which day it is.

An indefinite policy is not patient. It is a policy that has stopped keeping score because it does not like the score.