The first failure was at sea. The second was in the information between the ship and the people responsible for finding it.

The Virgo Transport 8 left Surabaya on Saturday for Banjarmasin with a manifest listing 213 passengers and 30 crew. In the Java Sea early Sunday, its captain reported waves reaching three meters, then a list, then distress. Contact was lost around 2 a.m. Authorities later received information that the vessel had overturned. Indonesia’s search agency said it learned of the loss two hours after contact disappeared.

By the latest accounting, 108 people had been rescued and six bodies recovered. At least 129 remained missing. Those figures may change as survivors reach shore and search teams reconcile names, but they already describe the awful geometry of the operation: more people unaccounted for than known alive.

A passenger manifest is supposed to remove one uncertainty from a disaster. It cannot calm the water or launch a helicopter, but it can tell rescuers how many lives they are seeking. In Indonesia, officials routinely warn that the number aboard a ferry can differ from the number written down. The country’s 17,000 islands make affordable sea travel essential; that same dependence means a weak count is not a clerical defect. It is a hole in the rescue plan.

The two-hour reporting delay matters for the same reason. An emergency system is not merely a collection of radios, beacons, ships and offices. It is a promise that distress will move through those components faster than the people in the water move away from their last known position. Every break in that chain enlarges the search area.

Sunday’s rescue unfolded after a grim week for regional ferry travel. In the Philippines, the reported death toll from a ferry fire near Palawan rose to 76, with 13 people still missing. Forty-three survived. A survivor described flames and smoke spreading so rapidly that passengers panicked before they could reach life jackets. Off Vanuatu, at least one person died and more than 30 were reported missing after the MV Matui sank between Ambae and Santo; the prime minister’s office cited high winds, warnings that may not have been followed and possible overloading.

These are separate accidents in separate regulatory systems. They should not be compressed into a single imagined cause. But together they expose the same neglected layer of public safety: the record before the wreck. Who boarded? What was loaded? Which warning was received? When did the rescue authority learn that the vessel was gone?

Shipping romanticizes the horizon. Rescue work cannot afford to. It depends on timestamps, coordinates, capacity limits and names written correctly before departure. The sea makes people disappear quickly. The state’s first duty is to know precisely who has not come back.

In much of maritime Southeast Asia and the Pacific, the ferry is not a nostalgic alternative to modern transport. It is the bus, the freight truck and sometimes the only practical connection to a hospital, job or family.

That is the context surrounding three ferry emergencies within days. Indonesia’s Virgo Transport 8 overturned with 243 people listed aboard. The Philippine ferry fire near Palawan killed at least 76 people, according to the latest published count, while 43 survived and 13 remained missing. In Vanuatu, the MV Matui sank in bad weather; 15 people were reported found alive, one dead and more than 30 missing.

The known circumstances differ. Philippine investigators had not determined the fire’s cause, though a survivor reported an apparent blast and fire spreading from the cargo area. Vanuatu’s prime minister’s office said the inquiry would examine high winds, disregarded marine warnings, possible negligence and overloading. Indonesian authorities reported rough seas before their vessel overturned.

The common condition is dependence. Island communities cannot regulate away the need to travel by water, and air service cannot cheaply substitute for ordinary passenger and cargo movement. The meaningful comparison is therefore not ferries versus planes. It is regulated, inspectable ferry service versus an economy forced to accept whatever vessel sails.

Safety begins before the gangway rises: accurate manifests, enforced capacity, cargo inspection, reachable distress equipment, weather authority and a rescue service notified immediately. These provisions look dull beside a new terminal or aircraft order. During a wreck, they become the entire government.

The clean headline is that Qualcomm has won a major AI customer. The revealing footnote is that the customer may earn an ownership position by placing the orders.

Qualcomm said Amazon could purchase as much as $60 billion in AI data-center chips and related products under a long-term agreement. The companies plan multiple generations of custom silicon for inference—the work of running trained models. Qualcomm, still diversifying beyond smartphones, gets a formidable cloud buyer and a route into a market dominated by Nvidia.

Amazon also received warrants reportedly worth about $4 billion. They vest as purchases are made and permit Amazon to buy Qualcomm shares at $161.26 each. That makes the transaction part supply contract, part customer rebate and part capital relationship.

There is nothing inherently mysterious about warrants in a strategic agreement. They align incentives and can compensate a buyer for committing volume to an emerging supplier. But they complicate the market signal. A large order ordinarily says the customer preferred a product at the quoted price. A warrant-backed order says the customer preferred the product, the financial inducement or some combination of both.

That distinction matters across the AI infrastructure boom. Cloud companies want alternatives to Nvidia’s expensive and supply-constrained processors. Challengers need anchor customers before manufacturing at scale. The result is a market in which the buyer may finance the vendor, the vendor may subsidize the buyer’s commitment, and both announce the gross purchase ceiling.

The useful number will not be $60 billion. It will be the purchases Amazon actually makes, the warrants that consequently vest and the performance of the chips once they enter production. The press release records the maximum. The transaction ledger will record the conviction.

Andrew Wiles’s proof of Fermat’s Last Theorem became one of the monuments of twentieth-century mathematics. An experimental Claude system has now turned the theorem into something different: a 13-million-line artifact checked from end to end by a computer.

Anthropic says its agents produced 29,500 intermediate theorems in Lean, a language designed to make every logical dependency explicit enough for mechanical verification. The work took 11 days. Nature reported that a comparable human formalization had been expected to consume years.

This was not the discovery of Fermat’s Last Theorem; the theorem was proved in the 1990s. It was the formalization of the proof—the translation of a vast mathematical argument into a structure whose gaps cannot hide inside phrases such as “it follows” or “by standard methods.” That distinction makes the achievement less cinematic and more consequential.

Mathematics has always depended on compression. A proof persuades because specialists can inspect its decisive moves without replaying every elementary consequence underneath them. Formal verification reverses the bargain. It expands the compressed argument into enough detail for software to check, producing assurance on a scale no human reader will traverse line by line.

The new division of labor is therefore not simply human conjecture followed by machine proof. It is likely to include human goals, machine-generated scaffolding, mechanical certification and then a second human task: extracting the explanation that deserves to survive.

Software engineering already knows this problem. A program can pass every test and remain an architectural horror. Mathematics is now acquiring its own version of machine-generated code debt—not because the result is false, but because a verified artifact is not automatically a teachable one.

Claude’s formalization deserves the word milestone. The next milestone will be smaller: a system that can find the argument inside its own thirteen million lines and tell mathematicians which few ideas did the real work.

A federal judge has ruled unlawful an administration plan directing the Federal Emergency Management Agency to prepare for a 50 percent staffing reduction.

The decision, issued by U.S. District Judge Susan Illston, came in litigation brought by the American Federation of Government Employees and other labor groups over reductions in the federal workforce. According to the Associated Press, senior Department of Homeland Security officials directed FEMA leadership to submit a plan for halving staff even after agency supervisors objected.

The legal question was institutional rather than meteorological. Congress had established protections intended to preserve FEMA’s independence and operational capacity. The court found that the staffing directive conflicted with those constraints.

A judicial ruling does not by itself establish the ideal size or organization of FEMA, and the decision should not be read as a performance audit of every position. It establishes that restructuring an agency must occur within the authority the law grants.

That boundary matters acutely in emergency management. Capacity that appears idle between disasters is the reserve used when multiple events arrive together. Training, regional knowledge and interagency relationships cannot be purchased on the morning a storm makes landfall.

The Atlantic’s unusually quiet hurricane season may reduce immediate operational pressure, but quiet weather does not erase statutory procedure. The court’s judgment returns the argument to its proper venues: Congress, lawful rulemaking and an accountable record of what disaster response is expected to do.

The most revealing detail in the campaign for Christopher Nolan’s The Odyssey was not its pantheon of actors or its source material. It was a four-letter industrial specification: IMAX.

Sight and Sound observed that advance posters largely withheld the usual inventory of faces, character names and breathless promise. Nolan’s name appeared prominently; so did the projection format. The movie was marketed less as an adaptation of Homer than as a reason to enter a particular kind of room.

That is an extraordinary reversal. For years, theaters have advertised themselves as neutral delivery systems for intellectual property whose real commercial life continues on home screens. Nolan’s post-Oppenheimer leverage allowed him to sell the apparatus itself: enormous film, enormous screen, communal attention and a showing that cannot be perfectly reproduced by waiting.

The bet also clarifies what theatrical cinema can still charge for. It is not merely early access. Streaming trained audiences to regard the title as portable and the screen as interchangeable. Premium-format filmmaking argues that the room is part of the work.

There is vanity in making a projection technology share top billing with Homer, certainly. There is also unusually honest advertising. The campaign did not promise that viewers needed another familiar narrative product. It promised scale, duration and surrender to a machine built for looking.

Hollywood spent a decade insisting that theaters would survive because people love movies. Nolan offered the sharper proposition: theaters survive when the theater changes the movie.

The word “manifest” sounds like paperwork because we have been taught to admire infrastructure only after concrete is poured. But a passenger list is infrastructure. So is a cargo declaration, a maintenance log and the timestamp showing when a distress call reached the office that could act on it.

When a ferry sinks, the manifest becomes a moral document. Every accurate name prevents a family from being asked to prove that someone existed aboard. Every enforced capacity figure converts a commercial temptation into a boundary. Every recorded warning establishes who knew what before weather became tragedy.

This is why contempt for administration is so often contempt for accountability wearing a work shirt. The form is mocked because it is slow, repetitive and incapable of inspiring a ribbon-cutting ceremony. Yet the form is where a promise acquires a named beneficiary and an official acquires a duty that can later be examined.

The week’s ferry disasters do not share a single established cause. They do share a demand for legibility. Investigators need to reconstruct passenger counts, cargo, weather warnings, communications and operator decisions. Without that record, grief is forced to compete with institutional improvisation.

A competent state does not merely arrive with boats. It knows whom it is looking for. It does not merely announce a safety limit. It records the load and stops the vessel from exceeding it. Bureaucracy fails when it becomes ritual detached from reality. Its answer is better administration, not romantic ignorance.

The manifest is the state because it converts a crowd into persons the system has agreed to count. Lose the list, and power can always claim it lost fewer people.

Thirteen million lines of formally verified mathematics are an achievement and a provocation. They prove that a machine can satisfy an immense ledger of logical obligations. They do not prove that thirteen million lines are the form in which knowledge should live.

Modern institutions increasingly confuse the ability to produce a receipt with the ability to explain the purchase. A model passes a benchmark; a compliance system fills every field; a theorem prover certifies every inference. The artifact is complete, and therefore the thinking is declared complete too.

It is not. Explanation is an act of selection. It identifies which facts matter, which steps surprise and which abstraction lets another mind carry the result elsewhere. Verification is deliberately indifferent to this hierarchy. Its strength is that it checks the boring step and the beautiful one with equal suspicion.

That makes formal systems indispensable partners and dreadful editors. Claude’s Fermat formalization may protect mathematics from hidden gaps and make previously fragile arguments reusable. But if the intellectual economy rewards only artifacts that a machine can certify, researchers will be pushed toward what produces a green check rather than what produces comprehension.

The proper ambition is not to choose between rigor and readability. It is to demand both as separate deliverables. Let machines expand arguments until no logical debt remains. Then require them—and us—to compress those arguments until the governing idea becomes visible.

A proof nobody can survey may still be true. A culture that no longer asks to understand why it is true has confused certainty with knowledge.