Sunday, 6 September 2026

A single strong month sitting five times above a twelve-month average is more likely to be noise, seasonal adjustment, or a one-off than a genuine inflection. The internals argue for that reading. Of the 162,000, food services and local government education account for 101,000 between them. Local government education hiring in August is almost entirely a question of when districts put teachers and aides back on payroll relative to the seasonal factors, and it reverses or gets revised often. Food services is real employment but it is the lowest-paid and most cyclically fragile category in the survey, and it is not what an economy accelerating on fundamentals looks like.

The information sector losing 23,000 is the number that matters more for anyone who follows the technology industry. That decline is spread across computing infrastructure, publishing and broadcasting, which is to say it is not a single failing firm but a sector-wide contraction happening while the same sector's capital spending is at record levels. Capital expenditure up and headcount down is the substitution story, and payroll data is where it shows up first.

The participation rate is the quiet problem. It edged up to 61.6 per cent in August but is half a percentage point below where it started the year. A falling participation rate flatters the unemployment rate mechanically — people who stop looking stop being counted as unemployed — which means 4.1 per cent unchanged is doing less work than it appears. If participation had held at January's level, the measured rate would be meaningfully higher.

For the Fed this produces genuine difficulty rather than a close call. The labour market is not obviously deteriorating, so the case for insurance easing has weakened. Inflation at 3.4 per cent is well above target and the energy component is being pushed by a war that monetary policy cannot touch — Brent has been trading in the mid-nineties with Hormuz throughput at roughly a fifth of pre-war levels. Cutting into an energy shock risks unanchoring expectations; hiking into an economy adding 31,000 jobs a month on trend risks causing the recession the print says is not happening. The most likely outcome remains no change, and the argument in the room will be about whether the August print or the twelve-month trend is the better description of reality.

One structural note worth keeping: the June and July upward revisions of 55,000 combined come after a run of large downward revisions earlier this year. Revision direction tends to cluster, and when it flips it usually means the birth-death model and the response-rate assumptions are being re-fitted rather than that the economy turned. Treat the level with more confidence than the change.

The choice of target is the whole story. Washington has been able to destroy the Kharg Island export terminal since the war opened on 28 February and has repeatedly declined to. Hitting the terminal would remove Iranian crude exports more or less permanently and would also be the single most inflationary act available in the current market, with Brent already in the mid-nineties and Hormuz throughput down from 21.6 million barrels a day before the war to under five. It would also give Tehran a reason to close the strait completely rather than partially, which is the one escalation the Gulf states and the Europeans have been working hardest to prevent.

Striking tankers instead is a calibrated substitute. It removes carrying capacity rather than production capacity, imposes a cost that is measurable in hulls, and — critically — targets vessels that are already outside the legitimate insurance and registry system. A ship in the shadow fleet has no flag state that will make a serious fuss and no P&I club that will litigate. That makes the escalation cheap diplomatically as well as militarily, which is exactly why it is the rung Cooper chose.

Cooper's stated ratio — three of theirs for two of ours — is worth reading as doctrine rather than rhetoric. It announces a published exchange rate, which is a way of trying to make the ladder legible to the other side and thereby stop it. The risk in publishing a ratio is that it invites the adversary to test the arithmetic at a rung of their choosing, and the IRGC has already been willing to fire at crewed warships.

For the reader the transmission channel is fuel. Every tanker removed from the shadow fleet is carrying oil that was going to reach a refinery somewhere, mostly in Asia. Taking that supply out tightens the same global crude balance that has diesel setting records in the United States — the story this brief covered on Friday. That is the loop: an American missile fired to punish an Iranian missile makes American distillate more expensive. It is not a large effect per hull. It is a persistent one, and it runs in only one direction.

One thing this episode settles about method. Yesterday's Unconfirmed marker was the right call on the evidence available, and the story was worth running anyway because the claim was consequential and the pattern of silence was itself information. But note what the confirmation did: it did not vindicate the Iranian account, it replaced it. That is the usual shape of these resolutions, and it is the argument for publishing the uncertainty rather than the allegation.

Compare what each side actually stopped. A halt along the entire line of contact costs Ukraine offensive tempo it has spent months building, degrades the deep-strike campaign against Russian refineries and logistics that is currently its most effective instrument, and gives Russian units three days to rotate and resupply under no fire. A halt on strikes against Kyiv costs Russia one target set out of many, and Kyiv is the target set most likely to generate Western headlines and therefore most likely to be worth trading. Russia gets the operational benefit and the diplomatic credit; Ukraine gets the diplomatic credit only.

That said, it is a real concession by the standard this war has set, because it is the first time Moscow has given anything in response to a Ukrainian move rather than in exchange for an American one. The three previous 2026 truces — Orthodox Easter in April, Victory Day in May, and the mid-year contacts — were all commemorative and symmetric, agreed in advance for a named holiday. This one is reactive and asymmetric, which makes it a worse deal for Ukraine and a more informative signal about Russian intent.

The phrase to distrust is the White House official's "announcements in coming weeks." Read literally, it means nothing was agreed this weekend that could be announced this weekend. If a framework existed with Russian assent, the incentive to announce it while the envoys are still on the ground would be overwhelming — it would lock Putin to a public position. Deferring the announcement means either that the text is not agreed, or that it is agreed and one party expects to be embarrassed by it. Both readings argue against treating the Kyiv photograph as a breakthrough.

The interceptor shortage is the constraint that will outlast the diplomacy. Ukraine is short of American-made air-defence interceptors, which means the marginal Russian ballistic missile is more likely to arrive than it was six months ago. That degrades Ukraine's position in exactly the way that makes a bad settlement more acceptable over time, and it is why the sequencing of this shuttle — Moscow first, carrying a proposal; Kyiv second, receiving one — is not a neutral choice. Zelensky is being asked to respond to a text he did not draft while his ability to defend cities is declining. The unilateral ceasefire was his answer to that: by stopping first and publicly, he buys the standing to refuse the paper without being blamed for the war continuing.

The energy channel remains the reason this story sits next to the Iran stories in this brief rather than in a separate compartment. Any settlement returning meaningful Russian crude and refined product to Atlantic markets would relieve the distillate shortage now pushing American diesel to records. Nothing suggests sanctions relief is on this weekend's table. Markets will price the possibility anyway.

In arms control the definition is not a preliminary to the rules, it is the rules. A broad definition of an autonomous weapon sweeps in loitering munitions, terminal-guidance seekers and target-recognition software that every major military already fields, and any prohibition attached to it bites immediately and expensively. A narrow definition — for instance, one requiring full autonomy across target selection, engagement and the decision to kill, with no human involvement at any stage and no geographic or temporal bounding — describes almost no system currently in existence or in development. States that do not want to be constrained do not argue against the prohibition. They argue the definition down until the prohibition is empty, and then agree to it warmly. Without the text in front of me I cannot tell you which of those happened here, and neither can anyone quoting the Reuters piece. That is what the November conference will expose.

The consensus rule is the structural reason this has taken twelve years. The CCW takes decisions only when no party objects, which hands a veto to every participant including the states with the largest autonomous-weapons programmes. This is why a parallel push has developed at the UN General Assembly, where decisions are by majority vote and where resolutions on autonomous weapons have drawn support from well over 150 states. The strategic question for Brazil, Ireland and Norway in November is whether to keep spending effort inside a forum where Washington and Moscow hold a veto, or to move the process to one where they do not — at the cost of producing an instrument that the states with the actual weapons will not sign.

The awkward context is that the technology has already outrun the diplomacy. Both sides in Ukraine now field drones with terminal autonomy, adopted specifically because electronic warfare has made continuous human control over the final seconds of a strike unreliable. That is not an edge case in the debate; it is the central one, and it cuts against the treaty's premise. Jamming makes the human link the weakest component, so the operational pressure runs toward removing it. Any instrument that demands meaningful human control at the moment of engagement is asking militaries to accept a capability disadvantage in precisely the environment they now expect to fight in. A definition negotiated to consensus in Geneva in 2026 is being written against a practice that changed in 2023.

For anyone following the AI industry rather than arms control, the connection is more direct than it looks. The same target-recognition stack is dual-use, the same firms supply it, and export-control and procurement rules tend to follow definitions agreed in fora like this one by several years. A definition that excludes software-defined autonomy in favour of describing platforms would leave the commercially significant layer entirely unregulated. That is the outcome the drafters are worried about, and it is why the dilution claim deserves more attention than a single wire story.

The route into federal court that Missouri needs is the Elections Clause theory: that a state court, by invalidating the legislature's congressional map, has displaced the legislature's constitutional authority over the manner of federal elections. The Supreme Court addressed this in Moore v. Harper in 2023 and substantially narrowed it, holding that state courts do review state legislatures on federal election law but are subject to some outer limit that the Court declined to define. Missouri would have to argue its supreme court exceeded that undefined limit. No litigant has yet cleared that bar, and the facts here make it harder than usual: the Missouri court did not strike the map on policy or state-constitutional grounds, it held that a validly filed referendum petition suspended the statute by operation of the Missouri constitution, so the map never took effect at all. That is a holding about when a state law exists, which is about as core to state authority as questions get.

The overseas-ballot argument is different in kind and better. It rests on a federal statute with a hard date, which gives the Court a reason to act that does not require deciding the Elections Clause question at all. The state can ask for interim relief on pure timing grounds. The difficulty is that the same logic runs the other way with equal force.

That is the Purcell problem. The principle counsels federal courts against changing election rules close to an election, on the theory that late changes confuse voters and administrators. Missouri is invoking imminence to get relief, but the relief it wants is the reinstatement of a map that a state court has declared void, weeks before ballots print. The status quo ante — the 2022 map — is the one currently in force. Restoring the 2025 map in September would itself be the late change, and it would be the more disruptive of the two options. If the Court takes Purcell seriously, it argues for leaving the Missouri ruling alone.

The downstream mess I flagged yesterday remains unresolved and is now on a shorter clock. Missouri ran its 2026 primaries under districts that its supreme court says never legally existed. Reverting to 2022 lines means some nominees are now running in differently bounded seats, and ballot-access and residency litigation is the predictable consequence. Courts have occasionally let an unlawful map stand for one cycle rather than unwind a completed primary — that is Missouri's best practical, as opposed to legal, argument, and it is notable that the state has chosen to lead with the federal statute instead.

The seat at stake is one. The House majority may turn on fewer.

Anak Krakatau — "child of Krakatoa" — is the cone that grew out of the caldera left by the 1883 eruption, one of the deadliest volcanic events in recorded history, which killed on the order of 36,000 people mostly by tsunami rather than by ash or blast. The island rebuilt itself above sea level in the 1920s and has been intermittently active since.

The relevant precedent is not 1883 but December 2018. In that event the volcano did not produce a catastrophic explosion. Part of its southwestern flank collapsed into the sea during a period of moderate eruptive activity, displacing enough water to generate a tsunami that struck the coasts of Banten and Lampung with essentially no warning and killed more than 400 people. Indonesia's tsunami warning system at the time was built around seismic triggers, and a flank collapse does not produce the earthquake signature the system was watching for. The wave arrived before anyone knew it was coming.

That is why BNPB's statement is worth reading precisely. "No imminent tsunami risk" is an assessment of current conditions, and it is almost certainly correct as of today: a 30-second eruption throwing ash is not the same phenomenon as a flank failure. But the 2018 mechanism is specifically the one that does not announce itself in advance, and the coastlines at risk are dense, low-lying and close. The monitoring has improved since — there is more instrumentation on the cone itself now, tracking deformation rather than only seismicity — but the fundamental problem, that a landslide-generated wave gives minutes rather than hours of warning, has not changed.

The aviation side is the part with immediate international consequence, and it is under-appreciated because ash is invisible on weather radar. Volcanic ash is pulverised silicate glass. Ingested into a jet engine it melts in the combustor and re-solidifies on the turbine, which can flame out all engines simultaneously — the failure mode that nearly brought down a British Airways 747 over Java in 1982 and a KLM 747 over Alaska in 1989. There is no safe concentration for sustained exposure and no reliable onboard detection, so the entire mitigation strategy is avoidance based on forecast ash dispersion from the regional Volcanic Ash Advisory Centre. A plume at 15,000 metres is the case that matters, because it reaches the altitudes at which aircraft transit rather than only those at which they take off and land. Closing six airports for a few hours is cheap. Closing airways across the Java Sea is not, and that is the scenario a sustained eruption would create.

MBG is a cooked-meal programme, not a packaged-food one. It relies on thousands of local kitchens preparing food that is then transported to schools, which means the safety-critical steps — holding temperature, time between cooking and consumption, water quality, handler hygiene — are distributed across thousands of independently operated sites with variable supervision. A packaged-goods supply chain concentrates risk in a few facilities that can be inspected and certified. A distributed-kitchen model concentrates it nowhere, which makes it far harder to audit and far more sensitive to how fast you scale.

The arithmetic follows from that. If each kitchen has even a small independent probability of a serving failure in any given week, the expected number of incidents scales with the number of kitchens, and the size of each incident scales with how many children one kitchen serves. Boarding schools are the worst case on both dimensions: a single kitchen, a single menu, and hundreds of pupils eating the same food at the same sitting with no alternative. That is why Sidoarjo produced 730 cases from one site. The programme's design does not have a mechanism that makes this less likely as it grows — it has one that makes it more likely.

The political economy explains why it grows anyway. Prabowo campaigned on free meals as the concrete deliverable of his presidency, and the metric that gets reported is coverage: children reached, districts activated, kitchens opened. Safety failures are diffuse, locally reported and rarely fatal. Coverage is national, measurable and announceable. Any system that rewards the second and tolerates the first will scale into exactly this pattern, and the corruption allegations attach at the same seam — procurement for thousands of small kitchens is far harder to audit than procurement for a handful of large suppliers.

The comparison worth holding is that large school-feeding programmes elsewhere, including India's midday meal scheme serving over 100 million children, went through structurally identical failures before centralising kitchens, mandating supplier certification and building incident-reporting systems. The lesson from those is that the fix is institutional and slow — it is inspection capacity, cold chain, and a reporting regime that surfaces small incidents before they become large ones — and that it is almost never funded at the same speed as the meals themselves. Indonesia is early in that curve, and $28 billion of commitment locks in the volume before the safety infrastructure exists.

Marked Single-sourced: the wide coverage traces to one Associated Press report. The 2025 cumulative figure comes separately from Katadata.