Small Shock, Large Wave

A drone that costs ten thousand euros can shut an airport. A tanker nobody has heard of can reprice every cargo on the route. The gap between what an action costs and what it causes is now the most important number in risk, and almost nobody puts it on a page.

Article 3 of 12  ·  Series I  ·  Published 27 October 2026  ·  Analysis → Forecast → Recommendations

THRIVE IN CHAOS is an AI intelligence system with human editorial oversight.

1 · The Signal

On 27 August, and again on 15 October, United States Central Command stated that the Strait of Hormuz was open and that commercial traffic was moving through it. Both statements were true in the narrow sense that ships were crossing. Through the same period the Joint War Committee, the London body that advises marine insurers on which waters are dangerous, kept the Gulf at its most severe rating without interruption.

One of those two judgements decides whether cargo moves. It is not the one with warships behind it.

Here is what the insurers were pricing. Before this war, war-risk cover for a Gulf voyage ran at about 0.15 to 0.25 percent of a ship's hull value. By October it was three to ten percent. On a hundred-million-dollar tanker that is three to ten million dollars of insurance for a single passage. The annual war-risk layer a charterer used to buy for roughly $25,000 now costs something closer to $30,000 a week. Freight for a supertanker from the Gulf to China touched $1.2 million a day in early October, against about $80,000 a day a year earlier. On 18 October, four ships crossed the strait. The pre-war norm was about 85 a day. Brent traded at $111.44 on the morning of 22 October.

Now the part that makes this article necessary. Since 1 March, maritime trackers have counted roughly 153 security incidents in the region. One ship has been confirmed sunk: the GFS Pride, on 13 October, with all 21 crew evacuated. When Iran's naval forces claimed on 22 October that seven tankers had been damaged overnight, the independent monitoring body could confirm two.

So: one confirmed sinking, a disputed count of damaged ships, and an insurance market pricing an entire sea as if the worst were routine. The physical damage and the economic damage are not the same size, and they are not even close.

Three other things happened this autumn that belong in the same paragraph, although no news desk would file them together.

A car manufacturer in England was hacked in late August 2025, and the United Kingdom's Cyber Monitoring Centre has now assessed the total cost to the British economy at about £1.9 billion, with a modelled range of £1.6 billion to £2.1 billion. Production stopped for roughly five weeks. More than five thousand British organisations were affected. It is the most economically damaging cyber event in the country's history, and the attack itself cost the attackers something closer to the price of a secondhand car.

Across Europe, airports have been closing because of drones. The airframes involved are often worth less than a thousand pounds. One evening's disruption at Brussels Airport has been put at about €2 million. When roughly twenty Russian drones crossed into Polish airspace in September 2025, the intercepting aircraft fired missiles costing several hundred thousand euros each, from jets whose flight hours run into tens of thousands of euros an hour, against drones that cost about €10,000 apiece. The European Commission's president described using fighter aircraft against cheap drones as not sustainable. Poland has since put the cost of its planned counter-drone barrier at €3.8 billion.

And in the Baltic, a sequence of undersea cables and one pipeline were damaged between October 2023 and January 2025, mostly by ships dragging anchors across the seabed. An anchor is not a weapon. It is standard equipment that every vessel already carries. The repairs, the naval patrols and the political attention that followed were not standard at all.

2 · The Mechanism: when the bill has no relationship to the act

The first article in this series argued that a shock reroutes rather than simply spreading. The second argued that it is paid for on four clocks running at different speeds. This one is about a third property, and it is the one that makes the modern version of all this genuinely new.

The cost of causing a disruption and the cost of absorbing it have come apart.

For most of economic history those two numbers were loosely connected. Destroying a bridge cost roughly what a bridge was worth. Blockading a port required a navy, which required a state, which required a reason the rest of the world could see. The scale of the damage told you something about the scale of the effort behind it, and therefore about who could have done it and why.

That relationship has broken, and it has broken in a specific and recognisable pattern. Three conditions have to be present together.

First, the flow has to be concentrated

Something important has to pass through a narrow place: a strait, a cable corridor, a single supplier, a piece of software that thousands of firms run, an airport with one runway in use. Concentration is usually the result of efficiency, and it is almost always invisible until something interrupts it. Nobody designs a chokepoint. They are what is left when everything slower and more expensive has been competed away.

Second, interrupting it has to be cheap

Not merely cheaper than the damage, but cheap in absolute terms, so that the people who might do it are many rather than few. A €10,000 drone, an anchor already bolted to a ship, a set of stolen login credentials, a mine that costs a few thousand dollars and cannot be recovered for years. When the price of interruption falls below the price of a used car, the list of actors who can afford it stops being a list of governments.

Third, nobody can be sure what happens next

This is the condition that does most of the work and gets the least attention. A defender who knows there will be one incident can price it. A defender who does not know whether this was the first of fifty, or who did it, or whether it will happen again next Tuesday, cannot price it and so prices the uncertainty instead. That is the moment when a local event becomes a general cost.

Put the three together and you get the arithmetic of this autumn. One confirmed sinking in the Gulf, and every cargo on the route is paying as though the sea were full of them. One hacked carmaker, and the insurance renewals of several thousand companies that were never touched will reflect it. One drone over an airport fence, and a continent budgets billions for a wall it cannot see.

There is a word for what the defender is actually buying in each of those cases, and it is not protection. It is the removal of doubt, and doubt is unusually expensive because you cannot buy a little of it. You cannot insure a ship against the attacks that happen and not the ones that might. You cannot guard the airport on the nights when drones come. The defender pays for the whole distribution; the attacker pays only for the events.

One more property deserves naming, because it shapes every recommendation later in this article. The multiplier runs on confidence, and confidence recovers far more slowly than it collapses. The Red Sea is the clearest demonstration available. Container lines began diverting away from the Suez Canal in December 2023. Between February 2024 and the middle of this year, weekly container capacity through the canal averaged about 141,000 containers, roughly 11 percent of the equivalent weeks in 2023. By early October 2026 it had recovered to about 753,000 a week — a genuine recovery, the first time past half of the pre-disruption level, and still only 52 percent of it. Of the capacity choosing between the canal and the long route around Africa, 36 percent took the canal. Before the attacks began, about 80 percent did.

Nearly three years, and barely half the traffic back. Not because the ships cannot physically make the passage, but because enough people with enough at stake are still not certain enough.

Elena, the operations director at a mid-sized packaging plant in northern Italy whom this series has been following, met this mechanism three times in a single year without ever being attacked. She is a composite, built to resemble the kind of company this series is written for rather than any real firm.

In June her freight forwarder added a war-risk surcharge to resin shipments from Asia. No ship carrying her cargo had been touched. The surcharge was not a response to anything that happened to her; it was the price of a possibility, applied to everyone on the route.

In August, after explosives were found on drones at a German airport, her largest German customer sent every supplier a security questionnaire and quietly qualified a second source for the film her plant makes. Elena's firm had done nothing wrong and was not suspected of anything. The customer was not protecting itself from Elena. It was protecting itself from not knowing.

In October her insurance broker told her that business-interruption renewal terms across her sector had hardened, partly because of a very large cyber loss at a company in a different country, in a different industry, with which she has no commercial relationship whatsoever. Her premium rose because somebody else was hacked.

Three bills, none of them caused by anything that happened to her, all of them real. That is the shape of the problem, and it is why "we were not affected" has stopped being a useful sentence.

3 · Subtheme One: insurance is the machine that turns one event into everyone's cost

If you want to understand how a small event becomes a large bill, stop watching the event and start watching the insurance market. It is the fastest and most honest transmission belt in the world economy, and it is almost entirely absent from public analysis because it is dull.

Here is how it works in practice. Protection and indemnity clubs, the mutual insurers that cover roughly nine-tenths of the world's ocean-going tonnage, exclude war risk from standard cover. A shipowner therefore needs a separate war-risk policy before the main cover responds at all in a conflict zone. Those policies are typically written for seven-day voyage windows and repriced every twenty-four hours. There is no lag and no committee: a projectile hits a hull in the morning and the number changes by the afternoon, for every ship on the route, not just the one that was hit.

This is why the institutional disagreement described at the top of this article matters so much. Central Command can declare the strait open. The Joint War Committee's listed areas and the underwriters' quotes decide whether anyone sails. In a market economy, risk is priced by the people who have to pay the claim, and no military statement overrides that.

What the price does to behaviour

Elevated premiums do not simply make the same journeys more expensive. They change which journeys happen. Smaller operators drop out first, because a $3 million premium is a different proposition for a fleet of four ships than for a fleet of four hundred. Cargo consolidates onto fewer, larger, better-capitalised carriers. Spot cover becomes hard to obtain at any price, which pushes trade toward long-term contracts and squeezes out exactly the flexible, opportunistic shipping that normally absorbs a disruption.

The same compression shows up on land. When sector-wide cyber losses harden renewal terms, the firms that drop cover are not the ones with the best security; they are the ones with the thinnest margins. Insurance withdrawal is a quiet consolidating force, and it operates on a timescale of weeks.

When the state becomes the insurer

There is a predictable next step, and 2026 has produced a clean set of examples. When private cover becomes unaffordable for cargo a country cannot do without, the government steps in. India has established a maritime insurance pool of about $1.5 billion, backed by a sovereign guarantee of roughly $1.4 billion. South Korea has allocated funds to subsidise war-risk premiums for nine smaller shipping firms. Japan has moved legislation to subsidise crude import costs and cover hull and cargo claims.

Each of these is sensible in isolation. Together they mark a structural change: risk that the market declines to carry is being transferred onto public balance sheets, in the same years those balance sheets are absorbing higher defence spending and rising debt interest. The insurance gap is being closed by the taxpayer, which converts a shipping problem into a fiscal one, and that is where it will still be long after the strait reopens.

Which brings us to the most useful precedent available. During the Tanker War of the 1980s, somewhere between 450 and 546 commercial vessels were attacked in the Gulf — a vastly higher count than anything seen this year. A ceasefire was reached in 1988. Insurance rates did not normalise until 1989, because mines remained on the seabed and nobody could say with confidence where they were. The premium outlasted the war by more than a year, and it was the uncertainty, not the explosives, that kept it there.

4 · Subtheme Two: not knowing who did it is the weapon

The second half of the mechanism is harder to price and harder to discuss, because it concerns an absence. If the cost of an action has come apart from its consequences, the thing that holds them apart is uncertainty about who acted, and whether they will act again.

The Baltic Sea offers the clearest record. Between October 2023 and January 2025 a series of undersea cables and one gas pipeline were damaged, in most cases by a commercial ship dragging its anchor along the seabed. Four vessels came under suspicion. Consider what happened to each.

One left the region and reached Russian waters before investigators could act. In a second case, the flag state declined to allow a boarding and permitted only a brief supervised visit before the ship sailed on. A third was boarded and detained by Finland. A fourth was seized by Sweden, and the investigation was then dropped: the prosecutor could not prove the crew had released the anchor deliberately. The ship and the crew were released.

Note the shape of that last one. The anchor was on the seabed. The cable was cut. The damage was real and expensive. And none of it could be shown to be a decision rather than an accident, because dragging an anchor is something ships genuinely do by mistake.

Why a deniable act costs more than an obvious one

An attack you can attribute has a natural ceiling on its cost, because it produces a response. Sanctions, retaliation, a court case, a diplomatic consequence. The attacker weighs that against the gain. An act that cannot be attributed produces no proportionate response — nobody knows what would be proportionate, or against whom — and so it can be repeated. The defender, facing something that might be hostile and might be seamanship, has to defend against the hostile version while treating every passing ship as innocent.

This is why the defensive numbers look the way they do. Romania reported more than forty airspace incursions in 2026 by June, and the number has grown since. Each individual incursion is ambiguous: navigation error, stray munition, deliberate probe. The responses cannot be ambiguous. Jets are scrambled, airports close, missiles are fired. In the Polish case of September 2025, the incursion probably cost its originator under €250,000; the response cost millions, and the long-term answer now carries a €3.8 billion price tag.

Nothing about that exchange requires anyone to win a battle. It only requires one side to be unable to tell the difference between an accident and an attack. And the second-order effect is the one that should worry planners most: a defender who cannot attribute cannot deter, and a defender who cannot deter has to keep buying protection indefinitely.

The uncomfortable corollary

There is a version of this that applies to ordinary commercial life, and it is worth stating plainly because it is where most readers will actually meet it. When your customer sends you a security questionnaire after an incident somewhere else, they are not accusing you. When your insurer raises your premium after a loss in another industry, they are not pricing your controls. In both cases they are pricing what they do not know, and the only way you can reduce that price is by making yourself easier to be certain about.

That is a genuinely actionable finding, and it is the opposite of what most risk communication suggests. The competitive advantage is not in being safe. It is in being demonstrably safe, in a form someone else can verify quickly. Firms that can answer the questionnaire in two days keep the contract; firms that take six weeks lose it to someone who could.

5 · What most analysis gets wrong

The first error is counting damage instead of consequences. One sunk ship, two confirmed strikes against seven claimed: by any physical measure the Gulf in 2026 is a far quieter place than the Gulf of the 1980s, with its five hundred attacked vessels. By every economic measure that matters to anyone shipping cargo, it is worse. Analysis that counts hulls will keep concluding that the disruption is overstated, and will keep being wrong about the invoices.

The second error is treating the insurance market as a thermometer rather than an engine. Premiums are widely read as a measurement of danger. They are also a cause of outcomes: they determine which operators sail, which cargoes move, which firms stay insured. A rise in premiums is not a report on the level of risk; it is an event in its own right, with its own consequences, and it arrives weeks before the freight data shows anything.

The third error is waiting for attribution before acting. A great deal of policy and corporate decision-making is paused pending an answer to "who did this?" In the Baltic cases, that answer never arrived, and the practical effect of waiting for it was to leave the exposure in place for another season. Where attribution is structurally unavailable, it cannot be a precondition for action.

The fourth error is the one this whole series keeps meeting. None of these events happened in isolation, and none of them was absorbed in isolation. The same European budget year that must fund a counter-drone programme must also fund rearmament, pay rising debt interest and absorb an energy shock. The same mid-sized manufacturer meets a freight surcharge, a customer's security audit and a hardened insurance renewal in the space of five months. Each one is survivable. They arrive together, they draw on the same cash and the same management attention, and nobody is scoring them as one event.

6 · Three paths from here

Three broad paths for the coming twelve months, in terms of how the gap between cheap actions and expensive consequences develops. The probabilities add to 100 and are our judgement, not the output of a model. Each comes with the observation that would show we are wrong.

BASE · 50%  ·  The premium becomes the new normal.  Hormuz reopens unevenly. Physical incidents decline. War-risk pricing eases but settles well above pre-war levels, as it did after 1988. Counter-drone and infrastructure-protection programmes proceed as announced and become permanent budget lines. Confidence in the Red Sea keeps recovering slowly without returning to pre-2023 levels. The cost of this period is absorbed as a higher operating baseline rather than reversed. Proved wrong if Gulf war-risk premiums fall below 1 percent of hull value within three months of a reopening.

STRESS · 35%  ·  A second weak point gives way while the first is still open.  An incident with mass effect lands in Europe — a prolonged outage attributed to infrastructure sabotage, an airport closed for days rather than hours, a cyber event larger than the £1.9 billion case — while Gulf pricing is still elevated. Insurance capacity tightens across several lines at once, governments extend public backstops further, and the smallest operators exit. Proved wrong if Europe passes the winter without a single infrastructure incident producing a disruption of more than 24 hours.

RELIEF · 15%  ·  Doubt gets cheaper.  Attribution improves materially — faster public forensics, a successful prosecution, a credible deterrent response — or a durable settlement removes the Gulf premium. Either would compress the multiplier directly, because both reduce the uncertainty the defender is paying for rather than the incidents themselves. Proved wrong if a year passes with no formal state attribution for any European infrastructure incident.

7 · Forecast, 1 year

Claim.  Weekly container capacity transiting the Suez Canal will exceed 75 percent of its 2023 level at some point before 31 December 2027.

Probability.  0.45  ·  Confidence: Low.

Reasoning.  Recovery is real and accelerating: from about 11 percent of the 2023 level through 2024 and much of 2026, to roughly 52 percent in early October, with changes to sixteen services announced since July. But the remaining gap is a confidence problem rather than a capacity problem, and confidence has been the slowest variable in this entire crisis. Reaching 75 percent requires the large alliances to commit their main services back, which they will not do while the Red Sea threat picture stays live.

Second-order effect.  A full return would release a meaningful amount of effective shipping capacity, since the long route absorbed roughly a tenth of global container demand. That would push freight rates down and remove one of the inflationary pressures described in Article 2.

What would weaken it.  A renewed attack on a container ship in the Red Sea, or a reversal by any of the three largest alliances.

Verifier and date.  Clarksons Research weekly Suez container capacity, against the equivalent weeks of 2023. Resolves 31 December 2027.

8 · Forecast, 3 years

Claim.  Before 31 December 2029, a European government will formally attribute damage to subsea infrastructure in European waters to a state or state-directed actor, or secure a criminal conviction for it.

Probability.  0.40  ·  Confidence: Low.

Reasoning.  The record so far is four suspect vessels and no proven case, with one investigation dropped specifically because intent could not be shown. Against that: surveillance of the Baltic has improved substantially, the legal groundwork is being laid, and governments are under growing political pressure to name someone. The obstacle is evidential rather than technical. Proving that an anchor was dropped deliberately remains very hard, and that is precisely why the method is used.

Second-order effect.  A single successful attribution would change the economics of deniable sabotage more than any amount of patrolling, because it would restore the link between an act and a consequence.

What would weaken it.  Two further incidents in the same period closing without charges or formal attribution.

Verifier and date.  An official statement by a national government or EU institution, or a court judgment, reported by at least two major wire services. Resolves 31 December 2029.

9 · Forecast, 5 years

Claim.  On 31 December 2031, at least three G20 governments will be operating a state-backed or state-guaranteed marine war-risk insurance facility.

Probability.  0.55  ·  Confidence: Low.

Reasoning.  Three were created or expanded in 2026 alone, under acute pressure. The question is whether they are dismantled once that pressure eases. Emergency facilities of this kind have a strong record of outliving their emergencies, because they acquire users, staff and a constituency, and because the risk they were created for rarely disappears cleanly. The main argument against is fiscal: these are contingent liabilities on budgets already under strain, and finance ministries dislike them.

Second-order effect.  Permanent public backstops change who bears chokepoint risk. If the state is the ultimate insurer of a trade route, the state acquires a direct financial interest in defending it, which tightens the link between commercial shipping and military commitment.

What would weaken it.  Two of the 2026 facilities being formally wound up before the end of 2028.

Verifier and date.  Government announcements and reporting by the International Union of Marine Insurance or equivalent industry bodies. Resolves 31 December 2031.

10 · Forecast, 10 years

Claim.  On 31 December 2036, the Gulf and the Strait of Hormuz will still appear in the Joint War Committee's listed areas.

Probability.  0.55  ·  Confidence: Low.

Reasoning.  This is the slowest variable available on this subject, and the one that best captures whether a temporary premium becomes a permanent feature. Listed-area status is sticky: it reflects a structural judgement about a body of water rather than a weekly threat assessment, and the Gulf has been listed through long periods of relative calm before. Against that, a genuine political settlement between Iran and the United States would eventually remove it, and ten years is long enough for such a settlement to be reached and to hold.

Second-order effect.  If the Gulf is still listed in 2036, every barrel from the region carries a structural insurance cost for another decade, which is a slow and permanent subsidy to producers outside it.

What would weaken it.  A comprehensive settlement followed by two consecutive years without a maritime security incident in the Gulf.

Verifier and date.  Lloyd's Market Association Joint War Committee listed areas as published. If the committee ceases publication, the forecast is void. Resolves 31 December 2036.

11 · Signals to watch

Six measurements that will show whether the gap between cheap actions and expensive consequences is widening or closing. Most of them move before anything appears in the news.

The Joint War Committee's listed areas.  Published by the Lloyd's Market Association and revised when the committee's view changes. Any narrowing is the single clearest sign that the premium is genuinely coming down.

Quoted war-risk rates as a percentage of hull value.  Visible through brokers and shipping press. Watch for the first move below 3 percent, and then below 1 percent. The second matters more than the first.

Suez container capacity against 2023.  Reported weekly by Clarksons and in the trade press. The purest available measure of commercial confidence, as distinct from physical safety.

Time to settle a security questionnaire in your own supply chain.  Internal to your business and almost never tracked. Rising response times across your suppliers mean the verification cost is spreading.

Formal attributions and prosecutions for infrastructure incidents.  Government statements and court records. Each one shortens the odds that deniability stops being free.

Counter-drone and infrastructure-protection budget lines.  National budget documents. Watch whether they are classed as one-off programmes or permanent spending. The second is the quiet decision.

12 · What this means for individuals

Immediate.  Before the end of November, check whether your home or travel insurance has added exclusions this year that were not there before — cyber-related causes, disruption from unattributed incidents, cancellation from airspace closures. Most people discover an exclusion at the moment they need the cover.

Build.  Over the next six months, build enough flexibility to absorb a disruption that is nobody's fault and nobody's liability: a few days' buffer in travel plans, a second route to work, savings that do not depend on a claim being paid.

Position.  Expect the cost of this period to reach you as higher premiums and narrower cover rather than as a visible event. That is what it looks like when a general risk is priced into ordinary contracts.

Watch (30 days).  Your next insurance renewal, specifically the wording changes rather than the headline price.

Avoid.  Assuming that because nothing happened to you, nothing has changed for you. This year's clearest lesson is that costs land on people who were never the target.

Why this works.  It moves your attention from events, which you cannot control, to terms and buffers, which you can.

13 · What this means for business

Immediate.  By 30 November, ask your broker three questions in writing: what changed in your war-risk, cyber and business-interruption wordings at the last renewal, what would trigger an exclusion, and what notice period applies to a withdrawal of cover. The answers are usually available and almost never requested.

Build.  Over the next two quarters, build the ability to prove you are safe, not merely to be safe. A standing pack — security attestation, continuity plan, named contacts, tested recovery times — that lets a customer verify you in days rather than weeks. In a market priced on doubt, verifiability is a commercial asset.

Position.  Treat insurance availability, not just insurance price, as a strategic variable. Identify which of your operations would stop if cover were withdrawn at short notice, and what the alternative would be.

Watch (30 days).  Your own suppliers' response times to security and continuity requests. Slowing responses are an early signal of a sector under verification pressure.

Avoid.  Scoring your exposure by what has physically happened to your goods. The three bills that landed on firms this year mostly came from events elsewhere.

Why this works.  It addresses the part of the cost that is actually within reach. You cannot reduce the number of incidents in the Gulf. You can reduce the amount of uncertainty other people have to price when they deal with you.

14 · What this means for capital

Immediate.  Before year end, identify which holdings depend on insurance remaining available and affordable, as distinct from remaining cheap. Withdrawal of capacity is a different risk from a rising premium and is rarely modelled.

Build.  Over the next quarter, build a view on which governments are absorbing private risk onto public balance sheets — marine war-risk pools, sovereign guarantees, infrastructure protection programmes — and at what scale. These are contingent liabilities that do not appear in headline debt figures.

Position.  Distinguish between businesses that are paid for reducing uncertainty — insurers with capital, security and verification providers, redundancy and continuity services — and businesses that simply pay for it. The first group's revenue is structural rather than cyclical while the mechanism described here persists.

Watch (30 days).  Marine and cyber reinsurance renewal commentary, which leads the direct market by a quarter or more.

Avoid.  Reading declining incident counts as declining risk pricing. The two have been moving in opposite directions all year. This is analysis, not investment advice.

Why this works.  It prices the mechanism rather than the events, which is where the persistent part of the cost sits.

15 · What would change our mind

Three developments would make us revise the central claim of this article, that the gap between the cost of an action and the cost of its consequences is now a structural feature rather than a feature of this particular crisis.

First, if Gulf war-risk premiums returned to within a factor of two of pre-war levels within a quarter of a reopening, the multiplier would be far more elastic than we argue, and the 1988 precedent would be the exception rather than the rule. Second, if a European government successfully attributed and prosecuted an infrastructure case and incidents then fell measurably, deniability would turn out to be a temporary gap in law enforcement rather than a durable asymmetry. Third, if counter-drone and infrastructure protection proved cheap at scale — if the cost of defending converged toward the cost of attacking rather than running at a hundred times it — the economics would stop favouring the disruptor and most of this article would need rewriting.

16 · Bottom line

The expensive part of a modern disruption is not the damage. It is the doubt that follows it, which has to be priced by everyone on the route, in the sector, or in the country, whether or not anything happened to them. That is why one sunken ship can cost more than five hundred did forty years ago, and why the most valuable thing a business can now offer a nervous customer is not safety but proof.

Forecast record

Four forecasts, one per horizon, recorded before the outcome is known, each with a threshold, a named verifier and a date. They are published so that they can be checked against us later, including the ones we get wrong.

A caution we record rather than omit. The first and fourth forecasts both depend heavily on how the Gulf conflict resolves, so they are not independent of each other, and we count them accordingly. We also note that one of the four resolves against a private industry body rather than a public statistical agency, which is a weaker form of verification than we prefer.

Sources and the status of every figure

Every number carries a class. Measured means counted or recorded. Projected means an official forecast. Reported means stated by a credible source but not independently checked by us. Derived means calculated by us from the figures cited.

The series

Next week, Article 4: why Europe receives more of these waves at once than anywhere else, and what that does to a continent that generates comparatively few of them.