THE COMPOUNDING

Eight pressures, one result, and what it looks like by 2031 and 2036

Series I Synthesis  ·  Published 7 October 2026  ·  Analysis → Forecast → Recommendations

What this is. Series I described eight pressures in eight articles. This is not a summary of them. It asks a different question: what happens when they all land on the same person, the same company, the same country, at the same time. Some of them make each other worse. A few cancel out. One pair, running together, produces something neither would produce alone. You do not need to have read the eight articles. Everything needed is explained here from the beginning.

1. A Permit That Takes Fourteen Months

Start with something ordinary.

A woman runs a small manufacturing business. Forty employees, one factory, a product that requires a permit to be renewed every five years. The last time she renewed it, in 2021, it took four months, which is roughly what the law says it should take. She has budgeted four months again.

It takes fourteen.

Nothing dramatic happens during those fourteen months. Nobody refuses her application. No official is rude to her, obstructive, or corrupt. The people handling her file are competent and overworked, and one of them tells her, apologetically, that they are dealing with three times the volume they handled five years ago with the same number of staff.

Somewhere in month nine, the rules change. Her category of business is folded into a broader one. The change is described as simplification and for most firms it genuinely is simpler. For hers it means a distinction that used to be assessed individually is now decided by a threshold, and she is on the wrong side of it by a margin that has nothing to do with anything real about her factory.

She appeals. The appeal goes to the same office that made the original decision, which is still short-staffed, and it takes another five months.

Meanwhile her bank has been acquired by a larger one, her payment provider has changed its terms, and a fault on a cable she has never heard of took her order system offline for eleven days in March.

Ask her what is wrong and she will tell you the system is broken. She is describing something real and she has the diagnosis slightly wrong.

Nothing in her story is a failure. The permit office is doing its job. The rule change was a reasonable response to a real problem. The bank merger was a commercial decision. The cable was repaired. Every individual part of the system worked more or less as designed.

What happened to her is that eight separate pressures, none of them aimed at her, arrived in the same eighteen months and reinforced each other. That is what this document is about.

2. The Eight, in Plain Language

Series I described each of these in detail. Here is each one in a sentence, followed by what it means for somebody standing in it.

One. We took out the spare capacity, and we were right to

For thirty years, businesses and governments removed everything that was not being used. Spare suppliers, extra inventory, backup systems, idle capacity. This was not foolishness. In a stable world, holding a spare costs money every day and pays off almost never.

What went with the spares was time. When something failed, there used to be a second option sitting ready. Now there is a second option that has to be found, qualified, contracted and brought online, and that takes months.

For the woman with the permit: her single supplier is fine until it is not, and then she discovers that finding another one takes eighteen months.

Two. Making work became cheap. Doing work did not

Producing a document that requires somebody's decision has become almost free. Applications, claims, appeals, submissions, filings. One person can now generate what used to take a department.

Deciding on that document has not become cheaper, and it cannot, because a decision that carries consequences has to be made by somebody who can be held responsible for it, explained afterwards, and accepted by the person it goes against. None of that gets faster when computers get faster.

For the permit office: three times the volume, the same number of people who can lawfully sign a decision.

Three. Overloaded institutions stop noticing differences

When an office cannot keep up, it does not seize up. It economises. It moves deadlines, replaces individual assessment with categories, raises thresholds so fewer cases qualify, and quietly stops making some kinds of judgement altogether.

This works. Most people get faster service than before. The cost lands entirely on the cases that do not fit the new categories.

For her: the rule change that folded her business into a broader category, which was better for almost everyone and worse for her.

Four. Some capacity can be bought. The important kind cannot

A government can buy a computer system in two years. It cannot buy an experienced assessor, because an assessor takes ten years to make.

So institutions get sharper in the places where the missing thing was equipment and blunter in the places where the missing thing was judgement. And the same change removes the middle layer of people who used to notice when a rule was producing a stupid result, because those people were expensive and their noticing was never in anyone's job description.

For her: the person who would once have looked at her file and said this rule was not written for a business like this no longer has the discretion to say it, or is no longer there.

Five. Numbers stop meaning what they meant

A measurement holds up as long as faking it costs roughly what doing the thing costs. When producing the appearance becomes much cheaper than producing the reality, the number drifts away from what it was measuring.

The dangerous part is the direction. A measurement that has come loose does not look broken. It looks excellent. Output rises, targets are met, the dashboard turns green, and nobody has any reason to be worried.

In the best-documented case we found, an institution's own record of its error rate was ten times lower than the estimate of the people who maintain it, and it takes about two years for a single error to show up at all.

Six. Correcting a mistake takes four steps, and they add up

For any system to fix a mistake, four things have to happen. Somebody has to notice. The information has to reach a person with the power to act. That person has to be able to change the rule rather than just the individual case. And the change has to be carried out.

Those four happen one after another, so the times add. A system that notices instantly, escalates instantly and decides instantly, but takes four years to implement, takes four years.

The correction times we could measure ran from about twenty-one to thirty-six months, and in every single case the fix was a postponement or a narrowing rather than a repair.

Seven. The state is coming apart into separate services

Money, identity, education, dispute resolution and record-keeping used to arrive from one place because there was no affordable way to deliver them separately. Now several of them can come from anywhere.

This is genuinely better in some ways. It is also a quieter trade than it looks. Your old provider owed you things it could not withdraw, gave you somewhere to complain, and could not decide you were no longer worth serving. Your new providers owe you what is in the terms of service.

For her: her bank changed hands and her payment terms changed with it, and nobody was obliged to ask her.

Eight. Everything separate runs on the same few cables

Here is the part almost nobody looks at. Her bank, her payment provider, her cloud software and her logistics platform are four different companies in three countries with four different regulators. Underneath, they run on the same fibre, the same power, and a handful of the same data centres.

There are around two hundred faults a year on the world's undersea cables. That number has barely moved in a decade, and the failure rate per kilometre has actually improved. What has changed is how long a repair takes: the median is now about forty days, more than double what it was ten years ago, and the longest single repair on record took 947 days.

For her: eleven days offline because of a cable she had never heard of, operated by a company she has no relationship with, which owes her nothing at all.

3. What All Eight Are Actually Doing

Line those eight up and something becomes obvious that is not obvious inside any one of them.

Six of the eight produce the same thing. Not more failures. Longer waits.

So the thing that is getting worse is not how often things break. It is how long they stay broken.

This distinction is not academic, and it changes what is worth doing.

If the world were becoming more fragile, the sensible response would be to prevent failures. Spend on reliability, reduce risk, make things less likely to go wrong.

But failures are not becoming more common. The reliability engineering is working. Cables fail less often per kilometre than they used to. What has changed is that when something does fail, everybody waits longer, and more things are waiting on each other.

Which means most prevention spending is aimed at the wrong target. The return sits somewhere else: in recovering faster, and in being able to keep going while you wait.

Why nobody notices this happening

A rising failure rate makes headlines. Something breaks, then something else breaks, and people start asking why.

A lengthening recovery time makes nothing at all. There is no day on which repairs got slower. There is only, eventually, a particular outage that lasts six weeks instead of two, and everybody experiences that as a bad outage rather than as a symptom of anything.

You cannot see a trend in recovery time from inside a single event. You can only see it in the numbers over a decade, and almost nobody publishes those numbers, which is a problem we will return to.

4. The Loop That Feeds Itself

Now the part that could not be seen from inside any single article, because each article held one link of it.

Four of the eight pressures form a circle. Follow it round.

Round the circle once

More arrives at an institution than it can properly handle, because producing a case got cheap and deciding one did not.

So the institution economises. It postpones, it replaces individual judgement with categories, it raises thresholds. This is the only move available: the thing it needs is trained people, and trained people take ten years, and the backlog is now.

But when you replace judgement with categories, you remove the people who were exercising judgement. And those people were doing a second job that was not in their job description. They were the ones who noticed when a rule was producing a result nobody intended. A caseworker looking at a file and thinking this cannot be right is a detection system, and nobody has ever budgeted for it as one.

With those people gone, problems take longer to surface. The institution now learns about its own performance mostly from numbers, and the numbers are produced by the process being measured. Errors go unnoticed for longer.

Which lengthens the time it takes to fix anything, because noticing is the first of the four steps and the other three cannot start without it.

And a slower institution is worse at spotting that it is falling behind. So the gap between what arrives and what can be handled gets wider.

Which is where we came in. The circle closes, and it does not need anyone to keep pushing it.

Nobody did anything wrong

Every step is a sensible decision made by a competent person for a defensible reason.

Postponing a deadline you cannot meet is better than missing it. Replacing individual assessment with a category is better than a queue that grows forever. Reducing headcount in a function that produces no measurable output is what any organisation under budget pressure does.

The loop is not in any of the decisions. It is in the order they come in, and no one person is standing where the order is visible.

Where you can already watch it running

The clearest case we found in Series I: a major piece of technology regulation was written, given a two-year lead time, and then postponed by sixteen months because the authorities meant to enforce it had not been set up and the technical standards were unfinished. The rules were not abandoned. There was simply nobody able to apply them.

The same law contained a provision specifically designed to help the regulator learn — supervised environments where companies could test new systems and regulators could find out what they did not know. Twenty-seven countries were given two years to set one up.

When the deadline was extended, one country out of twenty-seven had one working. The mechanism built to make the system smarter was blocked by exactly the shortage that was making it slower.

Why this will never feel urgent

Each trip round the circle takes two to three years. A decade contains four of them.

Something that compounds on a three-year cycle never produces a year in which the change is big enough to demand attention. Each year looks like the last one, slightly worse, in a way easily explained by circumstances. By the time the accumulated difference is obvious, the process has been running for a decade and its causes are ten years in the past.

This is the ordinary fate of slow problems. It is not a failure of anybody's intelligence.

5. Why the Gap Between Countries Is Widening

There is one way out of that circle, and it explains why some institutions are getting visibly better while others are getting visibly worse.

The escape route

If the thing an institution is short of can be bought rather than trained, it does not have to economise. It can buy the capacity and handle the volume.

A digital identity system can be procured in two to four years. A registry can be cleaned up. Computing capacity can be contracted. None of that needs a ten-year training pipeline, and all of it can be done with money and a decent project team.

So institutions facing a purchasable shortage escape the loop. Institutions facing a training shortage do not. And the same government can be doing both at once in different departments, which looks like incoherence and is not.

But the escape route requires something you may not have

Buying capacity means buying it from somewhere. Computers need electricity and a grid connection. Identity systems need registries that agree with each other. Machine assessment needs data that is actually machine-readable, and legal authority to use it.

Those four things are complements, not substitutes. Excellent registries with no identity system cannot be acted on. An identity system sitting on top of records that contradict each other will confirm who somebody is and then be unable to work out what they are entitled to.

Which means capability does not arrive gradually. It arrives all at once, when the last of the four pieces is in place, and not at all before that.

And this is why the gap grows rather than closes

The countries that can buy their way out are the ones that already had the infrastructure. Buying more increases their demand for it, which makes it more worth building there. The countries that cannot keep economising, which lowers their demand, which weakens the case for building anything near them.

The same input is both the entry ticket and the prize.

Twenty-seven countries with the same law, the same deadline, the same design and two years of notice produced one working result. Political system does not explain that, because political system was roughly the same across all twenty-seven. Wealth does not explain it well either. What varied was whether the plumbing was there.

What this does to the way countries get compared

Almost all comparison between countries uses categories that have nothing to do with this. Type of government, income level, governance rankings, corruption indices. All of those measure real things and none of them measures registries, identity infrastructure or grid capacity.

So a well-regarded country with fragmented records will get blunter, and a less-regarded one with clean records and a working identity system will get sharper, and every standard ranking will have them the other way round.

6. Where the Responsibility Goes

The third connection is not a circle and not a widening gap. It is a leak, and what leaks out is the question of who owes you anything.

Follow one function as it moves

Start with something the state used to provide, and watch what happens at each step.

While the state provided it, you were owed things that could not be taken away by a decision. Due process. A reason for a refusal. Somewhere to appeal. And the provider could not decide to stop serving you, or leave the country, or conclude that you were not commercially interesting.

Then the function moves to a private provider. Now you have a contract. It can be revised with notice. Your appeal is to the company, and after that to a regulator, possibly in another country, who has no duty to you at all if you do not live there.

Then the private providers, all of them, run on the same shared infrastructure. And the infrastructure owes you nothing whatsoever. A cable operator has obligations to its customers, which are telecoms companies, and none to the population whose connectivity depends on the cable. There is no appeal against a decision about where to lay capacity, because it was never a decision about you.

So the responsibility thins at each step, and it disappears completely at the last one — which is also the step where the most things depend on the fewest components.

This was not anybody's plan

No one decided that the most critical layer should be the least accountable one. It is what you get when separate services quietly converge on shared plumbing, one commercial decision at a time, over twenty years.

And nobody is watching the whole thing

Every layer has a regulator. That regulator supervises the companies operating on the layer. None of them supervises the layer itself, and none has a mandate that crosses layers.

So the concentration is visible to nobody. Series I ended up with three empty notebooks for exactly this reason. Nobody publishes how often automated decisions are wrong. Nobody publishes how long regulators take to fix a bad rule. Nobody publishes which services fail together.

In all three cases, the measurement would embarrass whoever would have to pay for it. Three unconnected fields producing the same blank space for the same reason is not a coincidence.

7. The Person Caught Between Two of Them

Some of these pressures do not interact in any clever way. They just land on the same person from different directions, and the person cannot tell them apart.

Squeezed from both sides

Institutions are recognising fewer distinctions. That makes an unusual situation more expensive, because the mechanism that used to look at your particular circumstances is the mechanism being removed.

At the same time, functions moving out of the state bundle are shedding the obligations that used to protect exactly those situations. Due process, a duty to give reasons, a route of appeal.

These two are unrelated. They come from different causes, they are decided by different people, and neither has anything to do with the other. They arrive at the same address.

If your circumstances are ordinary, you will experience the next decade as steady improvement, and you will be right about your own experience. If they are unusual in any way, you will find doors closing that were open in 2026, with no single decision to point at and nobody to ask.

This is the clearest thing Series I found about who pays, and it appears in none of the eight articles, because each of them held one half of it.

And the two clocks that queue

The second overlap is simpler and affects organisations more than individuals.

Institutional waiting and physical waiting are the same problem in two layers. A firm waiting eighteen months to qualify a replacement supplier, and forty days to have a cable repaired, is not waiting eighteen months. It is waiting eighteen months, with the possibility of a forty-day interruption at any point inside that, which is a worse shape than either number suggests.

Delays do not overlap helpfully. They queue.

8. What Is Getting Better

An account that finds only deterioration has been assembled rather than observed, and this one would be dishonest without this section.

Four things are genuinely working.

Engineering keeps improving reliability. Cables fail less often per kilometre than they did, because of better surveys, deeper burial and better armour. Most technical systems are more reliable than they were a decade ago, not less.

Common standards genuinely reduce friction. Ten countries converged on broadly the same rules for a new kind of money inside two years. Twenty-seven agreed on one specification for digital identity. That is real coordination and it did not have to happen.

Licensing puts obligations back on functions that left the state. When a payment rail appeared that no government had built, governments did not ban it. They licensed it, with rules about reserves, redemption and disclosure. The function left and some of the accountability came back.

And simplification really does help most people. A firm clearly inside the new categories gets a faster answer and a shorter form. That is not a rhetorical concession; it is why simplification keeps happening and why arguing against it is politically hopeless.

The pattern in all four

Each of these is strongest exactly where the problem is weakest.

Engineering improves reliability in systems that were already reliable. Standards coordinate the countries that were already capable — one specification, twenty-seven countries, one working result. Licensing protects people in places that already have working supervision. Simplification helps the case that was already straightforward.

None of them reaches the edge of the distribution, and the edge is where all the compounding lands.

So both things are true at once. The middle is getting better and the edge is getting worse, and every aggregate statistic will report the first. That is not a failure of the statistics. It is what an average does.

9. The Number That Decides It

If the thing getting worse is how long recovery takes, then one simple comparison determines whether any system stays healthy.

How long it takes to recover, against how long you get between problems.

Shorter recovery than the gap between events, and you return to normal each time. Your condition is stable. Every crisis looks survivable because every crisis was.

Longer recovery than the gap, and the next problem arrives while you are still dealing with the last one. Each event starts from a worse position than the one before.

There is nothing clever about this. It is how a person recovering from illness works, or a company's cash position, or a country's finances. What matters is that Series I found both halves moving the wrong way.

The recovery side is clearly rising

Every duration we could measure has lengthened or stayed stubbornly long. Repairs, detection, correction, requalification, training. That is section 3 and it is the better-evidenced half.

The gap between problems is probably shrinking, for a reason that is easy to miss

The number of underlying incidents is not obviously rising. But the number of disruptions any one organisation experiences is, and that happens without a single extra incident occurring.

Because everything runs on shared infrastructure, one cable fault surfaces as four separate service failures for the same company. And because bad rules take two to three years to fix, the same underlying problem keeps producing new incidents in the meantime.

So the gap between the problems you experience narrows while the number of actual causes stays flat. That is the worst combination available, and it is invisible in any incident count.

What this looks like in one company

Return to the manufacturer. She has an eighteen-month path to a replacement supplier, sixty-two days of stock, one logistics route, and a permit renewal that officially takes four months and actually takes fourteen.

Any one of those on its own is survivable. Stock covers a supplier gap for two months. A route can be changed. A permit eventually arrives.

Two of them overlapping is a completely different situation. Stock run down by a shipping disruption no longer covers a supplier failure. A permit that arrives ten months late collides with a loan condition written on the assumption of four.

She has not made worse decisions than a company that survives comfortably. She has had the same kind of events arrive slightly closer together — and the spacing between events is not something anybody plans for, because it appears in no risk assessment ever written.

And the crossing point produces no signal

This is the part worth sitting with.

There is no day on which a system moves from the safe side to the unsafe side. Nothing happens. There is no incident, because the change is not an incident. It is a change in whether the next event lands on a system that has recovered or one that has not.

You find out which side you were on from the second and third events in a cluster, not the first. Which means the useful preparation is not for a bigger shock. It is for a second one arriving before you are done with the first.

10. Five Years Out — to 2031

Five years is where this is most useful. The loops described above run on a two to three year cycle, so five years contains about two turns of each. The mechanisms are established, the evidence is recent, and confidence is moderate.

The probabilities below are the ones already recorded in the eight articles, each with a stated threshold and a date on which it can be checked.

Institutions

Simplification becomes the normal response, not the exception — 70%

Expect more rules narrowed rather than repaired. Higher thresholds, broader categories, longer transition periods, and each change presented as reducing burden or improving competitiveness.

Some of it will be genuine deregulation on the merits, and some will be an institution that cannot cope narrowing what it has to handle. From the outside these look identical, and there is one reliable way to tell them apart: ask whether the obligation survived with a new date and a smaller reach, or whether somebody argued that it should not exist and won.

The practical consequence for anyone running anything: whether a rule still applies to you in three years becomes a more important question than what the rule currently says.

Machines start making the routine decisions — 55%

Not because the technology arrives, which it already has, but because the alternative is a queue that never clears. Expect it first in the places where the facts are structured and the answers are unambiguous: routine tax, standard renewals, licence extensions, customs classification.

The limit will not be capability. It will be whether the reasoning can be reconstructed afterwards well enough to survive a challenge, because a decision nobody can explain is not a decision anybody can defend.

And the human staff will move to the exceptions, which is where the hard cases are, staffed by the people the easy queue no longer needs. Automating the simple half of a workload does not halve it. It concentrates the difficulty.

Numbers and measurement

The same measurement problem is documented outside research — 55%

Cheap submissions, an overwhelmed checking process, and a recorded failure rate far below the real one. Insurance claims, credit assessment, benefits and procurement are the obvious candidates.

This forecast is about somebody publishing a number, not about the problem starting. The problem does not need research-specific conditions and there is no reason to think it is confined there. What research has is a public record of its own corrections. The others have no number, and no number reads as no problem.

Physical infrastructure

An outage is officially blamed on a shared layer — 65%

Several unrelated services fail at once, somebody investigates, and the report names the cable, the settlement system or the data centre region rather than describing four coincidental incidents.

This single event would do more than a decade of argument, because it converts something abstract into an incident with a report attached. Every infrastructure regime in history was built that way, and it is not a satisfying way to build one.

Governments call infrastructure critical without paying for redundancy — 60%

Designation with reporting and security duties, and no funded obligation to build a second path. Reporting costs a government nothing and passes the burden to the operator. Redundancy costs money and produces something that earns nothing on a normal day.

Materials and power

Concentration in processing holds or increases — 70%

The bottleneck for critical materials is not in the ground, it is in refining, and refining capacity is built rather than found. One material in the set moved the other way over two years, with deliberate government support in more than one country, and produced a modest improvement. That is the realistic shape of a reversal: possible, slow, expensive and not automatic.

For at least one material the honest position is that there is no alternative route at scale this decade at any price. That is a constraint rather than a cost, and the two get discussed as though they were the same thing.

Electricity keeps limiting computing — 70%

Grid connections, transformers and cooling continue to constrain the build-out more tightly than demand does. Governments wanting machine capability compete for the same connections as commercial data centres, and neither can be hurried with money.

Who gets served, and who does not

The gap between the well-served and the rest becomes visible in ordinary statistics — 60%

People and firms with documents, assets and technical capability move to the better services. The old universal service is left to those who could not move, funded by fewer of the customers who used to subsidise it.

It will look unglamorous: offices closing, waiting times growing on the phone line, staffing cut where a digital route exists. Nobody will announce that a service is being withdrawn from anybody. It will simply become harder to use, one budget at a time.

What could change this picture

Three things would redirect it, and none is far-fetched.

—   One outage attributed clearly enough and severely enough to force somebody to start measuring shared dependencies. This is the fastest route and it requires an accident.

—   Automatic review clauses written into new rules as a matter of course — a date on which a rule must be re-examined whether or not anyone wants to. Almost free when drafted in, impossible to add later.

—   A serious expansion in trained assessors in two or more major countries, counted in qualified people rather than in budget announcements. This is the only intervention that fixes the cause rather than the symptoms, and it takes a decade.

An accident, a drafting habit, and a ten-year investment. They are listed in descending order of likelihood and ascending order of usefulness, which is the shape most structural problems have.

Put together

A world that works, and works measurably better for most people, while working measurably worse for those at the edges. Faster answers for standard cases. Coarser categories. Fewer people with the discretion to make an exception. More decisions made by machines that can explain themselves. Longer physical recovery times. And, probably, the first official admission that separately owned services share a single point of failure.

Nothing in the five-year picture is dramatic. That is the finding, not the reassurance. Nothing in this series has ever suggested that a crisis is required for old strategies to stop working.

11. Ten Years Out — to 2036

Before anything else, an honest statement about how much this is worth.

Confidence at ten years is low, and the mechanisms are the only thing carrying it. The loops turn every two to three years, so a decade holds three or four turns — enough for real compounding, and also enough for one unforeseeable event to redirect the whole thing. Everything below assumes no such event, which over ten years is itself an unlikely assumption.

We put numbers on it anyway. A number can be checked and be wrong in public; a paragraph describing possibilities cannot.

Institutions split into two kinds

Capability separates by infrastructure, not by type of government — 55%

Two countries with similar politics and different records diverge. Two with different politics and similar infrastructure end up able to do the same things, whatever they choose to do with them.

Both groups will be slow at correcting themselves, arrived at by opposite routes. The sharp ones removed the people who used to notice problems in order to become sharp. The blunt ones are slow for the reason they were always slow.

Comparative analysis will keep using political categories, because that is what the available data describes, and will keep getting this wrong in the same direction.

Infrastructure carries more and recovers no faster

Recovery time stays where it is or worsens — 55%

Reliability per unit keeps improving, because engineering improves what engineering can reach. Repair speed does not, because it depends on ships, crews, access and the economics of a thin maintenance market, none of which is a technical problem.

Regions that fund redundancy improve. Regions with one path do not, because every tool available for fixing a global layer is national or regional. Global averages will show improvement driven by the places that were already fine.

Nobody starts measuring

The measurement gap does not close — 60%

No international body publishes how long regulators take to correct themselves, how often automated decisions are wrong, or which services share a failure point.

This forecast runs against our own interest and we are stating it anyway. If it turns out right, the central things this series has been arguing about remain unmeasured for a decade, which makes the argument much less useful while confirming its diagnosis of why the measurements do not get built.

Two things that might happen and probably will not

Checking separates from the checked — 45%

Verification funded separately, reporting separately, and structurally impossible to cut by the body it embarrasses. The auditor model, extended into areas that currently certify themselves. The logic is sound; the timing is not, because this kind of separation historically follows a scandal rather than an argument, and scandals cannot be forecast.

A government accepts privately issued money for taxes — 35%

The sharpest available test of how far the money function has actually moved. Accepting private money for public obligations gives up the last thing that made sovereign money different, and hands the issuer a claim the state has agreed in advance to honour.

The three domains at ten years

For institutions: two populations rather than a range. One deciding at fine resolution and correcting slowly, one at coarse resolution and correcting slowly. Long correction times in both, reached by opposite paths.

For infrastructure: carrying more of everything than in 2026, more reliable per unit, no faster to repair, with redundancy improving where it was already adequate.

For a person: more services bought from companies, more of them running on the same few systems underneath, fewer obligations owed in any of them, and better service on an ordinary day. Somebody whose life fits the standard shape will look back on the decade as improvement and will be describing their own experience accurately. Somebody whose life does not will find that the accommodations available in 2026 are gone, without ever having been withdrawn by anybody.

Two things this forecast deliberately does not say

It does not predict a crisis. Nothing in these mechanisms produces one, and a decade-long argument that arrived at collapse after eight articles that each found adaptation would have been written backwards from a conclusion.

And it does not say which specific countries end up on which side. The deciding factor is infrastructure, which is measurable in principle and unmeasured in practice. Any list would be built from three examples per country and would look far more authoritative than it deserves. The method for assessing one system is in the articles. The league table is not, and leaving it out is a decision rather than an omission.

12. How to Tell If We Are Wrong

The eight articles each carry their own conditions. Three findings would break this synthesis specifically — the claim that these pressures feed each other rather than merely happening at the same time.

—   Recovery times fall in two or more of the areas measured while the pressures are still operating. The whole argument says they share an output. If the output moves the other way while the inputs remain, the connection is not real.

—   Institutions that buy their way to sharper capability also get better at noticing their own mistakes. The circle in section 4 depends on the opposite. If capability and self-correction rise together, the chain breaks at the third link and never closes.

—   The four things that are working start reaching the edge of the distribution rather than the middle. Most of the practical advice here rests on the claim that they do not.

And one honest word about this format. A synthesis is the easiest kind of document in which to build a pattern out of material chosen to fit it. Eight things noticed separately and later found to connect is a stronger claim than eight things described in order to connect, and only a reader can judge which this is. The connections in sections 4 to 7 were worked out after the eight articles were written, from material collected for other reasons. That is the best defence available and it is not a proof.

13. What to Actually Do

Each article carried its own advice. This is not a restatement of it. This is the short list of things that change once you take the connections into account.

If you are an individual

Prepare for the second problem, not the bigger one. Everything above says the change is about events arriving close together, not about events getting worse. A reserve sized for one disruption is sized for the world that is ending. What matters is duration: enough cash, documents, offline capability and alternative routes to get through a second thing arriving before the first is sorted out.

If anything about your situation is unusual, sort it out while you still can. Section 7 describes the one group being squeezed from two directions at once. Categories are getting narrower and obligations are getting thinner, and both move one way only. Whatever flexibility your situation currently gets is more available today than in three years, and it will not be taken away in a way you can appeal against.

Count what things share, not how many you have. Two banks that clear through the same system are one bank. Two internet connections over the same cable are one connection. This correction appears in all eight articles in eight different fields and it applies just as directly to your own arrangements. The only question that predicts what fails together is what sits underneath.

If you run a business

Stop planning around how likely a failure is and start planning around how long it lasts. Most continuity planning estimates probability and quietly assumes that restoration is quick. The evidence says probability is stable and restoration is getting slower. Plan software failures in hours and physical failures in weeks. And treat a service agreement as a document about who pays, not a promise about when service returns — a penalty clause has never repaired a cable.

Work out what your suppliers all run on, before anything else. Thirty to eighty vendors usually turn out to depend on five or six things: a few data centre regions, a few network routes, a few payment systems, one or two identity providers, one grid. Every other resilience exercise measures the wrong layer until this one is done.

Put your own reviews on a calendar, not on somebody's judgement. The circle in section 4 runs inside companies exactly as it runs inside governments, and the same defence works. A review that happens on a fixed date happens. A review that happens when someone decides it is needed happens when there is spare time, which is never. This costs nothing to write in at the start and is nearly impossible to add later.

If you allocate capital

Price how long an outage lasts, separately from how likely it is. For a business carrying debt, a four-day outage and a forty-day outage are different events with the same cause, and only one of them is a covenant problem. The duration is published for several kinds of infrastructure and appears in no valuation model we have seen.

Watch scope rather than rules. The five-year view expects rules to be narrowed rather than fixed, and narrowing is proposed, argued over and published months in advance in documents almost nobody reads. That is an unusual combination: a foreseeable regulatory event with a known lead time and very few people looking for it.

Look for the shared dependency that no amount of sector or country diversification touches. Holdings across different industries and continents can rest on the same cable route, the same settlement system, the same data centre region, the same reserve asset or the same auditor. Section 6 explains why nobody is measuring this, and section 5 explains why it is getting worse.

Founder's Lens

[ EDITORIAL GATE — WRITTEN BY HAND BEFORE PUBLICATION. Never generated. Replace this marker with the founder's text, or record a suspension. ]

14. The Short Version

Eight pressures. Six of them produce the same thing: longer waits. Not more failures — the failures are about as frequent as they always were, and in several areas less so. What has changed is how long everything stays broken.

That distinction decides what is worth doing. If things were breaking more often, the answer would be prevention. They are not. The answer is being able to keep going while you wait, and getting back faster.

Four of the eight form a circle. Too much arrives, so the institution economises. Economising removes the people who used to notice problems. Without them, problems surface later. Later noticing means slower fixing. Slower fixing means the institution is worse at spotting that it is falling behind. And round again.

Nobody does anything wrong at any point in that circle. Each step is what a sensible person does about the step before. The problem is the order, and nobody stands where the order is visible.

A second connection widens rather than closes: the way out of the circle is to buy capacity instead of training it, and you can only buy it if you already had the infrastructure. So the ones who can escape are the ones who least needed to, and their escaping makes the gap bigger.

A third leaks responsibility at every step, until the layer that everything depends on is the layer that owes nobody anything.

Four things are genuinely getting better, and every one of them is strongest exactly where the problem is weakest. Engineering improves reliability that was already good. Standards coordinate countries that were already capable. Licensing protects people where supervision already worked. Simplification helps cases that were already simple. The middle improves and the edge does not, and every average will report the first.

Which is what the whole series has been converging on, and what none of the eight articles quite contained. If your situation is ordinary, the next decade will be better and you will be right to say so. If it is unusual in any way — a business that does not fit a category, a person whose paperwork is complicated, a town at the end of one cable — you will find fewer options, less recourse and longer waits, arriving from directions that were never aimed at you, decided by people who have never heard of you.

And the thing that determines whether any of it becomes serious is a comparison so simple it barely looks like analysis. How long you take to recover, against how long you get between problems. The first is rising. The second is probably shrinking. And the point where they cross produces no event at all — it just changes whether the next thing lands on a system that has recovered or one that has not.

So: prepare for the second problem rather than the bigger one. Count what your options actually share. Plan using the timescales the evidence gives rather than the ones a contract promises. And if there is anything unusual about your position that currently gets accommodated, deal with it while the accommodation is still there.

The woman with the permit will get it eventually. She will also have learned something that no one told her, which is that the fourteen months were not about her, and that planning for four was the only real mistake she made.

Forecast record

This document introduces no new forecasts. The thirty-two forecasts of Series I stand as recorded in their own articles, each with a threshold, a named source that will settle it, and a date fixed before anyone knew the answer.

The percentages in sections 10 and 11 are those same forecasts, gathered here so the five and ten year pictures can be read whole instead of assembled from eight documents.

The second row is the one that matters. Everything this series says about the value of forecasts that can be checked is, at present, a claim about a method rather than a demonstration of one. It becomes evidence when forecasts start coming due and the wrong ones are published next to the right ones. Not before.

Sources

No new evidence appears here. Every figure comes from the article that established it, where it is listed with its source and with a note on what kind of number it is — something measured, something projected, something estimated, or something we worked out ourselves.

The connections described in sections 4 to 7 are our own reading and are not attributed to anyone. They are an interpretation of material collected for other purposes. A reader who accepts the underlying figures is free to reject the interpretation, and that is the property we want these documents to have.

In this series

—   Series I, eight articles, from The World Is Becoming Less Forgiving to After the Package.

—   Series II takes the same three questions to the physical layer — energy, computing, water, materials, connectivity, logistics — one at a time, building a single comparison table.

—   The method behind the Chaos Index and this series: /methodology

THRIVE IN CHAOS

Decision Intelligence for an Uncertain World

Analysis → Forecast → Recommendations  ·  Signal → Meaning → Action → Stability

Signal Over Noise  ·  thriveinchaos.ai

AI intelligence system with human editorial oversight.

Forecasts are probability-based analytical assessments, not certainties. This material supports independent judgment and does not constitute financial, legal or investment advice.