Why the house we have built may be outrunning our ability to repair it

An old fish swims past two young fish and says, “Morning, boys, how’s the water?” The young fish swim on for a while. Then one turns to the other and asks, “What the heck is water?”

Guarding the Wrong Door, the earlier companion to this paper, argues that national security has a water problem. The water is the habitat we have built for ourselves out of accelerating technology and ever denser interdependence, and this paper calls it the house. We live in it, so we stop seeing it, the way the young fish stop seeing the water.

What we do see are the monsters under the bed, in one bedroom, in our house.

They come in familiar shapes, and are genuine threats, and mostly understood, as we’ve had experience with them before: hostile states, criminal networks (some working as proxies for those states), terrorists (some aligned with the above, some not), and cyber actors who serve any of those camps or none.  These are all in view.  We’ve been briefed on them, funded and actioned responses. The newest arrival, however, is poorly understood: misaligned artificial general intelligence or superintelligence.  But it IS in sight.  Briefings have begun, taskings made, funding and actions are being prepared. It has pulled a whole field into the work of aligning it. Each deserves the attention it gets.

When a senior executive friend asked me recently how worried they should be about “AI apocalypse”, given recent headlines, I said, “very”.  But then I added, “but that’s not what you should really be worried about”.

Because the scarier monster is not those we see under the bed.  It doesn’t even fit under the bed.  The really dangerous monster is the house itself: the habitat we have built from accumulating and accelerating technology, and the dependence that comes with it. If the causal mechanism in this paper’s hypothesis is correct, the risk of severe, compounding failure rises. Yet the house sits outside the frame, and no one is charged with measuring or managing its resilience as a whole. The more complex and tightly coupled it grows, the more likely a failure becomes, and the greater chance that failure is of catastrophic proportions. Intruders and AI do not create that condition. They read it, and some of them pile on. The monsters under the bed are real, but so is the larger one, the house.

The house has no architect. Thousands of builders, each acting for sound local reasons, add complexity faster than anyone can understand what was added, much less repair what goes wrong.

Bottom Line Up Front

The hypothesis. National recovery capacity is falling behind the increasing complexity and coupling the country now depends on. If that holds, ordinary shocks will cost more and last longer each time, with or without an adversary. The paper treats this as a hypothesis to be tested. The indicators support it, and there is no public, cross-sector measure of national recovery capacity.

An understandable primary binding constraint is the replacement chain. How far a shock travels depends on how long it takes to replace what breaks, and on whether anyone still has the skills to do it. A large power transformer takes two and a half to nearly three years from order to delivery. A grid that loses transformers it cannot replace is living on a three-year horizon whether or not anyone calls it that. We face similar situations with respect to critical rare earth products, semiconductors, diesel, many pharmaceuticals, helium, phosphate-based fertilizers, and many more.

Adversaries do not need to start the crisis. A strategic competitor can wait for one and act inside it. Even a small, capable group with an obedient AI can hypothetically find the choke points that make one worse. And there are many scenarios in between, some described in this paper. Across all of them the condition of the house sets the stakes, and we face it regardless of adversarial action.

Washington alone will not fix it. The house has no architect to petition. Public awareness is thin, federal money is scarce, and federal coordination is constrained. The recommendations put forth here therefore start where agency already exists, with individuals, families, communities, companies, and industries. The ask of Washington is to lower the cost of and otherwise enable faster actions by these ground up entities.

The first federal move is to measure. There are things Washington can do. The Office of the DNI could sponsor a 90-day assessment of the national replacement chain: a ranked list of long-lead assets, their spares, and their replacement times. The DNI is the right sponsor because the question is adversarial and cross-sector, and the Intelligence Community already ranks targets by consequence. It costs little, it informs one funding decision, and it tests the paper’s central hypothesis. Previous efforts come close to this (DOE’s transformer reviews, the EO 14017 supply chain reviews, the National Risk Register ordered under EO 14239, CISA’s non-public list of systemically important entities). It’s possible that a classified version of this exists; this work would be a vulnerability roadmap if published.  But keeping it entirely nonpublic would limit a whole-of-nation response to improving national resilience.

A house whose condition no one measures will be judged only by the intruders it attracts.

A word to builders

This is not an argument against technology or innovation. It is an argument that those who build, adopt, and advocate technology, including myself and OODA members, should ask: innovate toward what, and move faster toward what? That’s the baseline responsibility. But technologists within the OODA community need to then go beyond the organizational goals and include national recovery capacity among the outcomes they intend to improve. That responsibility may mean changing a design or deployment path; it may also mean accelerating innovation in resilience, repairability, redundancy, and human capability. The point is not to put a brake on progress, but to steer it toward a more durable kind of progress.

What to expect from this executive version

This paper does three things. First, it diagnoses the house: why a habitat built from accumulating and accelerating technology fails more easily as it grows, and which few variables decide how far a failure spreads. Second, it applies that diagnosis to eight scenarios. Each is a path that runs from the condition of the house, through an ordinary or deliberate shock, to the people who exploit it, and each ends at a stated landing. Last, it turns to what can be done, starting with individuals, families, communities, companies, and industries, and ending with a low-cost federal role and a single first move.

This is the executive version. A reference edition (working version) of more than 100 pages, is available by request through OODA, carries what this version compresses: the full scenarios with their indicators and breakpoints, the rating method, historical stress tests, country profiles, and sources. The two are kept consistent. Where this version states a judgment in a sentence, the reference edition argues it.

A senior reader should finish this version with five things: an awareness of the water and an ability to see the house as the hazard; a test for when it is in trouble; the short list of variables that decide how far a shock travels; a common measure of readiness, the rule of threes; and a first move matched to the reader’s position. The paper runs in this order:

Section	What it does
The claim	Defines normal collapse and states the four propositions the rest of the paper tests
Why the house fails	Proposes a diagnostic, complexity outrunning slack, and names the five variables that decide how far any shock travels
Who piles on	Shows how adversaries read the house, from a small group with an obedient AI to a strategic competitor holding five moves
Why institutions miss it	Explains why the least plausible events get the least preparation, and how the paper rates risk to correct for it
Scenarios and endgames	Writes eight paths, A through G and X, from the condition of the house to where each could land, then examines the two that bracket the problem
The rule of threes	Extends the familiar survival rule to three months and three years, and shows who can act at each rung
What already exists	Takes inventory of what the United States and eight other countries already do
What to do	Sets out steps from individuals to industries, the federal government’s catalytic role, and the first move
The case against	States the strongest counterargument and what survives it

The paper is written for a particular kind of senior leader, serving or retired, with standing and the freedom to act without waiting for permission. Each reader starts from a different question. Some examples:

Reader	The question	Their move
Leaders with standing and no office	Which good, underweighted idea will I stake my name on?	Convene, publish, and carry the message to those in office
Serving leaders (IC, O-6 and above, SES)	What can I do inside my own authority, with no new money?	Sponsor compound exercises (power, GPS, and networks lost together) and the 90-day replacement-chain assessment
Industry and company leaders	How many days can we run with no outside help?	Measure it, publish it, and pool spares with peers

The claim

This paper tests the argument against the United States. The scope is deliberate: a bounded system is the only way to reason about cascades.

Here we join three ideas.

• Charles Perrow showed that systems both complex and tightly coupled fail in ways no one anticipated, out of ordinary parts, and he called the result normal accidents. His cases were plants, ships, and aircraft; applying his logic to a national house is this paper’s extension, and it should be tested as one.
• Joseph Tainter defined collapse as the loss of reliable complexity, and showed that societies face it when the cost of maintaining that complexity outruns its return.
• Per Bak’s sandpile supplies the trigger: one more grain may cause a small slide or an avalanche across much of the pile, and the difference lies in the pile, not the grain. In Bak’s model each grain falls only after the last avalanche has settled.

The paper’s concern is that technological change now arrives faster than institutions, habits, and infrastructure can settle, so new disturbances land while earlier ones are still running. That borrows Bak’s insight; it does not require the country to behave like his model. The OODA Network has put a blunter version on record: “most legacy institutions will systemically fail at the hands of exponential acceleration.” This paper makes the narrower, testable claim. Normal collapse is the three together. The country keeps its flag, its currency, and its institutions, and loses the ability to coordinate, repair, persuade, and protect. The paper’s central model of that loss is de-complexification by attrition. It is slow, uneven, and human, and no single event marks it. Four propositions organize the argument.

1. The house is the primary hazard. Concentration, coupling, maintenance debt, and thinning human skill determine how far any shock travels.
2. Exploitation follows fragility. Adversaries prepare for the house’s failures, or improvise when they see them. A competitor does not have to cause a crisis. Reading one is enough.
3. Implausibility is an opportunity. Events institutions dismiss as irrational get the least preparation, which makes them most valuable to anyone willing to try.
4. The cost curve has bent. A capable private group can use its own obedient AI to read public data, find choke points, and sequence small acts. Coordination was the historical bottleneck, and aligned AI, in the wrong hands, is a looming specter standing ready to break it.

Why the house fails: a race

Perrow set no threshold for when normal accidents begin, so this paper proposes a diagnostic, to be tested and not assumed. The house is in the normal-collapse zone when interactive complexity and coupling grow faster than slack and diagnostic capacity. Slack is the spares, skilled people, time, and money that absorb a loss. Tainter explains why slack lags: each added layer costs more to maintain, so budgets drift away from it. Bak explains why it fails to refill: when disturbances arrive faster than the house can settle from the last ones, crews, spares, and attention are still committed when the next ordinary trigger lands. Complexity has added resilience in places, such as multi-region cloud and interconnected grids, so the question is net. Five indicators bear on it. They support concern and do not prove the hypothesis.

Indicator	Trend	Source	What would show the opposite
Large power transformer lead time	About 150 weeks in mid-2024, roughly triple the 2021 level	Wood Mackenzie, June 2024	Lead times falling as new factory capacity opens
Net interest versus defense spending	CBO projected FY2024 net interest at $870 billion against $822 billion for defense	CBO, via CRFB, February 2024	Interest stable or falling as a share of the budget
Cloud concentration	The three largest providers took 63 percent of enterprise cloud spending in Q3 2025, up from 61 percent two years earlier	Synergy Research, via The Register	Falling share, or measured multi-provider failover
Overlapping open disaster recoveries	FEMA manages over 600 open major disaster declarations, some nearly 20 years old. In October 2024, only 9 percent of its response workforce was available for Hurricane Milton, with staff still deployed to Helene	GAO-25-108216, March 2025	Open recoveries falling, and surge workforce available for a second event
Human recovery capacity	No routine national measure exists	None	A routine measure, with rising scores

The last row is a finding in its own right: the skills the paper cares most about are the ones nobody counts.

Five variables decide how far a shock travels. The house’s conditions are many, and the reference edition describes them in detail. Five carry most of the argument. Everything else is a symptom of one of them, an amplifier, or an exploiter: AI amplifies all five, and China’s potential moves stand to act on them. The scenarios later in the paper are combinations of these variables under stress.

Variable	What it measures	Binding in scenarios
Concentration	How much of a critical function depends on one vendor, site, region, or design	D, F, G
Replacement lag	How long a lost asset takes to replace, given spares, factories, and crews	A, G, X
Human recovery capacity	The skilled people available to run, repair, and improvise, with and without the house’s tools	B, F, G
Coordination capacity	The ability to authenticate information, decide, and move resources across institutions	B, C, E, X
Fiscal and response capacity	The ability to finance repair and absorb a shock without a disorderly market	A, C

The 2003 Northeast blackout shows the pattern. Alarm software in an Ohio control room stalled unnoticed, and over ninety minutes three lines sagged into trees left too tall. Operators without alarms misread a deteriorating grid, and the failure spread across eight states and Ontario, cutting power to about 50 million people. Every link was plausible. The path was not.

The replacement chain is the binding variable. Metcalf (2013), the Pacific Northwest (2022), and Moore County (2022) showed that quotidian actions like rifle fire and ordinary tools can damage long-lead equipment. In 2013 operators rerouted power and replaced the equipment in weeks. Since then lead times have lengthened and spares have thinned. OODA’s Daniel Pereira describes the same race in cryptography as two clocks, one technical and one institutional, and warns that “the second clock may expire first.”

The resilience gap is the human half. It is the distance between what a stricken house needs its people to do and what they can do without its tools. It predates AI. At population scale, Americans no longer practice farming, making fire, or navigating without a device, and the supporting processes, such as saving seed, went with them. Trade press reports that Ford, after record recalls in 2025, concluded it had “mistakenly” thought AI alone would deliver quality, and added, at great difficulty and expense more than 350 experienced human specialists.

Who piles on

The house fails on its own, and people notice. Some actors prepare for the moment and others improvise when it arrives.

“Aligned” AI directed by a capable human adversary is the door most planning leaves open. Public conversation guards against misaligned AI. The sharper risk is a tool that maps dependencies and replacement times, simulates cascades, and schedules many small dispersed actions. Much of this needs patience and data, and little frontier capability. A defender who runs the analysis first finds the same list. Others aim lower in the stack. Emilio Iasiello has warned that adversaries are working to poison the information AI systems learn from, which strikes coordination capacity directly.

China holds five potential moves that could attach to any crisis. The paper does not predict their use. It asks what options a strategic competitor holds when it reads a strained United States: a Taiwan quarantine or seizure timed to a domestic crisis (the Big Bold Move); a coalition of the aggrieved built through Belt and Road finance and security ties (the Cortez coalition, after Cortés, who took Tenochtitlan chiefly with the Mexica’s resentful neighbors); small, quiet moves on U.S. debt; pre-positioned access of the Volt Typhoon kind in communications, energy, and water systems; and gray-zone pressure of the sort Bob Gourley’s Dragon Falls dramatizes. In the novel, aligned AI is the defender’s tool: Gourley’s protagonist builds a virtual intelligence center from open-source AI agents. In an adversary’s hands, the same tools would make that pressure faster, cheaper, and harder to see. A competitor holding all five need not build a crisis. It can wait for one.

Why institutions miss it

Pearl Harbor, the Yom Kippur War, 9/11, and Ukraine in 2022 illustrate a recurring problem: institutions can prepare least for what they judge irrational, and that unpreparedness is the payoff to anyone willing to try. The paper calls this the implausibility premium. The paper corrects for it by rating every risk on four axes, one of which measures neglect directly.

Each mechanism and scenario carries four ratings over about ten years: likelihood, in the estimative language of U.S. intelligence; consequence, from 1 (local nuisance) to 5 (durable loss of national coordination); discounting, from 1 (widely rehearsed) to 5 (dismissed as irrational); and confidence. Together they set an attention tier. Baseline covers what is already happening. Prepare covers likely events of moderate consequence. Stress-test covers high-consequence events that are at least unlikely, and calls for exercising the compound case. Surprise watch covers high consequence with high neglect at any likelihood, and calls for red teams and warning indicators. Boundary covers catastrophic events that are very unlikely, and serves to test assumptions. Neglect justifies investigation; it does not by itself justify spending. Every rating is a judgment, and the reference edition shows how each is assigned and challenged.

Scenarios: from the house to the endgame

A scenario is a path, and the paper builds each of its eight in the same order: the condition of the house first, the initiating event second, the exploiters third. The paper writes the paths and ends each one at a stated landing.

The landings form a ladder of ten endgames, ordered by how much national coordination survives: the ability to move goods, money, information, and people, and to repair what breaks. The ladder runs from the Long Grind (years of lost output, as in COVID), Strategic Eclipse (lost primacy abroad, as Britain discovered at Suez), and the Single-City Shock, through De-complexification by Attrition, the Confidence Collapse, the Dark Month (weeks without national power), and Fragmentation, to the endings fiction favors: Gibson’s Jackpot, the Electrical Dark Age of One Second After, and McCarthy’s Road. De-complexification by attrition is the primary landing in three of the eight scenarios and an alternative in the other five. That reflects how the scenarios were built and is not a probability.

Scenario	The path	Primary landing	Likelihood and tier
A. The energy, credit, and maintenance spiral	A sustained energy-shipping or refining shock raises costs across a house already carrying debt and deferred maintenance	De-complexification by Attrition	Roughly even; stress-test
B. The pandemic of incapacity	A pathogen with repeated waves and long recovery meets a depleted health and trades workforce	The Long Grind	Unlikely; stress-test
C. The confidence crisis and the gray zone	Weak auctions, a ratings action, or a debt-limit standoff erode the dollar’s safe-asset status	The Confidence Collapse	Unlikely; stress-test
D. The Taiwan compound contingency	A PRC quarantine, blockade, or seizure of Taiwan, possibly timed to a domestic crisis	Strategic Eclipse	Blockade unlikely, seizure very unlikely; homeland dimension on surprise watch
E. The alternative operating system	Allies drift toward Chinese-centered finance, standards, and maintenance during a stretch of U.S. weakness	Strategic Eclipse	Likely in partial form; prepare
F. The deskilled house	A mundane flaw in shared AI tools fails many sectors at once, run by people who rarely operate by hand	A short Dark Month	Likely as a contributing condition; surprise watch
G. The no-bang cascade	Familiar shocks stack faster than the house can repair them	De-complexification by Attrition	Roughly even in partial form; stress-test and surprise watch
X. The Aligned Planner	A capable non-state group with an obedient AI designs a campaign around the replacement chain	De-complexification by Attrition, through a brief Dark Month	Very unlikely in core form; surprise watch

Why G and X. Two scenarios bracket the problem, and the paper examines them in depth. Scenario G needs no adversary at all, and it is the paper’s strongest claim: the case against the paper touches it least, because it rests on ordinary shocks and depleted reserves. Scenario X sits at the other edge. It is the least likely path in the set and one of the most neglected, and the counterargument presses hardest against it. The other six mix ordinary shocks with exploiters and fall between the two. The paper argues that the investments that answer G also answer X, so a reader does not need to know which future arrives in order to know what to do.

Scenario G, the house without an adversary. Everything needed is already present: maintenance debt, concentration, thin expertise, a skills gap, and a public that struggles to authenticate what it is told. Familiar shocks arrive within months, such as an energy price spike, a regional health emergency, a cloud outage, and a hurricane. Each arrives before the last is repaired, so the backlog of open recoveries grows and planning horizons shorten. No national declaration comes, because emergency management is built around declared events. Opportunists adjust, and a bold move in the Western Pacific needs no coordination, because the window opens by itself. The partial form, prolonged recovery failures in several regions from stacked ordinary shocks, is roughly even in likelihood, with discounting 4 and moderate confidence. The remedy is a national recovery-capacity metric reported as routinely as unemployment.

Scenario X, the adversary on top. X asks what a capable, patient, non-state group with an obedient AI could do to a strained house. The idea has a long lineage: in 1996, Matt Devost, Brian Houghton, and Neal Pollard argued that sub-state groups could exploit a society’s dependence on interconnected information systems at low cost and low risk. What they lacked was cheap coordination. It is a multi-vector 9/11 built around an AI-enabled vulnerability chain, and its replacement chains: a dependency graph ranked by replacement time; dispersed physical attacks of the Metcalf type on long-lead equipment; intrusions that prompt precautionary shutdowns of unbreached systems, as at Colonial Pipeline; market positioning and attempts on leaders; and a response that becomes the third wave. The landing is a brief Dark Month and then a long attrition. X-core is rated very unlikely, consequence 4, discounting 4, low confidence, in the surprise-watch tier. It is not a prediction, a plan, or an accusation, and the paper contains no target lists or operational design. The first steps against it cost little, and the same investments defeat the natural failures in Scenario G.

The rule of threes: a common measure

People already carry a rule for the body: three minutes without air, three hours without shelter in harsh conditions, three days without water, three weeks without food. The paper extends it to the house: three months without the supply system (refills, fuel, parts, payments) and three years without the replacement chain and the skills to run it. Others have extended the rule before. Survival-education sites add a three-month rung for isolation, and a 2011 SurvivalBlog post uses three months and three years as planning horizons for supply, with a rural retreat as its answer. This paper builds on that work and differs in two ways: its rungs measure what systems can deliver and restore, and it assigns the upper rungs to industry and the nation. It found no earlier three-decade rung. The rungs are design horizons and not forecasts. The three-year rung is not arbitrary: a large power transformer takes about 128 to 150 weeks from order to delivery.

Rung	What fails	Who can close it
3 minutes to 3 hours	Air, bleeding, shelter	Individuals: CPR, bleeding control, a safe place
3 days	Water, light, communication	Families and neighbors
3 weeks	Food, medicine refills, local services	Communities, companies, mutual aid
3 months	Resupply: fuel, cold chain, parts, payments	Industries, sector pools, national reserves
3 years	Replacement chain and the skills to repair	Industry, the nation, and anyone who can teach a trade

Each rung changes who can act. Households can cover the first two, and help their communities cover the third. Only industry and the nation can reach the last two, and no household can. The skill requirement also changes: at three months, low-dependency skills of the Foxfire kind (preserving food, cooking and heating without a grid, treating water, simple repair) carry people through. At three years the house needs bridge skills, the trades and teachers that rebuild what wears out. A senior executive from the technology industry told the author that his makerspace friends could make almost anything, and that the limit was whether anyone could teach it. That is the bridge-skill problem in one line.

Extending beyond three years we find a generational horizon. At about three decades, large power transformers reach the end of a design life of roughly 30 to 40 years, and a working generation retires, so the question becomes whether anyone still knows how to rebuild. The paper flags that horizon and builds no rung on it.

Fiction is imagination and not evidence, and the paper uses it as a handle for public conversation. Alas, Babylon, One Second After, and Jericho dramatize the three-month rung and mislead mainly by making the end quick and the villain clear. Station Eleven and A Canticle for Leibowitz ask the three-year question: what survives, and who can teach it? Real cases are more instructive and less cinematic: Puerto Rico after Maria, Ukraine’s winters, Hurricane Helene, Cuba’s Special Period, and the post-Soviet household gardens. In disasters, people mostly help one another, and the failures come from supply, not from character.

The future is already here: what exists in the United States

William Gibson’s line that the future is already here, just not evenly distributed, is a useful test. Pockets of the three-year house exist now, and this paper takes inventory and proposes no new program. Individuals and families have CPR and Stop the Bleed training, hunters and anglers, home solar and batteries, ham radio operators, and the long-practiced storage habits of church communities. Communities have CERT teams, Medical Reserve Corps units, and neighbor-led groups that moved first after Helene. Companies include Walmart’s emergency operations, FORTIFIED building standards, and Sheltered Harbor in banking. Industries have electric mutual assistance through the Edison Electric Institute, sector reserve pools such as STEP, Grid Assurance, and RESTORE, GridEx exercises, and health care coalitions.

The gap is as clear. Only about 12 percent of Americans have planned with neighbors. CERT is largely unfunded. The federal sharing channel CIPAC ended in March 2025, and its replacement lacks liability protection. Fifty thousand lineworkers from 41 states can restore power after Helene because 41 states are not Helene. Almost nothing reaches past the three-week rung, and no one measures national recovery capacity in total.

What other countries do

Several countries have built, for their own reasons, parts of what this paper asks for. None targets the habitat as a whole.

Country	What it does that bears on the rungs	Why it does not transfer whole
Finland	Security-of-supply agency with industry pools; households advised to cope for 72 hours	Small, high-trust, conscript, long border with Russia
Switzerland	Compulsory stocks of staples, fuel, and drugs held by firms, for about 14 francs per person per year	Small, consensus-based; stocks erode under cost pressure
Sweden	Household brochure and volunteer defense groups	Compulsory duty; very high trust
Japan	Preparedness as a school and neighborhood habit; hazard is natural	Rests on recurring disasters in living memory
Taiwan	Whole-of-society resilience committee; six-month planning target in some ministries	Urgency of blockade; island energy dependence; early stage
Israel	Protected room required in new homes since 1992	Leaves older housing exposed; millions lack standard shelter
Singapore	Shelter and rice-stock mandates tied to permits and licences; behavior tested in the field	City-state capacity; state builds the housing
Estonia and Latvia	Data embassy abroad; civilian cyber volunteers	Willingness to act lags; trust across communities

Four patterns matter. Most machinery is public-private. Mandates attach to something already regulated, such as a building permit or an import licence, so they need no new appropriation. Willingness is not action: the Baltic polling, the Swiss household gap, and Singapore’s field tests (about 5 percent of people who noticed a planted suitcase did anything) all show it. And these are small, trusting countries with a named enemy, so none can be copied whole. Japan is the closest model on motive, because its hazard is the habitat.

What to do: start from the ground up

The paper’s own diagnosis explains why. In The End of Power, Moisés Naím argues that power has become easier to lose and harder to use, as more actors acquire the means to block. In New Power, Jeremy Heimans and Henry Timms contrast power held like a currency with power that moves like a current. A house built by thousands of hands will not be repaired from one office. It will be repaired by the same many hands, if enough of them can see what they are standing in. David Bray’s account of seven people in Poland who built a 500-node acoustic warning network for about $5,000 shows what that looks like at the edge. Friedman’s line applies: when a crisis comes, action depends on the ideas lying around.

Start now, no appropriation needed

Who	Action
Individuals	Cover 72 hours. Ask the medication-clock question: how many days of each daily medicine remain? Learn one skill that works without power, and teach another person a bridge skill
Families	Write a one-page plan. Count days of autonomy for water, food, fuel, and medicine, and raise the lowest. Rehearse a weekend without power and the router
Communities	Map who needs help and who can give it. Run a compound drill that cuts power, GPS, and networks together, and publish what failed. Count makerspaces, colleges, and trade schools as resilience assets
Companies	Extend business-impact analysis past the weeks rung. Keep a manual mode. Help employees’ households cover 72 hours. Train replacements
Industries	Build mutual aid that assumes simultaneous loss. Hold shared reserves of long-lead items. Publish days of autonomy by sector

What the federal government can do to catalyze this, mostly at low cost

Function	Action
Convene	Sponsor compound exercises that assume no national backstop
Publish	Release an unclassified assessment of the national replacement chain, ranked by lead time
Message	Extend the Ready campaign past 72 hours with the rule of threes
Give legal cover	Add liability protection for information sharing, and safe harbor for paid private responders
Measure	Add duration and behavior to national surveys, and test behavior in the field
Procure and reward	Build continuity into federal purchases; encourage insurers and lenders to price resilience (a proposal, not verified practice)

The first move. The Office of the DNI sponsors a 90-day assessment of the national replacement chain, with CISA and DOE jointly accountable under a named senior lead. The product is a ranked, cross-sector list of long-lead assets with their spares, replacement times, and costed options. It enables one decision, whether to fund regional reserves and hardening in the next cycle, and it tests the paper’s central hypothesis.

The ideal end state, shelf-ready. If every veto point cleared, the national package would keep a ranked dependency graph current (CISA with DOE laboratories, high implementability), create regional reserves of long-lead grid equipment under consume-and-replace rules (DOE with utilities, medium), extend physical security beyond the post-Metcalf standard (FERC, NERC, medium), and build terrestrial backup for positioning and timing while separating Pacific surge capacity from homeland response (DoD, FEMA, Commerce, HHS, low to medium). Implementability is a judgment: High means existing authority suffices, Medium means a new program or rule, Low means legislation or major funding. Perrow adds two cautions: reduce the size of targets, and judge each safeguard, defender-side AI included, on whether it adds slack or complexity.

Two cautions apply to the federal role. Centralizing everything repeats the mistake of the builders, because a single point of command is a single point of failure. The better standard, as David Bray and Ryan McLean put it, is systems that fail “gracefully rather than catastrophically.” And a public asked to act without trust may turn to panic or suspicion, so messaging should empower and not frighten.

The case against the paper

The strongest counterargument deserves full strength. The United States has absorbed civil war, depression, world wars, terrorism, financial crisis, pandemic, blackouts, and hurricanes. Scale, wealth, federalism, private capacity, and allies are real strengths, and complexity also repairs: grids reroute and mutual aid moves crews. Each argument is true in some range. Three structural points show where it narrows. Survivorship: the record lists the shocks the country survived and says nothing about shocks it has not met. Simultaneity: most recoveries followed one dominant shock, and the paper asks what happens if stacking goes further. Resilience is a stock: spares, skilled people, slack, and legitimacy can be spent, and COVID shows both halves, with extraordinary logistics and science and more than a million American deaths.

The paper survives in a narrower, falsifiable form. It does not claim that collapse is likely. It claims that past recovery is no guarantee, that several sources of resilience are degraded in ways earlier shocks did not meet, and that the interaction of ordinary shocks, misjudgment, and opportunism is under-rated. If regional reserves rise, drills show recovery within planned times, systems are separated, and a funded dependency graph guides spares, the ratings should fall. If they are still absent in several years, the discounting ratings should rise.

Evidence and limits

This is a conceptual synthesis from public sources, not a classified assessment. Every rating is a judgment; people rewarded for disagreeing should review the surprise-watch tier. The reference edition carries the scenarios, method, inventory, country profiles, claim register, and sources.

Popular preparedness writing appears in the paper as evidence of what some households do and expect, and not as authority on risk.

The house is everyone’s and no one’s. Each reader holds one room of it. Start by measuring yours.

The house is the hazard. The monster under the bed is only the loudest thing in it.