1. A Confession, Narrowly Made

On October 23, 2008, a committee of the United States House of Representatives held a hearing. The witness was a man who had just stepped down from the chairmanship of the central bank, a seat he had occupied for eighteen and a half years.

A member of Congress asked him whether he had discovered that his view of the world was wrong.

He gave an answer that would be quoted ever after: I found a flaw.

Pressed further, he became more specific. He had believed that institutions answerable for their own profits and losses would, out of self-interest, look after the interests of their shareholders and their firms. That belief, he said, had proven mistaken.

Here was a man of eighty-two, on the witness stand before Congress, admitting that a piece of judgment he had relied on for decades did not hold up in at least one respect.

The moment itself is worth recording. Across the eighteen essays before this one, the construct has rarely spoken this way. It renames itself. It declines to ask. It draws its boundaries inward. It gives the remainder an orbit that circles the ledger without ever landing on it. It demands documentation from everyone else. It almost never steps forward to say: the scale I have been using is wrong somewhere.

But notice exactly what he admitted. He said those institutions would not protect themselves the way he had assumed — which is to say, the flaw lay in his judgment about people. He did not say that the entire practice of compressing everything into a price was itself the problem. A construct admitting it misjudged one input, and a construct admitting that its whole method of measurement is unsound, are two very different confessions.

Rarer still is the timing. This was not an admission wrung out after he had been refuted, nor one dug up decades later by historians going through the archive. It came right after the fact, while he was still alive, while the losses were still growing, and it came from the chief custodian of that very scale.

This line is worth placing here first, because it fixes the vantage point for everything that follows.

The previous essay described how that machine was built: standardized contracts, calculable prices, a chain broken into seven links, and ledgers that had begun recording other ledgers.

What comes next is the story of the days the machine stopped — and of who, once it stopped, was caught, and who was not.

A machine breaking down is a technical problem. But who gets to decide, once it has broken down, whom to catch first — that is a different kind of question, and it does not belong to mechanics.

This is also where the crisis becomes most useful to think with. On an ordinary day, everyone is entered in the ledger, and the order of priority among them is invisible; only when resources run short, and people are forced to queue, does that order first show itself.

The crisis did not create that order. It had always existed. It is only that, ordinarily, no one has to use it.

A system wears two different faces: the one it shows on an ordinary day, and the one it shows when resources fall short. The second face is closer to the system's real structure, because on the first, everyone fits, and no one can see who comes before whom.

2. A Wager With No House Under It

To see that order clearly, we first need to see exactly how far this machine had gone.

By 2006, securitization was no longer a marginal financing technique. That year, US subprime originations reached $600 billion, 23.5 percent of all mortgage originations, and most of it was securitized; non-agency mortgage originations came to $1.48 trillion, more than 45 percent above the total for agency mortgages.

By the time a loan was made, its default destination was already to be packaged and sold.

That sentence deserves a slow reading. It does not simply mean that a loan could be sold; it means that the purpose of making the loan had itself changed. A bank used to write a loan because it wanted to collect interest on it for the next thirty years. Now an institution wrote a loan because it wanted to sell that loan within three months. The same contract, the same house, the same borrower — and the lender was looking at two entirely different things.

That shift pulled a relationship out of the transaction. A loan meant to be held for thirty years obliges the lender to care how the borrower will be doing over those thirty years; a loan meant to be sold within three months obliges the lender to care only about how it looks during those three months. The first kind of care is bound in place by self-interest and needs no one's goodwill to survive. The second kind expires on schedule.

What truly drove price out of all contact with reality, though, was a product known as synthetic.

An ordinary repackaged product has, underneath it, a real mortgage security; underneath that security, real loans; underneath those loans, real houses and real people who owe real money. The chain is a long one, but it is a connected one.

Being connected carries weight. As long as the chain holds, there remains a path — however arduous, however many of its seven or eight links one has to cross — leading back from the person who bought the security at the top to the person making a monthly payment at the bottom. It is a hard path to walk, but it exists, and later we will see that someone did in fact walk it.

A synthetic product is not like this. What sits beneath it is not those securities, but a wager on how those securities will perform.

The commission that investigated the crisis put it bluntly: once the real product on the market ran short, Wall Street began manufacturing a cheaper, more easily mass-produced synthetic substitute. The commission also noted that this kind of product financed not a single home purchase.

Not one.

To be precise, none of this makes synthetic products fraudulent as such. They serve a legitimate purpose: an institution already holding a large book of mortgage securities can use them to hedge its position without actually selling what it holds. That purpose is real, and it makes sense as risk management. The trouble is in the ratio. When little of it is bought to hedge and much of it is bought to bet, the character of the thing changes.

The scale involved was not small. One investment bank alone packaged and sold $73 billion of synthetic products between July 2004 and May 2007; among the more than 3,400 mortgage securities its products referenced, 610 were referenced at least twice.

Separate researchers tracking the same market found that roughly 5,500 mid-tier subprime housing bonds were placed into, or referenced by, structured repackaged products some 37,000 times — turning $64 billion of bonds into roughly $140 billion of assets.

Sixty-four billion dollars became one hundred forty billion. No additional house went up in the interim, and not one additional dollar was lent to anyone who actually wanted to buy one.

The previous essay put it this way: the ledger had begun recording other ledgers.

With the synthetic product, there is one further turn of the screw: the ledger no longer even needs a referent. It no longer records any actual debt; it records a bet on how that debt will perform, and that bet can be copied without limit.

One borrower's capacity to repay can appear, at the same moment, in dozens of contracts, though he himself borrowed only once.

By this point the construct no longer needs an external object at all. All it needs is something referenceable and people willing to take either side of a bet on it.

And neither side of that bet needs to be made up of bad actors. The buyer believes the loans will be repaid; the seller believes they will not; each prices according to his own judgment. At the level of any single contract, there is nothing to fault.

The problem sits in the aggregate. A loan can default only once, but the wagers surrounding it may number in the dozens; when it defaults, what comes due is not one payout but dozens.

The risk was not dispersed. It was copied.

This is where one thread running through the earlier essays reaches its limit. In Essay 10, the construct filed unanswerable questions outside its door; in Essay 12, it gave the remainder an orbit that circled the ledger; in Essay 18, the ledger began recording other ledgers. Here, what the construct is doing is no longer processing the remainder at all — it is reproducing itself. It requires nothing new to exist in the world; it only requires one more person willing to place one more bet on the same thing, and its books lengthen by another notch.

3. Even If Cows Built It

For this machine to keep running, it needed a checkpoint partway along: the rating.

A rating was originally a judgment. Someone would examine the quality of a pool of assets on behalf of investors who could not do so themselves, and hand down a grade that could be compared against others, so that a pension-fund manager in some other city would never have to go and look at those houses himself.

That function is real, and it is necessary. Without it, securitization could not work at all, because no one could afford the cost of running due diligence on thousands of strangers' loans.

This, too, is what commensuration genuinely achieves, and it should not be read the other way around. A pension fund in another country being able to lend money to a family it will never meet depends precisely on this checkpoint. The savings in the middle were enormous, and part of what was saved did land, as lower interest rates, in the hands of people buying houses.

The question is what this checkpoint later grew into.

One rating agency's volume of deals across its various repackaged products rose from 220 in 2004 to 749 in 2006; the value of those deals rose over the same span from $90 billion to $337 billion.

At the same agency, the investor-services division's revenue from structured products rose from $199 million in 2000 to $887 million in 2006; that revenue's share of the firm's total income rose from a third to 44 percent.

Staff did not grow to match. In the testimony that followed, one refrain kept recurring: not enough people, always fighting fires, almost no capacity left for actual research.

Among the investigative record is an internal instant-message exchange that would go on to be quoted more than any other. One analyst wrote that a particular deal was absurd, that the model had not even captured half the risk in it. Another analyst wrote back:

We rate every deal. It could be structured by cows and we would rate it.

What makes that line matter is not its crudeness. It is that it names a fact outright: rating was no longer a judgment about risk. It had become a step on an assembly line.

A step on an assembly line has one duty — to let things through, not to stop them.

And this particular step has a property rarely found elsewhere: what it produces is not a good but a judgment that other people will then take as fact. When a factory turns out a bad batch of parts, only that batch is bad. When a rating is wrong, everyone who acted on it is wrong along with it. When a judgment gets used as though it were a fact, the cost of being wrong falls on everyone but the one who made the judgment.

None of this required anyone to decide anything. It emerged from sheer volume: deals rose from 220 to 749, staff did not rise to match, and so the time available for each one shrank by several multiples. Past a certain point, looking closely at a deal and simply running it through the model stop being two different options.

And it was the issuer who paid, not the investor.

This does not need moral language to make the point; it is a structural fact. Whom you are looking out for and whose money you are taking had come apart. The rater was working on behalf of the investor. The check came from the issuer. This arrangement was public within the industry, written into the documents, known to everyone.

Essay 13 put it this way: the whole process can be entirely legal, and the outcome can still be radically unequal. Here it shows itself again.

The technical failure was equally specific. To smelt a pool of mediocre loans into a large block of highly rated securities requires one strong assumption: that these loans will not all go bad at the same moment, or at least not go bad in such lockstep.

The later reckoning showed that the rating models had estimated correlation far too low, and had imagined diversification as far more effective than it turned out to be.

By the end of 2008, the rating agencies had raised their key correlation assumptions to two to three times their pre-crisis levels.

Assumptions can be revised. The houses had already been repossessed.

Between those two sentences sits the time lag between the model and reality. A model can be revised at any moment — change one input, rerun it, and there is a new rating the next morning. But the loans priced under the old input had already gone out, the houses had already been bought, the contracts had already been signed for thirty years. The construct revises itself quickly. It cannot undo what it has already caused.

It is worth a word here about a common measure. The market did not wake up one day to some new piece of information. What it discovered was that the shared scale it thought it had was not, in fact, shared: the same letter meant very different things depending on which product it was stamped on.

A paper out of the Fed even stated outright that in this crisis, credit scores had not performed their supposed function of predicting real default risk.

A scale is at its most dangerous not when it registers one wrong reading, but when everyone is still relying on it after it has stopped measuring the thing it claims to measure.

4. All at Once

The most convenient story to tell after a crisis is that no one saw it coming.

The record does not support that story.

In 2004, the FBI had already labeled mortgage fraud a spreading epidemic, and noted that in just the first nine months of that year, reports of suspicious activity had already topped 12,000 — about three times the total for all of 2001.

In 2005, research from the Bank for International Settlements had already warned that the tranching instruments used in structured finance could produce unexpected concentrations of risk inside institutions' portfolios.

That same year, a large bond fund sent its credit analysts out to twenty cities to ask real-estate agents, mortgage brokers, and local investors, in person, what was actually happening in the housing and mortgage markets. Some called this approach shoe-leather research.

What they were looking for was precisely the information that no number had yet managed to capture.

And during that same stretch of time, in the spring of 2007, the chairman of the central bank was still telling Congress that the effects of the subprime problem on the broader economy and financial markets appeared likely to be contained.

The most important thing to see in this section is not who was right and who was wrong.

It is that all of this was happening at once.

That word — at once — carries the entire weight of this section. If the alarm had come first and the complacency after, this would be a story about someone who failed to heed a warning, and such a story always has a clear party to blame. But what the record shows has a different shape: in the same month, someone was writing the word epidemic, someone was touring twenty cities, someone was saying it all looked containable, and the machine was turning out tens of billions of dollars in new securities every single day, exactly as before.

This coexistence has a consequence. Afterward, whichever side anyone wants to take, the record from that time can supply the evidence. Those who say no one saw it coming can point to the optimistic judgments; those who say the warnings were there all along can point to the warnings. Both sides are citing real material, and it was all sitting in the very same months to begin with.

The alarm was sounding. The complacency continued. The argument went on. Business proceeded as usual. This is not a story of someone who warned first, was overruled, and was later vindicated to everyone's regret. It is a story in which all of these things existed side by side, and no single link in the chain had the authority to stop the machine.

Essay 16 described a structure much like this one: three places where someone saw it, three places where someone wrote it down, and the machine kept turning regardless.

But that case was one of withholding grain while knowing it was there. This one is different. This time, no single person held the switch.

An analyst who saw that some deal was absurd could say so in an internal chat, and that was all he could do. A fund that sent people to twenty cities could buy a little less itself, and that was all it could do. The Bureau wrote up its findings, and those findings entered an enforcement process that was already stretched past capacity. The regulators held real power, but they believed markets would discipline themselves, and at the time, that belief was the mainstream view.

This is often where the construct is hardest to fight. It does not need anyone to intend harm. It does not even need anyone to order that things continue. All it needs is for every person, at every post, to do the thing that is reasonable from where they happen to be standing.

Stopping the machine was nobody's job.

The distinction from Essay 16 needs to be stated plainly. In those three famines, the machine could be stopped — at the time, someone pointed out exactly how to stop it, and the power to do so really did sit in specific, identifiable hands. This case is different: there was no single switch that could halt the whole machine, only the switch for one's own segment of it. And the result of any one person shutting down their own segment was, as a rule, simply that the business moved to another firm.

This is why responsibility is so hard to apportion in this register. The trouble is not that no wrongdoer can be found. It is that everyone, at their own post, was doing the job that belonged to that post, and the sum of everyone's proper job was that outcome. Stopping the machine would have required someone to step outside their assigned place, and stepping outside one's place carries no institutional reward — and usually carries punishment instead.

5. The Weekend the Machine Stopped

The timeline needs to be set out clearly, because the difference between being rescued and not came down to a matter of days.

And this line only makes sense read as a continuous whole. Taken separately, each event looks like an isolated institutional crisis; read together, it is the same thing surfacing, in turn, at different points along the chain: first the dirtiest end of it stops selling, then the funds holding it cannot meet redemptions, then the banks financing those funds cannot borrow, and finally everyone who had been lending to those banks refuses to lend at all.

2007 was the year the machine began to stall. On February 27, a government-sponsored enterprise announced it would stop buying the riskiest subprime mortgages and related securities. On April 2, a company that had once been a leading subprime lender filed for bankruptcy protection. In June, the rating agencies began downgrading a batch of subprime-related bonds. On June 7, a highly leveraged structured-credit fund run by an investment bank suspended redemptions. On July 31, two of that bank's funds, both heavily weighted in mortgage securities, went into liquidation. On August 9, a French bank froze redemptions on three funds — a date now widely treated as marking one of the starting points of the global credit panic.

The significance of that date lies in where it came from: outside the country. Until then, every incident could still be explained away as a local problem in the American housing market. A French bank freezing redemptions showed that these instruments were already sitting on other people's balance sheets. The chain broken into seven links, described in Essay 18, revealed its full length at that moment.

The short-term funding markets were the first to give way. The outstanding balance of asset-backed commercial paper fell from $1.18 trillion in August 2007 to $745 billion in August 2008 — a decline of 37 percent.

In March 2008, that investment bank nearly went under. On March 13, it told the central bank it could no longer obtain financing in the market; on March 14, it survived, barely, on a bridge loan; on March 16, it accepted a takeover by a larger bank. The official post-mortem later put it without hedging: absent extraordinarily large liquidity support, or a takeover by a stronger institution, it likely would not have survived to open its doors on Monday, March 17.

On September 7, two government-sponsored housing-finance enterprises were placed into government conservatorship.

On Friday, September 12, another investment bank still had $18.5 billion outstanding against one central-bank facility; on Monday, September 15, it borrowed a further $28 billion from a second facility, and even that could not stem the flight of funds.

That weekend, history very nearly went a different way.

The investigative record shows that a British bank had been the most likely buyer, and that Britain's Chancellor of the Exchequer refused to let British taxpayers stand behind an American bank. According to the recollection of that investment bank's own lawyers, the US Treasury Secretary later told them: the British government was not going to do this.

On September 15, it filed for bankruptcy protection. The same day, another investment bank announced it was being acquired by a commercial bank.

A later researcher summed up that failure in the coldest possible terms: it went bankrupt in the small hours of September 15 because, that morning, there was not enough cash on hand to open for business.

That account moves the whole episode out of a story about insolvency and back into a story about a short-term funding line that simply snapped.

The difference between the two accounts is not a scholarly nicety. If an institution falls because it is insolvent, then rescuing it means rescuing something already rotten; if it falls because, overnight, it could no longer borrow, then rescuing it means patching a gap in time. The same event, read the first way, is just deserts; read the second way, is bad luck. And at the time, neither reading could actually be confirmed, because confirming it would have required knowing what those assets were truly worth — and that was precisely what no one could say.

The forty-eight hours after its fall were the real systemic thunderclap.

On September 16, the New York Fed was authorized to lend an insurance conglomerate up to $85 billion. That same day, a money-market fund, which held that investment bank's paper, saw its net asset value break below a dollar — later confirmed at 97 cents — and a run on money-market funds began.

On September 19, the central bank introduced a facility to support the purchase of high-quality commercial paper from money funds, and the Treasury introduced a guarantee program for money-market funds of up to $50 billion.

On September 21, the two largest investment banks were approved to convert into bank holding companies.

On October 3, the $700 billion rescue package was signed into law.

In little more than ten days, an entire set of boundaries that had once kept separate domains apart, each minding its own business, was redrawn. Who could draw on the central bank's money, who could change categories, who could receive the state's guarantee — all of it was settled within those ten-odd days.

What deserves the closest attention here is the word identity. The two largest investment banks being allowed to convert into bank holding companies meant they changed category: from then on they could walk through a different door, draw a different kind of money, and answer to a different set of rules. Essay 13 observed that the line the construct draws is never between activities but between organizational forms. This time, no one crossed that line. The line itself moved, for a chosen few.

6. The List

Now the question can be asked: why these, and not those.

The answer is not conspiracy — and it is more worth examining than a conspiracy would be.

A conspiracy can be corrected: find who colluded, remove them, and order is restored. But a scale that cannot see a certain thing will fail to see it no matter who is holding it.

Rescue, too, requires a scale.

In those days, the people making decisions genuinely had a calculable quantity in hand, and they really were calculating it. It was called systemic risk: how many counterparties this institution's collapse would drag down, how many financing chains would freeze, how many contracts would suddenly require simultaneous settlement, whether it would touch off a chain of runs.

By that measure, an insurance conglomerate was worth an $85 billion loan, because it stood as counterparty to countless credit-default-swap trades; its collapse would have meant yanking the other side out from under hundreds of contracts at once.

By that measure, an investment bank's worth depended on how many chains it was woven into, and on whether anyone stood ready to take it over.

By that measure, money-market funds had to be guaranteed, because tens of millions of people treated them as cash, and a run on them would spread straight into commercial paper, and from there into the money businesses use to meet payroll.

Every one of these could be calculated. Every one had a reason behind it. And most of those reasons held up, given the conditions of the moment.

It might even be said that the people making decisions during those weeks were doing the only thing they could do. Time was measured in hours, information was badly insufficient, the legal tools on hand were limited, and every additional hour might bring down one more institution. Under those conditions, catching first whatever threatened to pull everything else down with it was a reasonable judgment — and a responsible one.

The trouble is not that these calculations were wrong. The trouble is what this scale is able to see.

It can see institutions. It cannot see individuals.

An institution can be entered into systemic risk because its collapse propagates: it has counterparties, a balance sheet, a web of contracts, hundreds of firms tied to it. All of this can be counted.

A person who loses his house does not propagate. His default will not send three hundred other people into default at the same moment. He has no counterparty. His loss stops with himself and his family.

On the scale of systemic risk, his reading is zero.

And so the list took the shape it took. What qualified someone for rescue was never how badly they were hurt, never who was in the wrong, never who deserved it more. It was whether they could be entered into that quantity.

For something to enter this quantity, it must first meet one condition: it must be connected to others. And the tighter the connection, the higher the reading. Which gives this scale a property rarely spoken aloud: it rewards entanglement. The more an institution is bound up with everyone else, the closer to the front of the line it stands when the crisis comes.

An institution whose fall would set off a chain reaction had to be caught, however disreputably it had behaved along the way, because catching it was done for the sake of everyone else.

A man who signed a contract he could not read, and who now cannot make next month's payment, does not make that list even if he never did a single wrong thing — because saving him would only save him.

This logic holds together. In those weeks it was, in fact, the only logic that did hold together. Resources were limited, time was measured in hours, and catching first whatever threatened to pull everyone else down was a reasonable way to sort the field.

And it was, at the very same time, a table of values.

The instant a scale begins to rank things, it is announcing what counts. Systemic risk measures how tightly one thing is bound to everything else; it has no column for how that thing is doing on its own.

Essay 16 said that some people stand outside the register. This time the shape is different. This time, those people are not missing from the registers — they are in a great many of them: in the credit records, in the foreclosure statistics, in the unemployment figures. They simply are not in the one register that mattered.

The remainder here takes on a different way of existing. It is not being omitted. It is being placed further down the line.

And an order of priority is invisible on an ordinary day, because on an ordinary day, no one has to stand in line.

This also explains why this argument is so hard to settle. On an ordinary day, anyone can say that the system treats everyone alike, and produce the evidence for it: the same rate sheet, the same application process, the same contract language. All of that is true. The order of priority is not written into those documents. It is written into who gets the money first when resources run short — and that only happens once every few decades.

What a crisis does is compress resources down past the point where there is enough to go around, and so the order that had been there all along gets spoken aloud for the first time.

7. Two Numbers, Both Correct

The cost comes with numbers attached.

They belong here because everything discussed so far has concerned institutions, and institutions do not lose sleep.

All of these numbers share one thing: every one of them was tallied after the fact. At the moment decisions were being made during those weeks, not one of these figures yet existed. How many people would lose their jobs, how many families would have to move — nobody could work that out at the time, and nobody was asked to. Which institution to save was a question that needed an answer that same day. How large a price would be paid for it was something no one totaled up until years later.

From the first quarter of 2007 to the second quarter of 2011, the national average home price fell by more than a fifth.

Unemployment rose from 5 percent in December 2007 to 9.5 percent in June 2009, and reached 10 percent in October 2009; that month, more than 15 million people were out of work.

The number of foreclosures cannot be pinned down, and the fact that it cannot is itself worth noting.

One reserve bank estimated that roughly 3.8 million foreclosures took place between 2007 and 2010. Another data firm, looking back over a longer span, put the cumulative total of completed foreclosures since 2007 at about 7.8 million.

Both figures are correct. They measure different things: one counts within a fixed span of years, the other counts cumulatively from that point forward; one counts starts, the other counts completions.

The gap between these two numbers is roughly the size of every household in a mid-sized country.

Set this beside the figure carried out to seven decimal places back in the first section, and a pattern comes into view. What gets measured with extreme precision, and what cannot even settle on a common definition, has nothing to do with which matters more. It depends on which is easier to define as a countable event.

The scale running through the previous eighteen essays could carry a single loan out several decimal places, price a tranche's spread down to the basis point, mark a contract's present value to the day's close.

But when it came to counting how many families had been made to leave their own homes, the standards themselves began to disagree.

This is not because the statistical agencies were careless. It is because foreclosure is not a tidy event: some people sold the house before the process finished; some dragged it out for three years; some had the house taken after it had already sat empty for half a year; some never entered the formal process at all and simply mailed their keys back to the bank one day.

For something to be counted precisely, it first needs a shape precise enough to define. Losing one's home does not have that kind of shape.

A shapeless thing still happens, and happens completely. It is only that, in the recording of it, it must first be cut down into some countable form, and whatever gets trimmed away in the cutting drops out of the count. Essay 12 observed that where a scale cannot see, people assume there is nothing there. Here the situation is different: the scale did see it, and what it produced were two numbers that do not agree.

There were people inside the failed institutions too. After that investment bank went under, roughly 26,000 employees around the world lost their jobs.

There is something the central bank chairman of the time said himself. In a 2007 speech on the subprime market, he had already acknowledged that, for borrowers, the consequences of default could include losing both home equity and the home itself; and that, for neighbors, foreclosures concentrated in one area would depress the value of the surrounding properties.

A sitting central bank chairman saying this in a public address carries a different weight than a critic saying it. It shows the thing was seen at the time, seen at the very top, and set down in an official prepared text. It simply never turned into any decision that had to be made during those particular days.

This last half of the sentence is the one that matters.

When one house is repossessed, the house next door loses value along with it; once that happens, the family next door may no longer be able to borrow; unable to borrow, they too may fail to keep their house.

This is a form of transmission, and it transmits quite genuinely.

It simply never enters the quantity called systemic risk, because it travels sideways, street by street, rather than upward along a network of contracts.

It cannot move a single institution's balance sheet. So on that scale, it never registers.

This is the sharpest asymmetry in this entire history. Transmission along the vertical axis gets calculated, because it travels along contracts, and contracts carry reference numbers. Transmission along the horizontal axis does not, because it travels along streets, and streets have no counterparties. The same kind of event — one thing causing another — and one version enters the model while the other never even gets a variable.

8. Never the Subject

On causes, researchers have not converged even now, and the disagreement is not merely a matter of taste.

One line of argument centers on deregulation and regulatory failure. This is the main thread of the majority position in the official inquiry into the crisis, and it points not to a single piece of legislation but to a long institutional drift: faith in market self-discipline, the freedom of institutions to shop for the most lenient regulator, over-the-counter derivatives without transparency or collateral requirements, and a shadow-banking sector performing the same maturity transformation and liquidity mismatch that banks perform. One line in that inquiry's conclusions is written without hedging: three decades of deregulation and reliance on self-regulation had eroded key safeguards. The same conclusion insists, just as firmly, that regulators were not without power — they had power and did not use it.

A second line of argument centers on monetary policy held too loose. Some researchers argue that the central bank kept interest rates too low for too long in the first half of the 2000s, and that this significantly inflated the housing bubble. The federal funds target rate fell from 6.5 percent in December 2000 to 1.75 percent in December 2001, and down further to 1 percent by June 2003.

Researchers inside the Fed system, along with the chairs who came after, do not accept this as the primary cause. They argue that the short-term policy rate alone cannot explain the scale of the run-up in home prices, the narrowness of risk premiums, or the fact that bubbles inflated in step across many countries at once. The two sides are not arguing over whether rates played a role, but over whether they were the most important driver, and over whether raising them, given the real constraints of the moment, could actually have stopped the bubble from forming.

A third line of argument centers on global imbalances and a glut of savings. Capital surpluses from emerging economies and oil-exporting states flowed into the United States and Europe, pushing down the yields on safe assets and forcing the financial system to manufacture more assets that merely looked safe. Some researchers propose that the global imbalance and the financial crisis were both products of the same underlying cause — that the same policy distortions produced them side by side. Others caution against treating the global imbalance as mechanically equivalent to the crisis, arguing that what truly mattered was how the global financial system translated those capital flows into leverage, maturity mismatch, and mispricing.

A fourth line of argument centers on fraud and corrupted incentives. This line does not read the crisis primarily as a macroeconomic imbalance but emphasizes false information and deliberate concealment at the micro level. Senate testimony from 2007 cited industry samples in which, among so-called stated-income loans, 90 percent involved overstated income. Later academic research identified widespread misrepresentation of asset quality across the non-agency residential mortgage-security market: in a market worth roughly $2 trillion, intermediaries gave buyers false information, in the contract disclosures themselves, about the true quality of the underlying assets — whether a borrower actually lived in the house, whether the house carried a second lien — key variables that were significantly misreported.

A fifth line lays the primary blame on government housing policy. This is the most well-known dissent within that same official inquiry: that housing goals, agency incentives, and steadily loosened underwriting standards together produced an enormous quantity of high-risk mortgages, and that this was a necessary condition for the crisis. The majority opinion explicitly rejected treating the Community Reinvestment Act as the main cause, pointing out that most subprime originators were not even covered by that law, and that the share of high-cost loans connected to it was small.

No verdict will be handed down here.

But there is one thing common to every one of these explanations, and it is rarely pointed out on its own: almost no serious research places the responsibility on the borrower alone.

The disagreement is over whether the responsibility belongs mainly to macro policy, to regulatory structure, to institutional incentives, to international capital flows, or to some combination of these. And in every version of that combination, the person who signed the contract is never the protagonist.

This deserves to be looked at on its own, because it runs exactly opposite to how things turned out. In every explanation, he is the one things were done to: sold a product, handed loosened standards, offered a rate that had been pushed down, met by capital that had come pouring in, given material that had been misreported, chased for yield he never asked to chase. In no version of the story does he occupy the position of the subject.

He is not the protagonist. And yet the share he bore was the most complete of anyone's.

Back to the opening line.

A man on the witness stand said: I found a flaw.

That was indeed a flaw, and he named the right location for it: institutions answerable only for their own profits and losses will not necessarily protect even themselves, let alone anyone else.

Except that line can be pushed one step further.

That machine compressed one person's ability to repay into a credit score; compressed a house into a loan-to-value ratio; compressed a neighborhood into a probability of default; compressed thousands of loans into a single letter; then compressed that letter into a point value that moved every day. Every step of compression had its logic. Every step raised efficiency. Every step sent more money to people who otherwise could never have borrowed it.

But compression runs in one direction. It leaves behind whatever cannot be converted, layer after layer, and the higher up the chain you go, the fewer people can still see it.

By the day the rescue had to be decided, the scale used to determine who would be saved first was measuring in that very same direction: who was tightly connected, who stood at the front of the line.

One man was not tightly connected. He was simply himself, tied to one house, one job, one street, a handful of relatives.

Those ties are real, and to him they are everything. They simply do not appear on any balance sheet.

None of this needs to be written as an accusation. It is a fact anyone can check: on the scale actually used to make decisions during those weeks, there was, in truth, no column for this. For that column to exist, someone would first have had to build it — and in those days, with time measured in hours, no one had the time to build a new scale. At the moment that matters most, the construct always reaches for whichever scale is already in its hand.

The ledger has not yet balanced. It is still being kept.