Show me how you're paid, and I will show you what happens next.









A rat with no tail. My first three ideas: a disease, a fight, or a predator that grabs the tail and misses the rest. All three of mine were wrong.

The new plan was simpler and cheaper. Bring in a rat tail, get a coin. Anyone could do it. No wages, no supervisors, no arguments. You cannot count rats in a sewer. You can count tails on a table.

A modern sewer network opens beneath the city. Rats move in and thrive.
Paid hunters go down. On 12 June alone, 20,114 rats are killed.
The city opens the reward to everybody: a coin for every tail.
Tails arrive in their thousands. Live rats without tails appear in the streets.
Officials find rats being bred on the edge of the city.




The rat farmer was not even breaking the rule. The city had asked for tails. He was producing tails. He was the best supplier the scheme ever had.








A city pays one coin for every tail. Catchers bring in 500 tails a day for six months. That is about 90,000 coins and about 90,000 tails. How many fewer rats are there? Work it out before you turn over.

You cannot answer it. Nothing in the scheme measured rats. If every tail came from a dead rat, there are 90,000 fewer. If every tail came from a rat that was let go, there are more. The count is the same either way. The number could not tell the difference — so nobody could.




Nobody chose tails because tails were the point. They chose tails because tails could be counted, carried and checked. That is a completely different reason, and I did not notice it was a different reason.








It is three hours and fifty minutes. Two hundred people are still waiting and the clock is about to fail them. What is the fastest way to hit the target? Not the best way. The fastest.

Move them. Admit them to a ward, and the emergency clock stops. Nothing has been cured. Nobody has lied. The patient may be in exactly the same trouble on the other side of the doors. But the four-hour rule has been kept.




England's NHS Plan promises that nobody will wait over four hours.
The target goes live: 98% of patients finished within four hours.
Long waits fall sharply. On its own terms, the target works.
The threshold is eased from 98% to 95%.
Researchers publish the spike sitting just before the fourth hour.








An economist writes that a pattern collapses once it is used for control.
A psychologist writes that indicators corrupt the process they monitor.
An anthropologist gives the idea the short sentence everyone now quotes.
The idea is famous. It is also broken again every single year.

That is what makes it a law rather than a story. It is not about banks, or hospitals, or rats. It is about what happens to any measurement once somebody's reward depends on it. Change the field and it still happens. Change the century and it still happens.




The neat sentence — when a measure becomes a target, it stops being a good measure — is the anthropologist's, in 1997. The economist wrote something heavier in 1975. Being nearly right about who said it is still being wrong.








A branch is measured on accounts opened. It has 1,000 customers and a target of eight products each. That is 8,000 accounts. How many of the 8,000 does the target require anyone to use?

None of them. The target stopped at the word opened. An account nobody wanted, nobody used and nobody remembered counted exactly the same as an account that changed somebody's life. The number could not tell those two apart either.

Countable today. Reported by the person being paid for it.
One tidy figure for a whole bank, in one line.
Nobody could count it easily, so nobody counted it.
The measure had already stopped at 'opened'.




About 5,300 staff are dismissed for sales-practice breaches.
Regulators impose $185 million in penalties.
Product sales goals for branch staff are removed altogether.
Pay is re-linked to customer service and use, not products sold.
A review raises the estimate to about 3.5 million accounts.








I keep wanting there to be a villain. If there is a villain, I can be safe by not being him. There is no villain here, and that is exactly why it is frightening.

A system does not need a conspiracy. It needs one number, one consequence, and enough people. Give five thousand ordinary people the same squeeze and a few hundred will bend in the same direction — not because they planned it together, but because they were each standing under the same weight.




Counted daily, reported publicly, attached to your name.
Miss it often enough and you lose the job.
Everyone around you is somehow managing it.
Thousands of strangers, one identical squeeze.








The year before, the chain had put its service advisers on sales quotas and commissions. Their pay now moved with the size of the bill. Nothing about the brake pads changed. Only the person looking at them.

This is the quiet part. Almost nobody invents damage that is not there — that is a lie, and lies are frightening and rare. What happens instead is smaller. Every doubtful case, every borderline part, every it-would-probably-be-fine starts landing on the same side. Nobody notices doing it.




Service advisers are moved onto sales quotas and commissions.
Investigators report unneeded work in 34 of 38 undercover visits.
The company ends the quotas and commissions for those advisers.
About $8 million settles the California case.








Which evidence I pick up first. How long I keep looking. When I decide I have looked enough. Nobody audits any of those. And by the time I reach a conclusion, all three have already happened.

A liar can be caught. Ask again, ask differently, check the record. Somebody who is sincerely wrong passes every one of those tests, because there is nothing hidden to find. They will look you in the eye. They will be glad to explain. Sincerity is persuasive, and it is not evidence.




Knows the truth and says otherwise. Rare, and catchable.
Believes it completely. Passes every honesty test.
Not at the answer. At which evidence gets picked up.
It is contagious, and it feels like conviction.








Wrong again, and wrong in the same way as last time. I went looking for people who knew it was rotten. Chapter eight already told me what to look for instead: people who were paid to be sure, and were.

The pressure is gentle and constant. A borderline deal, a Friday deadline, a client who will walk. Each judgement is defensible on its own. Add up a hundred thousand of them and one agency stamped its top grade on about thirty mortgage deals every working day of 2006.




Rating agencies move to being paid by the firms they rate.
One agency awards its top grade to about 30 mortgage deals a working day.
The downgrades begin, and do not stop.
Investments graded very safe turn out not to have been.
A public inquiry calls the rating failures essential to the collapse.








Hospital A operates on everyone. 90 of 100 survive. Hospital B turns the hardest cases away. 98 of 100 survive. Which hospital is better? And what would you need to know to answer?

You cannot tell from the scores. B's number is better and B's patients were easier. A's number is worse and A took the people nobody else would take. The board shows both hospitals a single figure with no label on it. The score cannot say which stone was heavier to lift.

Real. Slow. Difficult. Exactly what was wanted.
Also real. Fast. Free. Invisible in the score.
Both of them added together, with no label.
The patients nobody was willing to take on.




Both of my answers this chapter were extreme, and both were wrong. Publish everything was wrong. Publish nothing was wrong. The dull answer in the middle is the one that actually helps people, and it is much more work.








Every other chapter rewarded an output: tails, accounts, repairs, grades. This one rewards telling somebody early. That is genuinely the thing the factory wants, and it is countable, and it cannot be faked into existence.

Invert the design and see it. Reward output only, and a fault gets hidden, passed down the line and found by a customer a year later. Reward the pull, and the same fault costs four minutes. The cheapest moment to find a problem is the moment it appears.




You run a help desk. You can pay people for A. calls answered per hour · B. problems actually solved · C. customers saying they would come back · D. nothing extra, just a fair wage. Each one can be bent. Which do you choose, and what goes wrong first?








1 · Pick one number. 2 · Make sure it is easy to count. 3 · Make sure the real thing is hard to check. 4 · Pay now, long before the consequences show up. 5 · Let the person being paid do the counting.

Every one of them had all five. Tails: countable, rats uncheckable, paid at the desk, counted by the catcher. Accounts, repairs, grades, survival rates — the same five, every time. She had not invented a trap. She had described the ones that already happened.




Your bonus depends on a number that you also report. You may: A. leave it · B. have somebody else measure it · C. be paid on three numbers that pull against each other · D. be paid in five years, once the results are in. Every option costs something. Which cost would you accept?








The accounts were opened between 2011 and 2016. The largest bill landed in 2020. A limit on the bank's size stayed on until 2025. The person who opened the four extra accounts had left years before any of it.

A reward that pays today for something judged in a decade is a kind of borrowing. It takes the good part now and posts the bill forward. Nobody feels dishonest doing it, because the bill has not arrived yet. The gap between the reward and the consequence is where every trap in this book has lived.

The extra accounts are opened, and paid for, month by month.
The penalties arrive: $185 million.
A regulator caps the bank's size until it fixes its controls.
A $3 billion settlement closes the case.
The size cap is lifted, seven years after it was imposed.




Arrives this month, in full, to you.
Arrives in years, often to somebody else.
Every trap in this book lives inside it.
Pay later, or measure something that answers sooner.








Three questions I can ask about any measure, and none of them need an expert. Who is paid by this number? What does it count that I do not want? And when does the bill arrive?

None of this makes rewards bad. The cord in the factory is a reward, and it works. Rewards built the sewer, the hospital, the workshop and the bank. The care goes into choosing what is being rewarded — because whatever it is, you will get more of it.




In the bank, thousands agreed. In the workshop, thirty-four of thirty-eight agreed. Among the graders, an entire industry agreed. They agreed because they were all paid the same way — and agreement that large is very easy to mistake for proof.



Siri can now ask who is paid by a number. Then she notices something harder. A thousand people, certain of the same thing, on the same morning. Is that evidence — or is everybody copying everybody?