Luck Is an Engineering Problem

Richard Wiseman handed people a newspaper and asked them to count the photographs. Planted inside was a line in plain type: stop counting, there are 43 photographs in this newspaper. A second planted line offered prize money to anyone who told the researcher they saw it. People who rated themselves lucky kept finding the lines. People who rated themselves unlucky counted photographs, carefully, to the end.

Both groups held the same newspaper. Attention split them. The old argument, luck against skill, treats luck as a single variable. Luck separates into three parts, and each part answers to a different control.

The cleanest decomposition I know is sancai, an old Chinese frame: heaven luck, earth luck, human luck. In poker terms: the cards you are dealt, the table you choose, the way you play the hand.

Getting the decomposition wrong costs in both directions. Call everything heaven luck and you stop trying, because effort looks pointless. Call everything human luck and you stop learning, because wins prove your genius and losses need a villain. Melvin Lerner named the engine under both errors: the just-world hypothesis, the need to read every outcome as deserved. Fatalism and self-congratulation are the same accounting error with opposite signs.

Most engineers I know spend the improvement budget on the play. They tune habits and read decision books, inside a game they never audited. The consensus is rational: the play is the only part you touch daily. I believe the table is the bigger lever, and the payoff math backs it.

The only move on the cards is the frame

Heaven luck is the birth lottery. Take two babies born today, one in Oslo and one in Freetown. The Oslo child gets a life expectancy of 83 years, the Freetown child 62. The odds of dying before age five differ by a factor of 47, and GDP per person by a factor of about 100. Neither child chose anything.

Warren Buffett calls his own draw the ovarian lottery. He was born in 1930, in the United States, male, analytical, exactly when American markets began to pay for analysis. He bought his first stock at 11 and stayed in the game for eight decades. Morgan Housel did the arithmetic on longevity: of Buffett's 84.5 billion dollars when Housel counted, 81.5 billion arrived after his mid-sixties. Buffett argues his two sisters matched him in intelligence and ambition, and their era offered them no market. Bill Gates likes to remind him that the same brain, born 100,000 years earlier, ends up as some animal's lunch.

John Rawls built a test for pricing this draw. Judge a society from behind a veil of ignorance, before you know which ticket you hold in it. A rule that survives the veil is fair. Most defenses of a lucky position do not survive it.

None of this is negotiable, and the trap is treating the draw as a verdict. Luck is a comparison against alternatives you imagine, so every draw is relative to the reference set you carry. Born in Freetown today beats born in Freetown during the civil war, which beats born there a century earlier. The farmer in the old Chinese parable answers every windfall and fracture with the same line: good luck, bad luck, who knows. The line is a refusal to grade a hand in the middle of it.

In 1978, Philip Brickman's team tracked lottery winners and paralyzed accident victims. The winners were barely happier than controls, and the victims were far less miserable than anyone predicted. The reference set moves to meet the draw. Even the extreme cards fade to baseline. The frame is the durable part.

The reference set also steers behavior. People who feel lucky scan wide, and a wide scan finds planted lines and useful strangers. Anxiety runs the reverse program in hardware: under stress the visual field narrows and the periphery goes dark. Napoleon asked his generals whether they felt lucky and promoted on the answer, because the answer proxied opportunity-spotting under chaos. The internet works against the frame by handing everyone thousands of curated reference lives, which is why fortunate people now feel poor.

So the move on heaven luck is frame discipline, and the exercise is gratitude, run mechanically. Widen the reference set on purpose until the current draw sits in context. Buffett lands on duty instead of guilt. A winning draw carries an obligation to spread the payout. Guilt about your cards helps nobody, and it plays them badly.

The table decides more than the play

Earth luck is the environment you install your life in: the city, the industry, the game, the partner. It is the hardest of the three to see and the most important to get right. One environment decision reprices thousands of downstream decisions at once. Poker professionals treat this as doctrine, and the biggest decision in the game is table selection, before a single card matters. Charlie Munger gives it to investors as fishing rules: fish where the fish are, and never forget rule one.

The reason is the shape of the payoff curve. Dentistry pays on a flat curve, where the best dentist alive earns two or three times the average one. Publishing pays on a power law. Mark Manson estimates his own book is about twice as good as the median self-help title, and it returned about 20,000 times the median result. Twice the quality, twenty thousand times the payoff. The curve does the work, so choosing the curve is the decision.

Founders meet the same law through Andy Rachleff. When a great team meets a lousy market, the market wins. When a lousy team meets a great market, the market wins again. Marc Andreessen built his product-market-fit essay on top of that law, and the law is table selection restated.

Power-law games arrive in waves, and birth years cluster at the crests. Gates and Jobs were born in 1955 and hit career start exactly when personal computing opened. The birth year was the card, and the school was the table. In 1968 Lakeside School put a teletype in front of an eighth-grade Gates, access that was rare at any age. He banked years of programming before the industry existed.

Zuckerberg arrived in 1984, Amodei in 1983, Altman in 1985. The pattern is not destiny, because the downside of chasing a crest early is capped. Zuckerberg built a Spotify-shaped product in high school, years too early, and it cost him a side project. The upside is the company that names the decade, and that asymmetry is the whole case for trying early.

AI is the current wave, and I moved my work onto it with open eyes. The gold rush pays storytellers before builders, and I stayed anyway. Conviction about the technology was half the call. The other half was table selection, and the honest ledger says the table will do more for my outcomes than my talent will.

Some of what reads as luck is a variable the observer failed to model. From the outside Napoleon looked like the luckiest general alive. Up close he ate and joked with his soldiers while enemy generals dined as princes, and morale won battles that the map said were lost. The analyst Ethan Arsht scored thousands of historical commanders against the outcomes their resources predicted. Napoleon sits so far above expectation that he reads as a data-entry error. Engineered variables, invisible to the opponent, settle as luck in the record.

Wiseman spent a decade on lucky people and found four behaviors, and behaviors is the operative word. Lucky people act on chance encounters, trust intuition where they hold real experience, expect good outcomes, and convert setbacks into lessons. Will Smith's father put the first behavior in one line decades ago: luck is when opportunity meets preparation. All four go on a calendar. Lucky is a behavior set, not a trait.

Louis Pasteur said the lab-bench version in 1854: in the fields of observation, chance favors only the prepared mind. The mold that drifted onto Alexander Fleming's plate in 1928 became penicillin because the man at the bench spent years on antibacterial agents. On any other bench, the plate goes in the sink.

The cheap implementation is surface area. Richard Hamming watched it at Bell Labs. The scientists who kept their doors open ate constant interruptions and did the work that mattered, because the door leaked in the problems worth solving. Mark Granovetter measured it in job markets. More than 80 percent of jobs found through contacts came through weak ties, the people seen occasionally or rarely. Close friends know the same openings you do, and strangers carry news from outside your graph.

Jason Roberts gave the mechanism a formula: luck surface area is what you do multiplied by who hears about it. The multiplication is the useful part, because a zero in either term zeroes the product. A builder who never tells runs at zero, and a teller who never builds runs at zero, no matter how large the other term grows. That formula is why I write.

Companies build surface area into the floor plan. Steve Jobs designed Pixar's headquarters around one central atrium so that animators and accountants collided on the way to coffee. Google gave engineers a fifth of their time for side projects, and Gmail fell out of that budget. Cities do it at a larger radius. Paul Graham wrote down their ambient messages: New York says be richer, and Silicon Valley says be more powerful. Pick the message you want repeated at you for a decade.

None of this needs a relocation budget. Take the dinner you planned to skip, and work from the room that has other people in it. If one encounter in ten matters, the job is to get through the other nine fast. A closed door processes zero per week.

The largest earth-luck decision has nothing to do with careers. A partner sets the operating context for thousands of decisions across decades. Marry chaos and every downstream bet inherits it. Marry well and the portfolio runs cheaper.

The strongest objection to table-first thinking is that it licenses quitting. If the table is always the suspect, every hard month reads as a bad game, and you hop tables forever. Table-hopping loses to almost any player who stays, so the objection is real. It lands on hopping and misses auditing, and the difference is a calendar and a record. Change tables on scheduled evidence. Change tables on a bad week, and the audit is an excuse.

Grade the bet, never the outcome

Human luck is play quality. Annie Duke played poker for a living and then built a second career on decision science. Her frame is the usable one: every decision is a bet. No choice in life sits at 100 percent or at zero, and the true probability is partly hidden from you. Two forces separate your choice from your result, bet quality and variance. Only bet quality takes instruction.

The corruption starts when you grade decisions by outcomes, which poker players call resulting. A hand that wins 95 percent of the time loses one night in twenty, on schedule. The losing night carries no lesson, and mining it for one produces superstition. Self-serving bias then finishes the job: wins file themselves under skill, losses under bad luck, and the feedback loop dies. Duke goes further and drops the adjectives, because a draw you call a disaster today can pay out for decades. She left graduate school sick and broke, played poker as a stopgap, and the stopgap built everything the PhD was supposed to.

The fix is mechanical rather than motivational. Move the forecast from implicit to explicit before you decide. Write down what you expect to happen and the condition that reverses the call. The act of writing reduces error on its own, and the record turns the outcome into a graded exam instead of a mood. Engineers already run this discipline for systems. We write down why we picked Postgres, and never why we picked the job.

Duke's clients run premortems before big bets: assume the bet failed a year out, then name what killed it while the naming is cheap. Daniel Kahneman adds the outside view: before forecasting your own attempt, look up what happened to everyone who tried it. Your inside story feels unique, and the base rate mostly wins.

Scale the effort by the cost of reversal. Jeff Bezos sorts decisions into one-way doors and two-way doors, and he sets the speed by the door. A two-way door gets decided fast, at about 70 percent of the information you want, because you can walk back through it. A one-way door gets the slow treatment. An intern hire is a two-way door with a checklist, and a CEO hire is a one-way door with a dossier.

Two biases fight the write-down. Prospect theory measured the first: a loss weighs about twice as much as an equal gain, so imagination over-renders the bad branch. The second is that nobody stress-tests the status quo.

Duke tells a story about an ER doctor who hated her job for years and froze on an offer to leave. The block was the classic one: what if the new job turns out bad too. Duke asked her to rate the odds of being happy in a year if she stayed. Zero, she said, with years of evidence. The new job, sight unseen, rated a fifty. Fifty beats zero, and she quit the next day.

Years of agonizing collapsed under one explicit comparison, because the status quo never had to sit for the exam.

Volume is the last lever on human luck. A 5 percent edge is invisible in one hand and near-certain across 1,500 hours. Venture capital is the same machine pointed at a power law: dozens of bets a fund, and one or two companies pay for everything. Repetition shifts the weight from variance to skill, and it shortens the life of every wrong path.

Volume has one precondition, survival. At Bell Labs in 1956, John Kelly worked out how much of a bankroll to stake on a given edge. The formula carries a warning. Bet too large and a positive edge still ruins you, because one bad streak ends the game. The averages only pay the players still at the table. Size every bet to survive the worst streak, and treat any bet that can end the game as mispriced at every price.

One inversion from the poker table is worth keeping. The better your decisions get, the more bad luck you take in public. Good play invites opponents to put money in behind the odds, and the odds still pay them sometimes. Duke reads a bad beat as evidence the strategy works. Check the decision once and skip the autopsy.

At the top, the scoreboard measures luck

Michael Mauboussin calls it the paradox of skill: as everyone in a field gets better, luck decides more of the outcome. Skill converges at the top and the spread between competitors shrinks. The variance that remains belongs to luck. Baseball shows it cleanly. Nobody hit .400 after Ted Williams in 1941, because pitching and fielding got better everywhere and the whole league packed into a narrower band.

Robert Frank ran tournament simulations to price the effect. Give skill 98 percent of the weight and luck 2, then add enough contestants, and the most skilled entrant almost never wins. At the tail everyone is skilled, so the winner is the skilled entrant who also drew the best luck. Every crowded field runs this tournament, and yours is a crowded field.

The evidence you study makes this worse, because winners are the only ones still visible. In the Second World War, the military asked the statistician Abraham Wald where to armor its bombers. The returning planes showed holes on the wings and the tail. Wald put the armor where the holes were absent, on the engines, because planes hit there never came home to be counted. Every success-habits article is a returning plane. The engine hits are in the postmortems nobody publishes.

So read winners with the luck subtracted. Copy the parts of a win that repeat across many winners, and ignore the parts unique to one, because unique parts are usually the luck. Study the median failure harder than the outlier success. Avoiding the standard ways to die pays better than imitating one survivor's quirks, and the failure list is shorter.

Act forward like an owner, audit backward like an accountant

Machiavelli compared fortune to a river that floods. Nothing stands against the water in flood season, and everything depends on the levees built while the river is low.

The three lucks compress to one operating rule, and the rule is levee work. Forward, act as if everything is in your control. Backward, give luck its true share of the outcome. The forward stance preserves agency, the only input you steer. The backward audit preserves calibration, the only way your bets improve. Most people run the rule in reverse. They take credit for the past and surrender the future to fate.

I do not know which of my own wins were skill, and the written record is the only instrument I trust to find out.

The protocol fits on an index card.

Before any hard-to-reverse decision, write four lines. Name the call, the odds you expect, the failure you already see, and the condition that reopens it. Date it, and read it when the outcome lands, before you assign credit.

Once a quarter, audit the table. Name the game you are in, the shape of its payoff curve, and the wave underneath it. If the curve is flat and your ambition is not, no quality of play fixes the mismatch.

Every week, buy surface area. Accept one invitation you defaulted to declining, and count the unplanned encounters at the end of the month. A count of zero means the newspaper in your hands has a message in it, and you never stopped counting photographs.