Fermi Problems: How Many Piano Tuners in Chicago?

At dinner last month I asked how many books sit on our shelves, and the table split into campaigns. My daughter said three hundred, my son said a thousand, and both defended their numbers like property. We counted the next day, shelf by shelf, and landed at four hundred and twelve. The winner was not the closer guess. The winner was the method my daughter used: so many shelves, so many books per shelf, multiply. She had estimated without knowing the word for it, and the shelf-by-shelf count proved her chain nearly right.

That chain has a famous name. A Fermi problem asks something nobody could know exactly, how many piano tuners work in Chicago, and answers by multiplying rough guesses until the errors cancel. Physicist Enrico Fermi was its patron saint: at the Trinity atomic test he dropped scraps of paper in the blast wind and estimated ten kilotons, against the now-accepted twenty-one, right within a factor of two from confetti (Wikipedia, Fermi problem). This article runs his game three ways, then hands it to the family.

We counted the next day, shelf by shelf, my son tallying while my daughter called numbers, and landed at four hundred and twelve against her chain's four hundred flat. Twelve books off on four hundred is three percent, from guesses that felt like jokes at dinner. The boy who said a thousand paid for ice cream. The shelf method has held the dinner crown ever since.

The piano tuner question

Chicago holds about three million people, so roughly a million and a half households at two apiece. Say one household in twenty owns a piano that gets tuned yearly: seventy-five thousand tunings a year. A tuner manages two a day across two hundred and fifty working days, five hundred tunings a year. Seventy-five thousand divided by five hundred is one hundred and fifty tuners. Every number in that chain is a guess, some off by half, and the answer still lands in the low hundreds, where the true count lives.

Notice what the chain never needed: a directory of tuners, a single phone call, one hard fact beyond the city's population. Each step converts an unknowable into two guessables, and the guesses are chosen where ordinary life supplies them. Households per person you know. Pianos per street you can picture. Tunings per day any craftsperson can sanity-check. Fermi problems are won in the splitting, not the guessing.

Audit the answer by asking whether it earns a living. Five hundred tunings a year at, say, $150 each is $75,000, a plausible craft income, neither starvation nor fantasy. If the chain had said fifteen tuners, each would need ten times the work or ten times the price, and the absurdity would show. Every Fermi answer deserves this second division: does anybody get paid, fed, or housed by it. Numbers that cannot survive their own economics are confessions, not estimates.

Rough times rough stays fine

Five sloppy guesses should make slop to the fifth power, yet they do not, because errors point both ways. An overestimate here meets an underestimate there, and multiplication lets them shake hands. The mathematics underneath: multiplying guesses adds their logarithms, so the misses wander like a random walk instead of marching off a cliff. A nine-step chain where every guess is off by double still lands within a factor of eight of truth, against a worst case of five hundred twelve. Cancellation is not luck. It is arithmetic with a forgiving shape.

The one way to break it is bias, every guess leaning the same direction. Five optimistic guesses multiply into a fantasy, five fearful ones into a panic. The defense is deliberate bracketing: for each step, name a low you are sure sits below truth and a high you are sure sits above, then take the middle. Brackets force the errors to straddle, and straddled errors cancel. Our big numbers guide teaches the language this answer speaks, orders of magnitude. Fermi chains are how you earn the right to speak it.

Think of nine guesses as nine coin flips. All nine landing optimistic is one chance in five hundred twelve, which is why the typical result clusters near truth while the nightmare stays theoretical. Flip honest coins and heads pay tails' debts. Load every coin the same way and the ledger never balances. Bracketing is how you keep the coins honest when your hopes are not.

Answer in ranges, not digits

A Fermi answer is a neighborhood, not an address: about one hundred fifty tuners, surely between fifty and five hundred. The range carries the honesty the digits cannot, because the chain's error budget is visible in it. Anyone who answers "one hundred fifty-two" has misunderstood the game; the twos digit is confetti. Our rounding guide sharpens numbers you already know. Fermi ranges bound numbers nobody knows, and the two skills meet exactly at the bracket.

Ranges also decide before precise work begins. Want to sell piano-tuning equipment and need ten thousand customers? The tuner estimate sits two orders of magnitude below, so the plan dies on a napkin instead of in a lease. Scientists run the same pre-check: a rough frame first, and if the careful calculation lands far outside it, the assumptions get audited, not celebrated. Our scientific notation guide handles answers too huge for commas, grains of sand and stars. Fermi chains are what you run before you need it.

Report every Fermi answer in the same uniform: middle, then walls. About one hundred fifty tuners, walls at fifty and five hundred. The middle is the headline and the walls are the honesty, and anyone skipping the walls is selling precision the chain never owned. Children learn the uniform in one evening and police it forever; my son now demands walls on my own claims at dinner, which is justice of a sort.

Three family games

First, the shelf count we already played: shelves times books per shelf, then count and compare. Second, the rice jar: layers times grains per layer-row, then cook it for dinner and count nothing, because the ratio of guess to truth is the prize, not the rice. Third, Friday pizzas in town: households times pizza Fridays per month divided by slices, checked against the pizzeria's own boast if they will share it. Three games, three chains, three brackets, and children who start splitting unknowables on sight.

Our guess-and-check guide guesses answers and tests them. Fermi inverts the habit: guess the inputs, multiply, and let the answer inherit the honesty. Same family, opposite direction, and the children who learn both stop fearing questions without given numbers.

Fourth game, for bath night: how many buckets fill the tub. Length times width times depth in bucketfuls, each dimension paced off in bucket widths, then run the tap and count. Volume is just three chained guesses wearing swim trunks, and the overflow line judges the brackets without mercy. Ours missed by one bucket, high, because nobody brackets splashing.

When Fermi fails

Honesty requires the failure modes. A chain is only as narrow as its widest bracket: if one guess spans a factor of a hundred, the answer spans it too, and four sharp guesses cannot rescue one foggy giant. Split the giant or admit the fog. Correlated guesses fail together as well: five guesses all fed by the same wrong newspaper err as one, and no cancellation rescues a single poisoned source. Independent guesses from independent life experiences are the requirement, not the decoration.

The one habit: estimate first

Before any calculator, any spreadsheet, any asked expert, run the chain on scrap paper. Five lines, five guesses, one bracket each, one multiplication. Then do the precise thing, and compare. When they agree, confidence is earned. When they differ by a factor of ten, the error is found before it ships, the way Fermi's paper scraps checked an atomic bomb before the instruments reported. My daughter now opens estimation fights at dinner on purpose. The shelves made her dangerous.

Teachers can run the same game on a board: every child brackets silently, the class median becomes the chain's guess, and the class average of brackets beats almost every individual. Thirty biased coins, averaged, behave like one honest coin. The lesson lands twice, once in the answer and once in the averaging.