The Lab·47 machines·12 wings
Ideas you can hold.
Forty-seven machines in twelve wings. Each one turns an idea into a decision you make yourself. One of them will remember you tomorrow.
Three ways in
Pick by the time you have.
-
An evening
Three deep simulations. Twenty minutes each.
Real companies, valued from their filings, live in Investing.
Twelve wings
Jump to any wing, or keep scrolling.
- Wing I, The marquee8 machinesThe machines people talk about afterward, from your life in weeks to a bomber full of holes.
- Wing II, The boardroom6 machinesMake the calls Blockbuster, Southwest, McDonald’s and Coca-Cola made, six more from Intel to Netflix, then a strategy-firm final round.
- Wing III, Capital11 machinesEleven machines about money: how it grows, shrinks, streaks and stampedes, what a fund’s fees do to one dollar, and what a short or an option costs while you wait.
- Wing IV, The mind4 machinesYour brain ships with known bugs, and these four machines find yours.
- Wing V, Other people3 machinesTrust, coincidence and contagion: what changes when other people are in the game.
- Wing VI, The stadium3 machinesThree calls where the math and the crowd disagree: a penalty, an empty net and a fourth down.
- Wing VII, Time2 machinesThe budget that never refills: what your memory can hold, and what you hand to a machine.
- Wing VIII, Twenty slots1 machineOne machine that keeps score for life. Twenty punches. No undo.
- Wing IX, The pit1 machineThe roulette table, back after closing with a gentler rule: every bet covers the losses plus $10. Watch it double anyway.
- Wing X, The corner office2 machinesRun the whole firm: a levered fund in the week the trade turns, and a buyer of tired software with a debt facility.
- Wing XI, The foundry4 machinesBuild the firm before you run it: Rockefeller’s refinery and Carnegie’s mill, a 12-week startup, the four machines that set modern prices, and an AI lab one model behind.
- Wing XII, The last mile2 machinesWinning the argument is the start: a campaign that needs votes in the right states, and a cell therapy that must survive the factory, the price and the drive to the hospital.
Wing I · 8 machines
The marquee
The machines people talk about afterward, from your life in weeks to a bomber full of holes.
Your life in weeks
A 90-year life, one square per week. Set your age. The dark squares are spent. The pulsing square is this week.
Every plan you are postponing lives in the empty squares. Count them.
The three doors
In 1990 Marilyn vos Savant printed the right answer to this puzzle. About 10,000 readers wrote in, nearly 1,000 of them with PhDs, and most said she was wrong. Play the game they argued about. The host knows where the prize is.
“Three doors. Behind one, a 100× return. Behind two, nothing. Pick, and I’ll show you a little secret about one of the others.”
Why switching wins, explained so a five-year-old nods
You pick door 1. Your door holds 1 chance in 3. The dashed box, doors 2 and 3 together, holds the other 2 chances in 3. Nobody argues with this part.
Monty opens door 3. Empty, and he knew it; he was never allowed to open yours or the prize. The box does not lose its 2⁄3 just because he showed you which half of it was junk. All of it now sits on door 2.
The real choice: keep 1 door, or take both of the others. Switching means taking both doors, and Monty has already opened the empty one for you.
Two rounds out of three, your first pick was wrong, and on exactly those rounds the door Monty leaves shut is the prize. When Marilyn vos Savant printed this in Parade in 1990, about 10,000 readers wrote in, nearly 1,000 with doctorates, and most said she was wrong. Many readers changed their minds after running the game themselves. The tally above does the same job.
Still arguing? Play it with a hundred doors.
Same rules, bigger room. One prize, ninety-nine empties. Pick any door.
“Go on. Any of the hundred.”
Watch Monty open 98 of 100 doors and the switch stops feeling like a coin flip. Three doors is the same game, smaller.
Six in a row
Monte Carlo, August 18, 1913. Black came up 26 times in a row, and gamblers lost millions of francs betting that red was due. Bet the same hunch with a doubling system, $1,000 and $10 chips.
Brief
The table is rebuilt: a European wheel with one green zero. The system waits for six of one color, bets the other, and doubles after every loss so the first win puts you ahead. Find out who ends up with the money.
“Table’s open. The system needs six in a row before a single chip moves.”
Seven losses deep you have pushed $1,270 across the felt and the system asks for $1,280 more, all to win the ten dollars you wanted at the start. A $1,000 bankroll dies on rung seven. And the green zero belongs to nobody: not red, not black, not odd, not even. When it lands, every bet on the table loses at once.
After six blacks, or 26, red is still 18 in 37. Doubling only chooses how you lose: often and small, or once and completely.
Fold to the Moon
Paper is a tenth of a millimeter thick. Fold it in half, then again. The slider stops at 42. Guess where that lands before you drag.
Each fold doubles the stack. Fold 27 passes Everest, and fold 42 passes the Moon.
A billion seconds
Million, billion and trillion sound like neighbors. Count each one aloud, one number per second, and see when you would finish.
A million seconds is 11.6 days; a billion is 31.7 years. That is the distance between a millionaire and a billionaire.
Powers of ten
One slider, 41 orders of magnitude, 12 stops. Drag from a proton to the observable universe and find yourself on the ruler.
You are here. Hello.
A small universe
Conway’s Game of Life runs on four rules and no ruler. Paint some cells, press Run, and watch patterns nobody designed take shape.
Draw on it. The rules ignore what you meant.
The armor problem
1943. Bombers return from Europe full of bullet holes, marked on the plan below. Armor is heavy, so you can protect only one part of the plane. Choose.
Count where the holes are. Then ask which planes never came back to be counted.
Wing II · 6 machines
The boardroom
Make the calls Blockbuster, Southwest, McDonald’s and Coca-Cola made, six more from Intel to Netflix, then a strategy-firm final round.
The fifty million dollar email
Dallas, 2000. You run Blockbuster: about nine thousand stores and a river of late fees. Netflix, a money-losing DVD-by-mail startup, offers to sell itself to you for $50 million. Face every fork Blockbuster faced, with what it knew then.
Every scene here happened. Marc Randolph, who was in the room, wrote that Blockbuster’s team fought not to laugh.
The ten minute turn
Texas, 1972. Southwest is short of cash and sells its fourth plane but keeps the four-plane schedule. You run ground operations. A plane earns money only in the air. Make three planes fly four planes’ schedule.
At the industry’s 45-minute turn, three planes fly 24 legs a day and the schedule needs 32. Drag the turn down and find the airline.
You are not in the burger business
Chicago, 1956. McDonald’s is franchising fast and nearly broke: a $950 fee and a 1.9% royalty barely cover the lights. Harry Sonneborn, a former Tastee-Freez finance man, has an idea that becomes the company. Take the quiz first.
Where does McDonald’s make its money today?
The sip test
Atlanta, 1984. Pepsi is winning blind taste tests on television, and Coca-Cola is losing share. Your sweeter new formula beats Pepsi and your century-old recipe in blind sips. Run the tests, then make the call Coca-Cola made.
The counters are waiting. Your researchers have the cups poured.
Every metric is a proxy. The tests removed the brand to stay objective, and the brand was the product.
Make the call
Six hard calls, from FedEx in 1973 to Netflix in 2011. See what they knew and choose. Then see what they did and what it cost.
Laying out the case files.
Final round
Four cases from a strategy-firm final round. Find the number that matters, say what it means, and stop.
Brief
The cases: a cinema chain whose profit fell while revenue rose, a European grocer weighing the American South, a buyout fund with a week to price a chain of animal hospitals, and a startup with no idea what to charge. Play the question back, build the issue tree, pull the exhibits, do the math in your head and give the answer first. You are scored on structure, analytics, judgment and communication, with a reason for every score. The firm, the clients and the case numbers are fictional. The public facts in each case file are real.
Setting up the interview room.
The interviewer is grading one habit: the answer first, then the math that holds it up.
Wing III · 11 machines
Capital
Eleven machines about money: how it grows, shrinks, streaks and stampedes, what a fund’s fees do to one dollar, and what a short or an option costs while you wait.
The compounding machine
Two people start today. One gets 1% better each day, the other 1% worse. After a year the first is 37.8 times better. The second has 2.6% of what they started with.
For weeks the two curves look the same. That flat early stretch is where most people quit.
The shrinking hundred
At 3% inflation, $100 buys about $55 of goods after 20 years. Set your own rate and horizon.
Rule of 72: at 3%, prices double in 24 years. Cash in a drawer does not.
The honest coin
A fair coin, flipped 100 times. Count how long the streaks run, then think of every chart you have been told a story about.
Press the button. The coin has no memory, and you will still see patterns in its streaks.
The patience market
A market that exists only on this page. Buy in, watch it move, and sell when your nerve runs out. After you sell, the chart keeps going and shows you what leaving cost.
This machine measures temperament in seconds.
Venture odds
Most startups return nothing, and a rare few return the whole fund. Spread your fund across 1 to 20 bets, then run 1,000 futures and see how the decade could have gone.
One bet is a lottery ticket. Twenty bets carry the same odds with a far narrower spread of outcomes.
Follow one dollar
A pension fund writes a check. Ten years later, some of it comes back. Here is the trip, and who gets paid along the way.
Wiring the fund.
The herd
Two hundred and forty traders, each watching their neighbors. The dial sets how much they copy each other. Turn it up slowly and watch a calm market learn to stampede.
At low herding, the crowd’s errors cancel and the price drifts. Push past 70% and the traders move as one.
The rich get richer
A network grows one newcomer at a time. In the fair world, newcomers link at random. In the real one, they link to whoever already has the most links. Run both and compare the maps.
This rule, preferential attachment, shapes citation counts, follower graphs and platform markets. Nobody planned any of them.
The generous coin
A fair coin. Heads, you win $150. Tails, you lose $100. One flip, real stakes. Decide now.
On average this coin pays you $25 a flip. Kahneman and Tversky found that losses weigh about twice as much as gains.
The short squeeze
You are short 100 shares at $100, with $10,000 in the account. The trade is probably right. Cover whenever you like. Wait too long and your broker covers for you.
A long position can lose what you paid. A short can lose more than you have.
The option clock
You bought a 30-day call: the right to buy a $100 stock at $105. It cost $320. With the stock below $105, every dollar of that price is time. Press start and watch it melt.
The gray bar only shrinks. The seller collects it while the clock runs.
Wing IV · 4 machines
The mind
Your brain ships with known bugs, and these four machines find yours.
The marshmallow clock
The famous test, rebuilt for adults with cursors. Press the button and hold it for 30 full seconds. The host will keep talking the whole time.
Mischel’s Stanford children waited up to 15 minutes, and a 2018 replication found the link to later life much weaker than first reported. You only need 30 seconds.
The reaction tax
Click the box, wait for it to change color, then click again as fast as you can. Your number is what you bring to a race against trading machines.
A fast human needs about a quarter of a second. Exchanges match orders in microseconds.
The gauntlet
Five rooms, each with one correct exit and several confident wrong ones. Your reasoning is the only equipment. The score comes at the end.
Room 1 is a pitch meeting. Go in.
Predictions, checked
Seven famous forecasts, each with a date and a source. Call each one hit or miss before you see what happened.
Pulling the clippings.
Wing V · 3 machines
Other people
Trust, coincidence and contagion: what changes when other people are in the game.
The trust machine
Ten rounds against a hidden opponent. Each round you both choose: cooperate or betray. Both cooperate: 3 points each. Betray a cooperator: 5 for you, 0 for them. Both betray: 1 each. The machine has a strategy. Find it.
Round 1 of 10. The machine moves when you do.
The birthday party
How many guests until two of them probably share a birthday? Most people guess far too high. Drag the slider, watch the odds, then throw the party and meet the twins.
Twenty-three guests give better than even odds: 50.7%. Trust the arithmetic over the instinct.
The threshold
A forest, a match and one dial: how densely the trees are planted. Below a certain density, fires starve. Above it, nearly everything burns. Set the density, plant, then click anywhere to strike the match.
Physicists call the line near 59% the percolation threshold. Epidemiologists watch the same tipping point at R₀ = 1.
Wing VI · 3 machines
The stadium
Three calls where the math and the crowd disagree: a penalty, an empty net and a fourth down.
The penalty spot
You are the goalkeeper, facing 10 penalties. Before each shot, dive left, dive right or hold the middle. In a study of 286 penalties, about three in ten went down the middle, and keepers dove 94% of the time.
Bar-Eli called it action bias: a goal conceded after a dive feels better than one conceded standing still.
The pulled goalie
Hockey, down one goal, clock running out. Pulling your goalie adds an attacker and leaves an empty net. Tradition pulls at about one minute. Pick your moment, play 500 games, and compare yourself to tradition.
Losing by one and losing by two pay the same: nothing. Once you accept that, the goalie comes out earlier.
Fourth down
Three fourth-down calls of the kind NFL coaches face every Sunday. Make each call, then see what expected points said. In 2006 the economist David Romer showed that coaches kick far more often than the math supports.
Your headset is on, and 60,000 fans will boo the correct call.
Wing VII · 2 machines
Time
The budget that never refills: what your memory can hold, and what you hand to a machine.
The vanishing number
A number appears for a few seconds, then vanishes, and you type it back. Every round adds a digit. Most people hit a wall around seven.
George Miller put the span of working memory at about seven items, plus or minus two, in 1956.
The delegation dial
Turn the dial and hand your daily choices to the machine, one notch at a time. Every notch is more convenient. Read the list at the end.
All yours. Heavy, isn’t it.
Wing VIII · 1 machine
Twenty slots
One machine that keeps score for life. Twenty punches. No undo.
The punch card
Buffett tells students to imagine a card with twenty decisions for a lifetime. This one is yours. Each punch is permanent, and this page remembers it.
No undo.
Wing IX · 1 machine
The pit
The roulette table, back after closing with a gentler rule: every bet covers the losses plus $10. Watch it double anyway.
The covered ladder
Wait for six of one color, then make every bet cover your losses plus $10. Watch that sensible rule turn into the doubling ladder from Experiment 03.
Brief
Lose $10 and the rule asks for $20. Lose that and you are down $30, so it asks for $40. Then $80, $160, $320, $640. Each bet roughly equals the whole hole, so the hole doubles and the bet doubles with it. Nobody chose to double. Coverage chose it. The martingale is the cheapest ladder that keeps the promise, and this table proves it to the dollar. A bigger bankroll never beats the wheel. It only chooses which rung ruins you, how rarely, and how much you post as collateral to chase $10.
“After hours. Same wheel, same six-in-a-row rule, one iron demand: every bet covers everything lost, plus ten.”
Every rung keeps the promise: win anywhere and the whole ladder ends exactly $10 ahead. And read the stakes out loud: 10, 20, 40, 80, 160, 320, 640. They are the doubling ladder from Experiment 03, digit for digit, because covering the hole plus ten is doubling, whether you meant it or not. Doubling the bankroll buys one more rung: the busts get rarer, the craters get deeper, and the average gets slightly worse, because deeper ladders push more money across the felt and the wheel taxes every dollar of it. Staying power multiplies the sign of your edge, and on this felt the sign is negative. The green zero still belongs to nobody.
No ladder of any shape changes 18 in 37. The wheel keeps 2.7 cents of every dollar.
Wing X · 2 machines
The corner office
Run the whole firm: a levered fund in the week the trade turns, and a buyer of tired software with a debt facility.
Conviction: the margin call
You run a levered AI fund. Six trading days, a prime broker asking for more collateral. Being right is not enough.
Brief
You founded an AI-focused hedge fund: long the companies that build intelligence, short the software it replaces. The thesis returned triple digits, and you levered it. On Monday before the open, three things are on your desk: an overnight market briefing, a memo from your risk officer and a margin notice. To win you have to stay solvent, liquid and in control long enough for being right to matter. Inspired by a real fund’s reported July 2026 drawdown. Everything playable below is a labeled simulation, not a trading record.
Case file · the real story and where the fiction begins
Reported facts: Reuters reported that Situational Awareness, the AI-focused fund founded by Leopold Aschenbrenner, lost about 67% in July 2026 during an AI stock rout while remaining up roughly 80% for the year, sold substantial public equities with Citadel among the buyers, and removed its leverage (Reuters, July 31, 2026; Reuters on the Citadel purchase).
Everything else here is fiction: the tickers, prices, brokers, characters, dialogue, financing terms, deadlines, and every number the game generates. The starting book, margin rules, and market paths are labeled model assumptions, not a reconstruction. Your playthrough is a counterfactual; the game does not force the reported 67% on you, and no outcome here says anything about what the real fund should have done.
Model assumptions
Synthetic book: $2,800M long across six AI-infrastructure names, $600M short software, $100M private, $100M central cash, $600M short proceeds held at brokers, $2,000M financing debt, NAV $1,000M. Gross public exposure 3.4x NAV, net 2.2x, debt 2.0x; these are three different numbers. Positions, proceeds, and debt split across two fictional prime brokers. Illustrative maintenance: 20% on longs, 30% on shorts, add-ons disclosed when applied. Sale proceeds count as collateral at the executing broker immediately but transfer elsewhere next day. Prices follow a seeded two-factor model (market and AI sentiment) with regime shifts; your orders pay spread and impact that grow with size, scarcity, and stress, and large positions cannot exit in one tick. Same seed plus same actions reproduces the same week.
The trade only needs to be right eventually. The fund needs to be alive on Thursday at 3 p.m.
The acquirer
A company in Milan got rich buying tired software people still love and running one hard playbook on each. Run a fictional version: eight years, $50M of your own money, a debt facility and a market of faded products.
Brief
The playbook: move each product onto shared technology, cut costs, including some users will notice, and raise prices on the people who stayed because they love it. The cash funds the next deal. Every lever is a trade-off. Price increases spend loyalty, cuts risk quality, a replatform costs money now to save money later, and each new product makes the shared platform cheaper for the rest. Build the empire before the debt builds a case against you.
Height is revenue. Glass is debt, shared out by revenue. The dashed ring is the covenant: debt at 1.3 times revenue. Drag to turn the street; tap a building to read it.
Case file · the real playbook this is modeled on
The real story: Bending Spoons, founded in Milan in 2013 and led by CEO Luca Ferrari, buys mature digital products with loyal audiences that their owners under-invest in, then applies a standard playbook: migrate the product onto its central technology platform, cut operating costs deeply, optimize pricing and paywalls, and run product, marketing, and technology through centralized teams. Acquisitions include Evernote (2023), Meetup, WeTransfer (2024), komoot and Brightcove (2025), Vimeo (agreed 2025, about $1.38B), and AOL (agreed 2025), with deals financed substantially by debt facilities; the company raised funding in 2025 at a valuation around $11B. Everything you can click above is fictional: the companies, numbers, elasticities, events, and outcomes are model assumptions built to teach the shape of the strategy, not a record of any real transaction.
The Milan playbook bets that an old product, bought right and run hard, can outgrow a new one. It works only while users love the product, and every pull of a lever spends some of that love.
Wing XI · 4 machines
The foundry
Build the firm before you run it: Rockefeller’s refinery and Carnegie’s mill, a 12-week startup, the four machines that set modern prices, and an AI lab one model behind.
The titan’s ledger
Play Rockefeller from a Cleveland refinery in 1868 to the Supreme Court in 1911, or Carnegie from his 1872 bet on steel to the 1901 sale to J.P. Morgan. The machine keeps two ledgers, and only one is money.
Brief
Almost every crossroads is real and sourced: the secret rebates, the Cleveland buyout, the pipelines, the trust, Homestead, the Mesabi ore and the sale. The levers are the ones they pulled: cost, scale, price, integration, rebates, labor and law. The roads they did not take are open too.
Opening the ledgers.
Kerosene fell to about six cents a gallon and steel rails to under $20 a ton. Every line in the other ledger is also true, and this machine keeps both books open.
The batch
Y Combinator gives each company $500,000 on two SAFEs, three months and one number to watch: weekly growth. Pick an idea and a cofounder, survive 12 weeks, raise after Demo Day, and find out whether you are default alive.
Brief
Each week you choose between talking to users and building, launching ugly and launching late, the early check and Demo Day. The program facts and the dilution math are YC’s own and sourced. Every company and person in it is fiction. Make something people want, and do the arithmetic before you sign.
Clearing a desk for the batch.
Twelve weeks is long enough to build the wrong thing well, and long enough to find ten people who would be very disappointed if you vanished. Demo Day grades only the second.
Four machines that changed what things cost
Prices are set by machines most people never see. Turn each one over and take it apart.
Wheeling in the machines.
Frontier: the scaling bet
Run Aldermoor, a fictional AI lab one model behind the frontier, with $5.0B and 40,000 GPUs. Eight half-years.
Brief
The levers are the industry’s own: compute for the training run or for research that makes the next run cheaper, a bigger model or a leaner one that is cheaper to serve, which compute contracts to sign, what to charge per million tokens, how much safety work each capability level demands, and whose money to take. Staying at the frontier is a spending decision. Staying trustworthy there is a series of them. The lab, its rivals and every number are fiction. The case file keeps the real history of OpenAI, Anthropic and the rest separate, sourced and neutral.
Booking the compute.
Ten times the compute buys the same step, every time. The scaling law never says who pays for the tenfold, or what you promised on the way.
Wing XII · 2 machines
The last mile
Winning the argument is the start: a campaign that needs votes in the right states, and a cell therapy that must survive the factory, the price and the drive to the hospital.
Two seventy
Two invented parties, the real 538-elector map. Nine weeks, a finite war chest, polls that can miss together. You need 270, not a majority.
Brief
The map has 538 electors, apportioned by the 2020 census. You need 270, and Maine and Nebraska split theirs by district. Your candidate can be in three places a week. Persuasion fades in days; turnout takes weeks to build. Survive the debates and the October story, then sit through an election night where the early lead mostly tells you who counted first. The parties, candidates and map politics are fiction. The rules, the arithmetic and the history in the case file are real.
Printing the map.
The electoral college asks where your voters live, one state at a time. The first ticket to 270 wins.
The living drug
CAR-T is a living drug: a patient’s own T cells, collected, flown to a factory, given a new receptor, grown and returned. Choose the target, design the receptor, build the factory, run the trial and set the price.
Brief
In the phase 1 trial, the toxicity you cause comes from the same biology as the cure. Every lever is a trade-off that real design reviews argue about, and none has an obvious answer. Educational only. This is not medical advice, the numbers in the model are labeled assumptions, and nothing here should inform a decision about a real person.
Warming the incubator.
A pill is a molecule. This drug is alive and runs on a delivery schedule, and the schedule is what usually kills the program.
Forty-seven machines. Start with one.
You walked all 12 wings.
The essays these machines came from are at the front door. The rest of the collection is in the World. The chalk wipes clean. The punch card does not.