The Game Is Rigged

I'm an autistic Black kid from the outskirts of Dallas, Texas. Hear me out.

When I was a kid in Frisco, Preston Road was a dirt road. Frisco had about six thousand people in 1990. The 2020 census counted more than two hundred thousand. I watched a town turn into a city from the inside, and it taught me early that the ground under a place can change completely inside one childhood while the adults standing on it tell you nothing's moving.

The ground moving while the adults said it wasn't was the first thing I knew about the world, before I had words for it. The second thing I knew was that I didn't fit in it.

I design data platforms: the plumbing that takes everything a business knows, scattered across forty tools and three people's heads, and turns it into something a person or a machine can reason over. Before that I was a performance marketer, and I still am. The work is programmatic advertising and conversion rate optimization, which in plain terms means getting the right strangers to show up and then getting as many of them as possible to say yes, for as little money as possible. My edge in that business came from treating it like an engineering problem while a lot of my peers treated it like an art form, and the engineering won often enough to pay for my mistakes.

I've been poor for most of my life and rich a few times. I've sold knives and soap door to door in Dallas, worked years in nightlife, driven rideshare for a living, lived in more than twenty countries, and had stretches with no address at all. Across all of that I've interacted with somewhere between fifty and eighty thousand people, and I've had real conversations, even short ones, with around ten thousand of them.

This memo's working title was "Being poor and being dumb are both mostly your fault." I wrote it to start a fight, and I also believe a version of it. Starting a fight and believing something are different things, so before I published that claim I did what I do with an ad hypothesis before I spend real money on it: I tried to kill it.

I ran the premise through eight rounds with a research engine I'd told to argue with me. Each round built on what the last one taught me, and I checked every study it cited for what the study actually found. I pointed two of the rounds at my toolkit instead of the premise, and those went worse for me than the ones about the argument. I lost some of the eight rounds. One claim I was planning to use turned out to be a replication failure. One of my framings got called an overstatement even after I'd softened it once. One citation the engine handed me had the wrong authors attached, and I only caught it because I went back to the journal. You'll see every loss in here, because a memo that only reports the rounds its author won is marketing.

Years into untangling my life, I wrote a line for myself on the whiteboard in my room:

Both halves are true at the same time, and most people I've met only manage to hold one of them. The ones who hold only the first half stay stuck and feel righteous about it. The ones who hold only the second half stay stuck and hate themselves for it. I've lived long stretches in both camps, and neither one ever got me a single dollar or a single IQ point.

So I'll start with the first half, because I'm about to agree with every person who read the working title and wanted to throw their phone.

Then I'll show you the machinery I actually use: the ladder I run every operation on, the instruments I read a situation with and which of them are scientifically worthless, the four shapes every decision comes in, and the numbers I'm holding myself to this year. A memo that tells you the next move is yours and never shows you how a move gets picked is a poster.

In Dungeons & Dragons you make a character by rolling four six-sided dice for each of six ability scores (Strength, Dexterity, Constitution, Intelligence, Wisdom, Charisma), throwing out the lowest die, and adding up the other three. You keep whatever comes up. Nobody at the table blames you for rolling a 7 in Charisma. They might laugh at you, and you'll deserve that part, but nobody thinks you chose it.

I rolled Black in a small Texas town, in the years I grew up there. I rolled autism before anybody around me had a word for it. I rolled a household where calm was the scariest weather on the forecast, because calm is what comes right before it isn't.

I won't itemize any of it, because an itemized list turns into a misery résumé and nobody learns anything from those. Those rolls ship with a standard package: the conversation before you leave the house, the store where you get followed, the traffic stop where your hands are the variable, the one thing the kids at school find that you can't hide. I got the standard package, and plenty of people got a heavier one. I mention it only so you know I'm not typing this from a seminar stage in a rented suit.

I didn't choose any of that. Neither did you, whatever your version is.

The data on what those dice do is some of the best social science has ever collected, and it's brutal.

A quarter of a country got measurably smarter because somebody changed what went into the salt shaker. Who earned that? Nobody in that quarter did a single push-up for it. And a whole generation lost a couple of IQ points each because the car in front of them at the stoplight was burning leaded gas, which nobody in that generation chose either.

Not one line of that table was chosen by the kid it happened to. Where you were born, what you breathed, who raised you, what the cops saw when they looked at you, and what the kids at school decided you were before you opened your mouth all have real, measurable, and in several cases causal effects on your income and your test scores. Anybody who tells you otherwise is selling something, and it's usually a $997 course filmed in front of a rented Lamborghini.

I spent a stretch of my early twenties cold-calling people about insurance, and insurance has the cleanest definition of fault I know: fault is who caused the collision. The adjuster doesn't ask who designed the intersection, who put the stop sign behind a tree, or who decided that particular road didn't need a streetlight. He finds the car that hit the other car and assigns the number. If that's what fault means, my working title is false as a claim about the starting hand, and conceding it took about one sentence, not a research engine.

The law has no-fault divorce and no-fault insurance, and therapy has a whole vocabulary built around the words "it's not your fault." The one place almost nobody has managed to run a no-fault system is inside their head, where the only two verdicts available are total innocence and total guilt, court never adjourns, and nobody gets a defense attorney.

So what survived?

Two crowds will fight about the original premise in the comments, and I've spent enough time in both to know their scripts by heart.

The first crowd says everybody who's broke chose it: get up earlier, stop buying coffee, read a book, and by the way, here's my course. This crowd ignores that table completely, which is impressive, because the table is mostly the work of economists at Harvard and the Census Bureau, not some activist with a podcast.

The second crowd says it's the system. This crowd is right about the table and then stops, as if being right about the cause of your situation were the same thing as being out of it. I know that stop well. I lived there for years and it was warm, and nobody there ever asked me to do anything uncomfortable.

Both crowds are answering a question about the past: whose fault is it? One says you, the other says them, and they'll argue about it until the heat death of the universe. Neither answer tells you what to do tomorrow morning. "Whose fault" is the first-level question. The second-level question goes: given the hand I've actually got, what's the best move available to me right now, and who's the only person who can make it?

I walked into the first research round with a softened thesis. I'd already backed off from "mostly your fault" to "mostly yours to move," and I figured that version would survive. It didn't. The engine said even that overstated individual control, and then it handed me a sentence I've repeated to myself ever since:

Anyone who has built an optimization model knows this shape. There's a region defined by constraints you didn't pick: budget, capacity, physics, law. Call that region the box. Inside it you choose a point. The constraints are real and you don't get to argue with them, and you still choose the point, and the gap between a good point and a bad point inside the same box is usually enormous. Some points even move the walls. Moving before thirteen moved a wall. An extra year of school moves a wall. Getting into a room with people who have more than you moves a wall.

The research beat the original claim into something more precise. The dice are not your fault. The next move is. For the rest of this memo, fault belongs to the move and never to the roll.

Then why not say "responsibility" and spare everybody the aggravation? Because "responsibility" is the polite word, and polite words let you nod along without anything happening in your body. "Fault" stings. And where it stings is diagnostic.

I've spent years mapping how people get stuck, first for customers and eventually for myself, and they almost always get stuck in the same loop. Something painful happens. A fear gets installed to make sure it never happens again, and a fear is an investment, and most people run a terrible portfolio. The fear drives avoidance. The avoidance produces a bad outcome. The bad outcome produces shame, which is different from guilt: guilt says I did a bad thing, and shame says I am a bad thing. Nobody can live with that second sentence out in the open, so it gets buried under denial, and then under blame: the parents, the economy, the ex, the algorithm, the government. When I first described this loop out loud, in the talk that became my essay Echolocation, I said the one thing a person in it can't do is take accountability, "because to take accountability would mean that you have to face that shame you just buried."

The word "fault" hurts because it lands right on the spot where the shame is buried, and landing there is also what makes it useful, because that spot is where the work is.

There's a trap on the other side of that loop too, and I think more people live in it than in the loop itself. I call it the cycle of tolerance. Almost nobody I've ever talked to is ignorant of their problem. They know exactly what's wrong, and they've built a whole life that accommodates it. Then they run self-flagellation instead of change. They beat themselves up every night, and it feels like paying the bill. Self-flagellation is the emotional equivalent of jerking off: strenuous, private, repeatable, and nothing is ever born from it. The payment is a subscription, because it comes due again tomorrow night and buys nothing but another day of the same arrangement. Shame dresses up as responsibility, and the person gets to feel serious about their problem while the problem stays exactly where it was. I did this for about a decade with my procrastination and got nothing out of it except a detailed vocabulary for how bad I was.

On the whiteboard, next to the loop, I wrote remove blame, increase clarity. Taking the blame off the hand you were dealt clears the fog so you can see where the actual move is. Crossing from the loop into accountability takes something else, and the word for it is courage, which is annoyingly old-fashioned and has no substitute that I've found. On the far side, the loop runs the other way: truth, then responsibility, which I defined in that essay as authorship, taking ownership of your reaction even to things that weren't your fault.

Good Will Hunting has the most famous version of the first half. Sean sits across from Will and says "It's not your fault" over and over until the armor cracks. That scene is about the hand, about what was done to a kid, and it's correct. Then the movie keeps going. The last thing Will does is get in a car and drive across the country toward a girl, which is a move nobody on earth could make for him. The film holds both halves, and most of the people who quote it only remember one.

Across the ten thousand or so real conversations I mentioned, I've rarely met anyone who reached adulthood without bleeding somewhere: through their family, their community, their culture, their body, or their mind. Some of it is acute and some of it isn't, and I don't want to flatten that, because the kid who grew up in a war zone and the kid whose parents were cold at dinner didn't roll the same numbers. But everybody's character sheet has damage on it. The job is to record the damage without letting it become your identity. And damage was never the variable that separated the people I watched move from the people I watched stay. The dice set the box, and people split on what they did inside it.

Say that out loud, and the first thing most people want to argue about is whether they're smart enough to do anything inside the box at all.

A data platform keeps a schema separate from a reading. A schema says what a thing is: this is a customer, it has a name, an account number, a date it signed up. A reading is a measurement some instrument took at some moment under some conditions: the thermostat said 71 degrees, the scale said 190 pounds right after lunch. You never write a reading into the schema as if it were the thing's permanent nature, because instruments drift, instruments have bias, and conditions change what they report. I wrote a whole series, The Shape of Data, on one consequence of keeping them separate: the shape you give a record decides which questions you can ever ask of it.

People treat an IQ score like a schema field: Intelligence: 92, typed onto you like your birthday. It's a reading. It's a real reading, it predicts real things, and I won't pretend it's astrology. It's still a reading, and the research on how much it moves is a lot more interesting than either the "IQ is destiny" crowd or the "IQ is fake" crowd lets on.

Start with heritability, the word everyone misuses. In 2003, Eric Turkheimer and colleagues looked at twins across the income spectrum in the United States and found that the heritability of IQ was around 0.10 in the poorest families and around 0.70 in the most affluent ones (Psychological Science). That's the opposite of what the "it's genetic" crowd assumes. In the poorest homes, genes explained almost none of the difference between kids. The environment was doing nearly all the work. Genes only got to show up once the environment stopped being the bottleneck. A later meta-analysis found that pattern strongly in American data and not reliably in Western Europe or Australia (Tucker-Drob & Bates, Psychological Science, 2016), which says something grim about how low the floor sits in America.

I wrote a field note called Space Crystals about a physics experiment in Italy that spent roughly 25 years hunting a real effect in empty space and came back with nothing but limits on how big the effect could be, because its magnets were too weak to make the structure visible. Meanwhile dead stars with fields billions of times stronger were broadcasting the effect for free. What I took from it is that when a reading comes back null, you ask whether the medium is empty or your field is weak. A kid tested inside a starved environment is a weak-field reading. The structure was there, and the field wasn't strong enough to show it.

So how far does a reading actually move? Look at what happened across whole populations. Through most of the twentieth century, average IQ scores rose by roughly 0.28 points a year, and faster on the fluid reasoning parts of the tests (Pietschnig & Voracek, meta-analysis, 2015). That rise is called the Flynn effect. Genes don't change on that schedule, and environments do: more school, better nutrition, smaller families, jobs and machines that ask for more abstract thinking. Then, in several countries, the rise stalled and reversed. In Norway, Bernt Bratsberg and Ole Rogeberg found the reversal shows up between brothers: younger brothers born into the later period score lower than their own older brothers (PNAS, 2018). They had the same parents, the same genes on average, and lower readings. I wrote a whole piece, "The 39-Bit Brain in a Petabyte World," about what that implies for a world that spends its evenings scrolling. The short version is that whatever is making us dumber lives in the environment, and the environment is something a person can do something about.

The evidence holds at the level of individuals too.

That last row is personal. As an autistic kid I could've ended up in the special class, drooling on myself, and I saw early what the life on that track looked like. The standard intelligence tests lean hard on following verbal instructions from a stranger in a room, which is close to a perfect test of the exact thing autism makes hard. Michelle Dawson, Laurent Mottron, and their colleagues gave autistic children both the Wechsler and Raven's Progressive Matrices, a nonverbal pattern-reasoning test, and the same kids ranked on average about thirty percentile points higher on Raven's. The brain was the same, the instrument changed, and a different kid showed up on paper. Which one is the real kid?

I've always been built unevenly. Simple life things are really hard for me, and I have a lot of fun with hard things other people wouldn't dare try. I couldn't tell you why. A test that averages the two describes somebody who doesn't exist.

An average that describes nobody is also my whole problem with how people use group averages on race. In marketing I spent years watching teams build campaigns around one average customer: Average Alex, 38.4 years old, 1.7 children, prefers educational content, a man who has never existed and has never bought a single fucking thing. It's the most reliable way I know to aim a budget at an empty spot. A gap between two group averages, whatever caused it and however contested it is, tells you close to nothing about the individual in front of you, and nothing about how far that individual's reading can move. I put it this way in Hyperrelevance Cartography, and I meant it for customers and for people: demography shouldn't quietly become destiny.

The research engine also made sure I didn't skip the ceiling. School raises everybody's scores, and it doesn't close the gaps between people on its own: one large study found schooling improved intelligence across the board and showed no meaningful interaction with family income or genetic propensity (Judd, Sauce & Klingberg, npj Science of Learning, 2022). And by adulthood, differences in measured intelligence line up substantially with genetic differences, with heritability estimates around 0.6 in many datasets. So I'm not telling you that you'll out-test everyone who started ahead of you. You might not. What the evidence does say is that your own reading moves, by amounts people have measured, when your environment and your inputs change. The gap between you and somebody else is a bad scoreboard anyway. The useful comparison is you against the you who didn't change anything.

A language model starts with pretraining: a huge base of patterns it didn't choose. Then it gets trained further on narrower data, and whatever it's rewarded for, it becomes. Humans run a rough version of the same process. Genes are the pretraining. Your family and your town were the first training set, and they came with a reward system nobody asked you about: what got you warmth, what got you hit, what got you left alone. You don't get to pick your pretraining, but you do get to curate your fine-tuning data: what you read, who you talk to, what you practice, what you let yourself avoid. By the time you're an adult, that curation is most of what's actually in your hands, and it's a lot more than nothing.

There's a failure in data systems that haunts me more than any other, and it explains why so few people curate anything. In Vector Engineering I described an account owner who left a company eighteen months ago. The system never found out. His name still sits on the record, so escalations still route to him, reports still count the account in his book, and renewal reminders still go out over his signature. Nobody decided any of that. Somebody wrote a fact down once, and the frozen fact kept governing long after the living one changed. People carry records like that. Slow kid. Special class. Poor kid. Not a math person. Somebody wrote it down when you were nine, and it's still routing your decisions at thirty. The fix in software is to stop trusting the stored label and compute the answer from the trail: who actually touched the account this quarter. The fix for a person is the same: whatever the label said, ask what the record of the last ninety days says.

When I'm learning something new I stand in front of a mirror, look myself in the face, and see how confidently I can explain it simply. If I can't, I don't know it yet. It sounds ridiculous, and I kept doing it because it worked. And I hold myself to one rule about information: the function of information is to change how you act. If you gather information and it doesn't modify your actions, you've wasted your time.

That rule gives me my working definition of dumb, and it's one you can do something about: information that never becomes action, repeated for years, until the reading drops to match.

I've defined alpha the same way for years, long before I had a company to hang it on. (The trilogy that ends in The Alpha Is a Compiler is the long version.)

Imagine a whole team of hundreds of people working in perfect coordination. They're smarter than you, better funded than you, they've been doing this longer than you, and they're willing to wait far longer than you. How do you beat them? You find the thing they don't do only because they don't want to. They know it exists. They've probably tried it. They may have even seen it work. It just doesn't make sense for them, for whatever reason. That thing is alpha. The front door's locked and the back door's locked, so you go find a third door, and finding it usually takes some creativity and a willingness to look stupid for a while.

That definition quietly favors whoever has the least. The reasons a rich, coordinated competitor won't do something are almost always about scale, dignity, or time. The job's too small to move their numbers, too weird to explain to a board, too tedious for people who bill four hundred dollars an hour, too embarrassing for somebody with a reputation, or too slow to show up on this quarter's report. Every one of those is a door a broke person can walk through. You don't have a reputation to protect, a board to explain things to, or an hourly rate that makes the tedious work irrational. Being poor takes a lot away from you, and the one thing it gives you is permission to do the work nobody with options wants.

My whole career in performance marketing ran through those doors. I've had clients come to me with fancy websites they were proud of, fifty thousand dollars and up from an award-winning design agency, and I had to be the one to tell them their baby was ugly, because no matter how pretty the site was, it didn't convert. On one store, the ugliest page I tested won. It had a simple offer and no polish, and it did $600,000 in 60 days. On a metal art brand, one ad had a word spelled wrong in it. We made clean copies and replicated it every way we could think of, and the version with the typo kept printing money, across more than one store, for a long time. Nobody with taste would have shipped that ad. Nobody with taste would've kept it running after they noticed. Taste is a luxury good. On a budget, testing beats it every damn time.

The biggest wins I ever had came from doing the thing that looked dumb on paper. On a lead generation account, we made the application longer, which meant more questions, more friction, and more chances for a lead to give up. Cost per lead dropped from $200 to between $10 and $18, because the long form did two jobs at once: it filtered out people who were never going to buy, and it taught the ad platform's algorithm what a serious buyer looked like so it could go find more of them. Every marketing instinct says shorter forms convert better. The instinct is right on average and wrong for that business, and you only find out which by testing it. Back on that metal art brand, the move that took them from about $10,000 a month to $150,000 a month inside 90 days was a CSV export and a few afternoons in a Google Sheet. The team had been copying a competitor's charismatic founder and writing ads for design-conscious millennials who clicked and never bought. The cohorts said the actual buyer was a woman over forty-five, on an iPhone, buying a gift that would mean something on a wall. The product and the factory stayed the same, and the promise changed. It took no platform and no agency retainer, just somebody willing to stare at a spreadsheet until the imagined customer turned into the real one.

None of those wins took a genius, just a willingness to test the ugly thing, read the result without ego, and do the tedious analysis that better-funded people delegate to someone who doesn't care. That's an edge made almost entirely out of attention. Poverty can make that attention harder to spare, but studying one group closely doesn't require matching a competitor's budget. Point that attention at one specific group of people for long enough and it compounds into what I call hyperrelevance: you know that group so well, and so currently, that what you make fits them because it was made for them. A competitor with ten times your budget can't buy that overnight, because it's built out of time. What most poor kids never get is somebody telling them where to point it.

The other thing poor kids usually don't get is rooms. In 2022, Raj Chetty and a large team went looking for whatever predicted a poor kid's odds of moving up, using about 21 billion Facebook friendships (Nature, 2022). One measure stood out, which they called economic connectedness: how many of a low-income person's friends are high-income. They described it as among the strongest predictors of upward mobility identified to date. In their model, give low-income children the connectedness that high-income children already have and their adult incomes come out about 20% higher. That's a model, not a vending machine where you insert one rich friend and receive a raise, but it's still one of the loudest signals in the data, and it's a signal about rooms.

I didn't know that paper existed when I was a kid. I just ended up in rooms by accident and by necessity. I knocked on doors in Dallas selling knives and soap, which taught me to hear "no" as data. I cold-called thousands of people about insurance, which taught me you get about four seconds to earn the right to speak for another thirty. Somewhere in there I was raising capital for commercial real estate from investors three times my age. That job taught me to build a pitch around what's in it for them, not what's in it for me. Later I spent five years in nightlife, a lot of it in Las Vegas. Every class of person on earth walks through a nightclub door drunk, high, and more candid than they will ever be in daylight. Every one of those jobs put me in a room with people who had more than I did, and every one of them paid me to study how those people think.

Taking those jobs was a feasible-set move. On a résumé, that kind of networking looks like a sales job nobody wanted.

I've spent a lot of time working for and around people with serious money, and the biggest surprise, which took me years to believe, is how rarely the people with the money were the smartest people in the building.

One solar client got results good enough that their investor handed us what amounted to a blank check, as long as we kept leads under $25. We scaled that account to $50,000 a day in ad spend at its peak. It felt like being handed the keys to a Formula One car. What I learned from the inside is that the money was buying runway, and very little intelligence came with it. The financing behind that business let it keep spending and testing long after a normal company would've had to stop, and when the financing ran out, the ceiling vanished with it. The same money made everyone jumpier. An ad account froze for a few hours once, and the investor was blowing up the client's phone like the building was on fire.

On the other side of the ledger, I once audited a company spending over $100,000 a month on engagement ads, the kind that buy likes and do almost nothing for sales. That's bottle service for an algorithm: you buy the whole room a round, everybody cheers while your card is open, and not one of them knows your name in the morning. The account was run by a white-collar marketing manager who had clearly never touched advertising before and had been handed a budget and told to do Facebook with it. I remember sitting there trying to figure out how to tell them they were lighting money on fire. They might as well have walked up to Mark Zuckerberg and handed him the keys to a mansion.

And in crypto I watched the worst version of it. On one project I ended up as the final CEO. The investors and the founder ran off with nearly everything. Then they tried to pin it on me and the CEO before me, and that didn't work. Everybody in that community knew us. We'd been there every day. We'd begged them to just pay the developers and build what we'd promised. They chose to rob everybody instead, including us. Around the same time I watched grant money in that corner of crypto go to the same few well-connected firms, over and over, tens of millions at a time. Everyone else wondered where it went. I helped a team in Africa raise $350,000 for a crypto radio show in that climate. It's harder for a team in Africa to raise money than for a team in Switzerland. We did it by being brutally pragmatic and consistent, and it's still one of the raises I'm proudest of.

One line from those years has paid for itself more than anything else I picked up in them. A wealthy man told me there's always a market for the best. He said it the way people say things they've tested rather than things they've read, and I've had a decade and a half since to check it against every market I've worked in. I haven't found the exception yet.

What I was doing before that, and what most broke people do, is aim cheap and hope volume covers the gap. That's a reasonable-sounding plan and it puts you in the most crowded room in any industry. He was describing the opposite bet: get into the top slice of whatever you actually do, because the buyers with money are shopping in that slice and there are fewer of you in it. You don't want to be the worst. You don't want to be anywhere near the worst.

It's one rich guy's heuristic, and survivorship bias means the ones it failed aren't around to tell me. It still changed what I was willing to charge, which was the only test available to me at the time.

Watching it up close taught me that money buys runway, rooms, and the right to be wrong more times. It rarely buys judgment. Often it buys the absence of consequences for bad judgment, which is a different product, and over time that product makes some people worse at the thing they got rich doing.

Being broke is a shortage of runway, and the reason it feels like it makes you dumber is mechanical. Strategy games have a word for an economy that spends faster than it earns: a stall. In a stall nothing stops completely, and every build queue slows at once. A person stalls the same way when cash timing starts deciding every other choice, from what you eat to whether you can afford to fail at something new. Does that make you dumber? No. It makes you slower at everything at once, which feels identical from the inside. I wanted to cite the famous study claiming money worry costs about thirteen IQ points, and it didn't survive my research rounds. A larger study tested low-income American households before and after payday and found no difference in cognitive performance (Carvalho, Meier & Wang, American Economic Review, 2016). So I'll claim only what I can defend: a stall is a rate problem. It says nothing about how fast you'd run with your resources refilled, and refilling them is usually the first move.

The gap between money and brains shows up in every field I've worked in: technical, business, and creative. The people in the lab, the people on the front line, the people talking to the customer every day almost always know more than the people with the title and the budget. Between the two sits bureaucracy, infighting, a staggering amount of waste, and a standing circle jerk of people who have confused attending meetings with doing work. And when power sits somewhere with little transparency and little oversight, it eventually finds a person who will use it badly. Then what counts as normal gets a little more extreme every year after. That's a claim about all humans: every country, century, race, and political party I've studied. In everything I've read and everywhere I've worked, I've yet to find a group that held unchecked power for long and didn't produce this. So nobody should be scandalized to find it in the room they're trying to get into.

That pattern matters to a broke person for a practical reason: the people above you are beatable. What they have over you is runway, rooms, and forgiveness, not a better brain. All three are things you can build or borrow in pieces. Which one are you actually short on?

Luck is enormous, and the bootstrap crowd loves to skip it. Economists have studied the earnings histories of millions of American workers. Lifetime earnings risk is large and lopsided: a small number of people catch big positive shocks, and a lot of people catch bad ones (Guvenen, Karahan, Ozkan & Song, Econometrica, 2021). What nobody can give you is a percentage. There's no credible study that says your outcome is 60% effort and 40% luck. Anybody quoting you a number like that is quoting their politics.

When I was raising real estate capital in my early twenties, every pitch had to say "past performance is no indication of future results." The warning belonged there, and I've carried that sentence into marketing, where it drives people crazy, because I'll tell a prospect not to trust my case studies either. I can't guarantee anyone results, only what I do: the system, the process, the testing. I can guarantee that even the tests that lose will teach us something and leave us closer than before. The one time I forgot that, I got humbled on a garden art brand: a $5,000 budget, then another $5,000, three rounds of testing, and not one campaign caught. I'd sold millions of dollars of product in that exact niche. It didn't matter.

Now turn the disclaimer around and point it at yourself. Past performance is no indication of future results. Your past can't settle what happens next, either. Your bad years are a record, not a forecast, and on that point the warning has told you the truth more reliably than most of the people in your life.

The same idea has an engineering version, and it's been my whole career. Mathematicians can spend decades, sometimes close to a century, trying to prove whether something is possible, and they'll wander into quantum physics to do it. Meanwhile engineers ship systems that work 99.99% of the time, and the internet, cars, and satellites all run on that. What you need is a method that works reliably enough for you, tested cheaply enough that a failure doesn't kill you, rather than a proof that escape from poverty is possible for everyone in every case. Mathematics learned the same thing about its instruments. It has two great ways to measure shape: the sharper one is provably impossible to compute in general, while the coarser one runs on ordinary linear algebra and became the daily driver. I wrote about that in The Shape of Data II. A partial fingerprint you can take at scale beats a perfect one you can never take.

That's why I set a price floor on purpose when I build offers for small businesses. One of my services starts at $2,000, and I tell people plainly: if a $2,000 decision is a crisis for you, you're not ready for this, and that's fine. Your first move is to get to a place where $2,000 isn't a crisis. I mean that as the most useful thing I can tell someone at that stage. A person who can't afford a small bet can't afford to test. A person who can't test is stuck relying on luck.

Every operation I run, whether it's a software build, a marketing campaign, or a team of AI agents, runs on a framework I built and call the operational hierarchy. Picture a ladder with nine rungs. The two side rails are purpose, the reason the thing exists at all, which you hold onto at every step and never stand on. The rungs run from the most abstract at the top to the most concrete at the bottom. In a business, a plan that's missing a rung is broken, not partial, because whoever executes it fills the missing rung with a guess.

It took me an embarrassingly long time to notice that the same ladder describes a life.

Most lives I've looked at through that table, mine included for most of my twenties, are holding the rails at the Mission rung, with almost nothing beneath their feet. They have a purpose, sometimes a beautiful one. They have a mission, usually some version of "I want freedom" or "I want to take care of my family." And the bottom of the ladder is missing entirely. There's no objective with a number on it, no project with a finish line, no action small enough to start on a bad day, no decision rule made in advance, and no record of what actually happened.

The failure I'm most afraid of in software is when the data layer gets secretly simulated. An upstream step fails quietly and writes placeholder values instead of an error, and every system above it keeps running as if the placeholders were real. The dashboards look healthy. The demos look great. Then real data shows up and the whole thing falls over, because everything above the bottom rung was reasoning about a fiction.

People simulate the data rung of their lives constantly. It gets filled with a story instead of a record: how hard I've been working, how close I am, how it's all about to come together. The story is the placeholder, and it keeps the rungs above it looking healthy right up until reality arrives.

I've written about the fix from two directions. In Vector Engineering, the rule is that a visible blank tells you what you don't know, while a plausible guess erases the question. "I don't actually know what I did most of last week" is a better record than "I've been grinding," because the first one can be fixed. And in The Shape of Data III, I laid out how I grade a piece of stored knowledge. A draft stays low-grade no matter how good it looks or how long it took, and only something happening against it in the real world promotes it. A plan with nothing happening against it is a draft. So what has happened against yours?

I've heard versions of one sentence in more than one house, said flatly, with no irony anywhere near it: "This is my home, and I deserve to be safe from facts and data." That's the most direct statement of the suffering loop I've ever heard. I respect it a little. Almost everybody lives by it, and almost nobody has the stones to say it out loud.

When somebody tells me they want their life to change, I ask them a few questions, and I'm asking you the same ones now.

What did you do yesterday? I mean what happened, as opposed to what you meant to do. What about the last week? The last month? Walk me through your routines over the last year and how they've changed, if they have. If you've followed me for the last couple of years, where would you say your life is headed, judging by what's actually occurred rather than by what you want? And don't just list the obstacles. There have been other things too, good things, things you did right that you've stopped counting. Give me the whole record. And if there's no record, start one today. A record started today answers nothing about the past. It starts a clock, and the history you'll be able to read a year from now runs from the day you started it.

Most people will never have that conversation with anyone, including themselves, and the people willing to have it are the people I'm looking for.

This ladder isn't a law of physics. You can skip a rung. People do it all the time, and some of them get away with it. It's harder and more dangerous, and you can't skip two in a row and still call it climbing. At that point you're jumping. You need consecutive rungs under your feet. Going straight from "I want freedom" to a burst of frantic activity, with no objective to measure it against, is jumping. Going from a clear objective to "I'll just be more disciplined," with no tasks or actions underneath, is a wish with a spreadsheet attached. What is under your feet right now?

I learned the consecutive-rung rule in my body before I learned it in software. After spine surgery and four titanium rods, I had to learn to walk again, and it took the better part of eighteen months. Stairs became a full-body problem. Then came a backpack with two books in it, then more weight, flat ground before hills, and familiar ground before new terrain. Each step held only because the one before it had been tested under load first.

The rung most people skip is the decision rung, and it's the one with the best evidence behind it. At work, a decision has two parts: the rule for making it, and who's allowed to make it. Most people make every decision about their behavior live, in the moment, by mood, which means their worst mood gets a vote every day. Psychologists have a name for pre-deciding: an implementation intention, a plan shaped like if this happens, then I will do that. A meta-analysis of the research found a medium-to-large effect on actually reaching goals, around d = 0.65 (Gollwitzer & Sheeran, Advances in Experimental Social Psychology, 2006). It works because the decision was already made by a calmer version of you, and the tired version only has to follow it. I wrote the general rule in a field note about why the data mesh failed: sustained discipline shows up when the cheap path and the disciplined path are the same path, not when somebody writes a values statement.

I've struggled with hyperprocrastination for most of my life. I could blame it on childhood trauma from growing up in a dysfunctional household in a racist community, and that was a real obstacle. It isn't an excuse I invented, and it never did the work for me.

I'm not going to pretend I've beaten it, either. I once wrote in my notes, after a relapse, that I'd "once again stabbed myself in the foot with the long sharp spear of my relentless spirit of self-sabotage," and then the next line was "but I gotta cook." That's about as close to a method as I have: own it flat, no speech, then go cook.

The mechanics I've found that actually help are boring, which is why they work. When I freeze on a task, I name what's happening as a nervous-system event rather than a character flaw. I ground my body before I try to use my head. I strip the task of stakes. A task that has to be perfect is a task that never starts. I take about two minutes for a physical transition: a walk, a stretch, water on the face. Then I do a minimum viable session: the smallest version of the work that still counts. Most days the minimum turns into more. On the days it doesn't, the log still says I showed up, and the log is the only rung I trust. A miss I wrote down narrows the next attempt, and a miss I didn't write down is just waste.

If you procrastinate, you're in enormous company. Piers Steel's meta-analysis estimated that 80 to 95% of college students procrastinate at least sometimes, around half do it chronically, and something like 15 to 20% of adults are chronic procrastinators (Psychological Bulletin, 2007). Nobody's getting a medal for this. And the reading moves here too. In the long-running Dunedin study in New Zealand, children's self-control predicted their adult health, wealth, and criminal records even after accounting for IQ and family background. And the kids whose self-control improved as they grew up did better than their early scores predicted (Moffitt et al., PNAS, 2011). It's an observational study, so it doesn't prove that forcing yourself to improve produces the whole effect. It does say self-control is another reading, not a verdict.

Say you have a new problem in front of you, which is worse than a hard one. You don't know where to start, so you start where you started last time. It doesn't work, and you can't tell whether it failed because the idea was wrong, the execution was wrong, the timing was wrong, or the market was never there. So you try something else. That doesn't work either, and it teaches you nothing, because you changed four things at once and never wrote down what you expected. Which of the four was it? You'll never know.

Drill a few more layers into that failure and it stops being about the problem at all. Where does it actually bottom out? You start in the wrong place because you're matching on the last thing that worked. You match on the last thing that worked because you never built a way to tell which variable mattered. You never built one because looking closely might show that two years of effort went somewhere stupid. At the bottom, every outcome arrives uninterpretable, so no amount of experience compounds. Ten years in, you have one year of experience ten times.

The missing piece is a procedure for opening an unfamiliar situation, not intelligence or effort, and without one the results never stack.

I know that trap well. I lived in it for most of a decade, working hard and learning almost nothing.

People who watch me work now tell me I read situations fast. I'm slow, and what they're seeing is reading I did in advance, not instinct.

I could never walk into a room and just absorb it, so I have to build the read out of parts. Over about fifteen years I collected instruments, mostly borrowed, and learned which one to pick up for which kind of problem. That collection is most of what I actually sell. The rest is the willingness to run it on a Tuesday when nobody's watching.

Used well, those instruments give you four things you can have by Friday: a named guess about what the person in front of you is avoiding, a written record of which variable you changed, a rule you decided before you were tired, and a way to tell next month whether you were right.

Several of the instruments have no scientific standing, and one is flatly pseudoscience. I use them anyway, on the same argument I already made about the engineers and the mathematicians.

Every problem somebody pays me to solve has a person at the middle of it who's avoiding something.

Cognitive behavioral therapy gave me the vocabulary for that before any marketing book did. CBT's working claim, in Aaron Beck's late formulation of it, is that beliefs, situations, and reactions form a loop, and the loop can be edited at the belief (Beck & Haigh, Annual Review of Clinical Psychology, 2014). Clinicians work from a catalog of predictable distortions: catastrophizing, black-and-white thinking, mind reading, fortune telling, emotional reasoning, should statements, labeling, and disqualifying the positive.

Read that list again as a marketer. It stops being a clinical artifact and turns into the list of reasons your prospect says no.

I already used one piece of that clinical vocabulary on you: guilt says I did a bad thing, and shame says I am a bad thing. The distinction comes straight out of the clinical literature rather than from me, and it's the most useful sentence I know for working out why a smart person won't move.

On a sales call, a prospect tells me Facebook ads don't work for their industry. That's overgeneralization, and underneath it's one bad run with one agency in 2021. Another tells me their product is too complicated to explain. That's usually mind reading: a prediction about what a stranger can follow, made by someone who's never watched a stranger try. Another has a real problem and a real budget and keeps rescheduling. That one's almost always fortune telling. They've already run the movie where they spend the money and it fails, and they're declining to buy the ending they invented.

Those people don't need information, and that took me years to learn. They're running a pattern, the pattern is old, and it usually predates the business by about two decades.

The whole procedure fits on an index card.

Ask what specifically happened the last time they tried this, as opposed to what they think about it. Then write down which distortion you think is running, by name, before they finish talking. That written guess is the entire trick, because a guess you wrote down can be wrong in a way a feeling in your chest can't. Decide in advance what you do with each answer: if it's a bad prior experience you go get the numbers from that experience, and if it's a prediction about the future you go find one counter-example they already own. Then check the guess against what they do next, not against how the call felt.

Name it, write it, pre-decide the response, check it later: those four moves are how I make my reads checkable.

I keep a second catalog for my failures, and I stole it from quantitative finance. It includes look-ahead bias, where you accidentally let your backtest see the future; survivorship, where you only study the funds that lived; and overfitting, where you tune a strategy until it perfectly explains a past that will never recur. There are seven in the version I use, and at some point I noticed they line up almost one for one with the seven deadly sins the medieval theologians wrote down. Overfitting is gluttony. Survivorship is envy. Look-ahead is pride, in the precise sense of assuming you knew back then what you know now.

The finance version of that list isn't folklore. Bailey, Borwein and Lopez de Prado worked out how to compute the probability that a given backtest is overfit, and showed that if you try enough variations you can always produce one that looks excellent and means nothing (The Probability of Backtest Overfitting, 2014). Run enough ad creatives and the same thing happens. The winner you found may be the winner the search guaranteed you would find.

Matching backtest errors to sins sounds like a party trick right up until you use it on a campaign that's quietly dying.

The second instrument is dialectical behavior therapy, and specifically the dialectic part, which is the only piece of therapy language I've found that survives contact with a business problem unchanged.

A dialectic is two things that appear to contradict each other, held as both true, without collapsing to either one. Marsha Linehan built a whole treatment around it for people whose emotional regulation had come apart, and the central move is refusing the collapse: you keep both, at once, on purpose, instead of averaging them or picking the comfortable one.

You've already seen my dialectic. It's been sitting at the top of this memo since before I made a single argument.

I'm not to blame for my programming, AND I'm the architect of my own self-sabotage.

That AND is capitalized on my wall, because the sentence stops working the second it becomes an or. Collapse it to either side and you get one of the two crowds: the one that's right about the table and never moves, or the one that hates itself productively and calls that discipline. The only reason this memo has two halves, and a table of research in the first one, is that I refuse to collapse it. Everything else here is downstream of that refusal.

The shape transfers directly to client work. Suppose a client's market really is harder than it was three years ago, AND their offer is the reason they're losing. Both are true. Say only the first and they feel understood and change nothing, and the second on its own gets you fired before the second call. The work is holding both in the same sentence and staying in the room while it lands.

I use the same structure on myself when a launch fails. The market conditions were real, AND I shipped late. I practiced that sentence shape until I could produce it under pressure, and that practice is the flat description of most of what gets called emotional maturity.

There's one more piece of DBT I've taken wholesale, which is distress tolerance: the idea that you can be taught to survive a feeling without acting on it. Anyone who's held a position through a drawdown, or watched a client's ad account get restricted on a Friday afternoon, is doing distress tolerance whether they have a name for it or not. The name helps for one boring reason: a named thing can be practiced.

I ran the whole kit past the research engine the same way I ran the original premise and told it to be unkind, and it was. Here's the scoreboard.

So why is a guy who spent a whole section of this memo checking citations still carrying a pseudoscientific instrument in his bag?

Because an instrument's output is a reading too. I told you earlier not to write a reading into the schema as if it were the thing's permanent nature. That rule applies to my tools as hard as it applies to a kid's test score. Spiral Dynamics is a weak instrument. A weak instrument can still be pointed at something and produce a number worth checking. The failure is treating the number as the territory.

One rule makes the whole kit safe to carry.

An instrument gets to produce a hypothesis. It never gets to produce a claim.

Spiral Dynamics can tell me which of four ad angles to test first. It never gets to tell me which one won. The test tells me that, in money, and if the framework picked wrong I find out in about nine days for a few hundred dollars. Hawkins is the clearest case. Don't think too hard about where his numbers come from. There's some pseudoscience wizardry in that basement, and I won't defend a single digit of it.

What I needed was an ordered inventory of emotional states. Anger, grief, sadness, and excitement aren't the same thing, and even the ones that sit close together differ. Building an index like that from scratch is a career. So I picked up the work of a guy who's reasonably bright, spent decades arguing for his map, and put it into multiple books, and I took his ordering. I'm an enthusiast for making information visible too, so I can appreciate a good design without signing off on the methodology that produced it.

The piece I actually use is the shape: emotions sort into the ones that shrink a person and the ones that open them up, and the hinge between the two sets is courage. You met that idea earlier in this memo at the crossing point of the suffering loop, and I put it there without his name attached, because the shape survives and the calibration doesn't.

Carrying Hawkins is the engineer's position again: instead of a proof that the map is true, I need to know the failure mode when it's wrong, the cost of being wrong, and roughly how fast I find out. Mathematicians earn their keep by proving things impossible, and I earn mine by shipping something that works often enough, at a price where the misses don't kill me.

One version of reading people is seductive, and I have to talk you out of it because I've done it.

Once you start reading people for a living, you start reading their taste. Somebody tells you what they play, what they watch, what they have on the shelf, and it feels like data. Grinding games, cozy games, bleak fiction, and board games with a thousand tiny rules all seem to mean something. Everybody who reads people for a living has a private table of what each preference means, and mine used to be quite confident.

Then I tried to write my table down here and caught myself committing the exact crime I spent a section of this memo prosecuting.

I told you that a gap between two group averages tells you close to nothing about the individual in front of you. If I now tell you that people who like idle games are autistic, I've run the same bad inference with a smaller sample and worse data, and I've done it to a group I'm a member of. Average Alex is back, and this time he has a Steam library.

What the research supports is thinner and more useful than the folk version. Studies of autistic adults find real patterns in why people play, including stress relief, immersion, achievement, and creativity, with role-playing and action-adventure among the most common favorites rather than any distinctive grinding profile (Mazurek, Engelhardt & Clark, Computers in Human Behavior, 2015). Later work found associations between autistic traits and some genre patterns, varying by trait measure and by gender, all of it correlational (Frontiers in Psychology, 2022). That's a real signal, and it's nowhere near a diagnosis.

The genre is noise, and the data is the why, which you get by asking.

When somebody tells me they play a game where the numbers go up slowly and nothing can kill them, I don't conclude anything; I ask what a good session feels like. Sometimes the answer is that it's the only place all week where effort reliably turns into progress. That answer tells me nothing about the genre and everything about their job, and it's the most useful thing I'll learn on that call. A preference is a question generator, not a verdict.

Here's mine, since it's cheap to read other people's taste and expensive to hand over your own. I like the bleak stuff: Berserk, Attack on Titan, 40K. What I get out of it is that a world that shows how bad it can get is restful to me in a way a gentler one isn't. That's a report about me, and it would be a bad basis for a claim about you, and I notice the pull to make one anyway.

If you run every instrument on every problem you get a checklist, and a checklist is what people produce when they want to look thorough. It's the analytical equivalent of running every optimization pass in a compiler on a program that prints hello world. It's slow and impressive-looking, and it's worse than doing nothing, because now you've got forty pages and no damn decision.

So how do I pick? A technical architecture question wants the failure catalog and the decision shapes, and it has nothing to do with what a prospect is afraid of. A stalled sales conversation is almost entirely fear and almost never architecture. A content problem sits in the middle and mostly wants to know what the reader already believes. Getting that routing right is worth more than any single framework in the bag, and it's the part nobody teaches, because it looks like having no method at all.

Bloom's taxonomy is how I decide whether I'm allowed to skip something. The 2001 version runs remember, understand, apply, analyze, evaluate, create (Krathwohl, Theory Into Practice, 2002), and I use it as a blunt self-check rather than a ladder. Can I remember that Spiral Dynamics exists? Sure. Can I apply it to a live account without looking anything up? Also yes. Could I create something new inside it, or catch a subtle error in somebody else's use of it? Not close. So Spiral Dynamics gets to suggest, not decide, and my confidence in the output is capped at my level on the thing that produced it.

The version of Bloom I carry in my head has an extra rung at the bottom that the official one doesn't, which is awareness: knowing a thing exists at all, before you know anything about it. That rung is mine, not theirs. Most of what separated me from people with more schooling was inventory, not depth. You can't reach for a tool you've never heard of, and a surprising amount of what looks like intelligence is just a longer list of things you know exist.

Everything so far tells you how to read a situation and nothing about what to do, and analysis that never becomes a decision is the most expensive habit in this business. I know because I spent years there, well-informed and broke.

My fix came from Warren Powell, a Princeton professor who spent a career noticing that every field with a name for decision-making under uncertainty had invented its own vocabulary for the same four things. He wrote the framework up as Reinforcement Learning and Stochastic Optimization, and his lab keeps a free companion site for people who don't want to buy a Wiley textbook. Operations research, control theory, reinforcement learning, and economics use different notation for the same four shapes. His framework says that any way of turning a situation into a decision falls into one of four classes, and once you can see the four, you stop reaching for the expensive one out of habit.

Here they are in plain terms, cheapest first.

The first class is a rule you decided in advance. If this happens, I do that. You do no calculating at the moment of truth, because the calculating already happened, back when you were calm.

The second is a rule with dials. It's the same idea, but now you optimize the immediate move: how aggressive, how big, how wide, given what today looks like. You're solving for right now and deliberately ignoring next month.

The third weighs the future against the present: a reward now that costs you later, against a cost now that compounds in your favor. This class is where patience lives, and where infrastructure decisions get made.

The fourth simulates the decision forward. You play the tape: if I do A, they do B, and then the ground looks like this. It's expensive to run and worth it only when the decision is big and hard to reverse.

So why am I putting a decision taxonomy in a memo about being poor and being dumb?

You already met the first class in this memo, and I gave you the evidence for it without the name. The implementation intention, the if-then plan that showed a medium-to-large effect on actually reaching goals, is a rule you decided in advance. The most-researched self-discipline intervention I know of and the cheapest decision class in a Princeton textbook are the same object, because they solve the same problem: your worst mood shouldn't get a vote at the moment of truth.

The failure I see constantly, in businesses and in people, is reaching for the fourth class when the first one was sitting right there. Somebody with no rule about what they do on a bad morning will happily spend six hours simulating five-year scenarios. Call that strategy if you like. It's the most sophisticated available method of never starting.

The discipline is to start with the cheapest class that could work and escalate only when it fails, specifically, in a way you can point at, not when it feels too simple or when a simple rule embarrasses you in front of people with MBAs.

Almost everything I do for a client in the first ninety days is class one: a rule about which leads get called back within the hour, a rule about what happens to an ad when it crosses a cost threshold, a rule about which week the offer gets tested. Rules like that are boring, cheap to monitor, and easy to debug when they break. Most businesses I audit are running no rules at all and calling it flexibility, which is a hell of a word for winging it.

A rule with no record attached is just an intention wearing a uniform, so what else ships with it? Two things ship every time. The rule names who owns it, because a rule everybody owns is a rule nobody follows. And it names what gets written down each time it fires, because in ninety days the only question that matters is how often the rule ran and what happened after. The log of every firing is what a client actually buys from me. The strategy deck is free.

And the framework travels, which is the part I find beautiful. A soldier pinned down with a wrecked leg, no signal, and a drone overhead is running the same four shapes as a guy in an office tuning a model for a slightly better return on some affiliate offer. Their variables aren't remotely comparable, and their stakes are obscene to put in the same sentence, but the shapes are identical. One of them needs a pre-made rule more than anybody has ever needed anything, and the other one has the luxury of simulating.

Poverty pushes you toward class one for a reason worth naming. The cheaper classes need less compute, less data, and less time, and they fail in ways you can see coming. When your margin for error is thin, a simple rule you actually follow beats an elegant model you can't afford to be wrong about. That's the correct choice given the constraint rather than a compromise, and it stays correct until the constraint moves.

The seven sins of backtesting line up with the seven the theologians wrote down. The loop that keeps a person stuck looks like the loop that keeps a company stuck. A pre-made if-then rule is the simplest class of policy in a Princeton textbook and also the most-studied trick in behavior change. A stalled economy in a strategy game and a person who can't afford to test anything have the same rate problem.

Either I'm pattern-matching on noise, which is the most likely explanation and the one you should hold onto, or some of these shapes really are the same shape.

For about a decade I assumed noise. I had no mathematical training, I was reading about fluid dynamics and reinforcement learning for fun, and I kept getting the feeling that a trading strategy behaved like a fluid. You can't publish a feeling, and I couldn't make it go away either.

Then Ben Goertzel published an argument that made me sit down. His argument, in his words and under his term FluQNet, is that dynamic programming on a particular group of transformations gives you the incompressible Navier-Stokes equations, and that the same Hamilton-Jacobi-Bellman equation with its variables complexified becomes, in his framing, the Schrodinger equation. In that framing, optimal decisions, fluid flow, and quantum dynamics are three views of one structure. He lays it out in Rethinking (Classical + Quantum) Brain Dynamics and in recorded talks, and it's a stated research program rather than an offhand remark.

I can't check his proof. It runs through category theory and I don't have that math. I can follow the shape of the argument and I can read the people who do have the math, and what I'm doing when I repeat it is reporting, not verifying. The population of people who have read that work, understood it, and thought about what it means for building things is small enough that I'd be embarrassed to guess the number out loud. A tiny audience isn't evidence that something is deep. Plenty of things nobody reads are nothing. I mention the size of the room because it should lower your confidence in me, not raise it.

What I'll defend is much smaller and entirely practical. If several domains genuinely share a structure, then learning one of them deeply buys you the others at a discount. The practical payoff doesn't require the proof to be right, only the resemblance to be useful, and usefulness is a bar I can test.

That bet has paid, repeatedly, in ways I can point at. Understanding how a fluid goes from smooth to chaotic at a certain speed is why I can look at a market that's been calm for three months and stop trusting my averages. Understanding value functions is why I can explain to a client that the cheap fix now is a loan against next year. Understanding that a loop needs a crossing point is why the suffering loop earlier in this memo has an exit rather than just a diagram. I got those from studying something that looked unrelated and noticing the shape again, not from business books.

Depth in one strange field, plus the habit of looking for its shapes somewhere else, is the most available advantage left to anybody with no credentials and an internet connection. It will beat broad familiarity with your own industry more often than anybody in your industry wants to admit. Your competitor read the same eleven marketing books you did. Nobody in your market read fluid dynamics.

Pick one field that isn't yours and has real mathematics under it. Give it an hour a week for six months, which costs you nothing but the hours. Keep a running list of every place its structure shows up in your actual work, dated. At six months you'll either have a list with entries on it or you'll have an empty page, and both of those are answers. The empty page means you picked a field too far from your work, so pick a closer one and run it again. My list started with three entries in a year and it's the most valuable document I own.

I keep two kinds of math going for that reason. Topology I learn for fun, because Goertzel's enthusiasm for it got me curious and because I like knowing how things are put together. Probability and statistics I learn for work, because that's what actually decides whether a campaign was real or lucky. I don't have a degree in either one. I have a decade of reading the wrong books on purpose.

If the shapes repeat, then the thing worth building stops being another report and becomes a model of the world you operate in, stored somewhere a machine can query, that keeps getting more correct as things happen against it. That's what I mean by vector engineering: the discipline of getting facts, relationships, and their timestamps into a shape where a question can actually be answered instead of guessed.

Most businesses don't have a model like that. They have a CRM full of stale fields, a folder of decks, and three people who remember why things are the way they are. Ask that company a question that crosses two departments and watch what happens. Somebody schedules a meeting.

Three pieces I've written separately each describe one face of the same object. Echolocation is how you get a read when you can't see: you send something out and listen to what comes back, which is what a test is. Hyperrelevance Cartography is what you do with the read once you have it, which is to map a specific group of people precisely enough that what you make for them fits. Vector engineering is where the map lives so it doesn't evaporate when somebody quits. I wrote them as three essays because I understood them one at a time, and they're one system.

Put the three together, point a few AI agents at the result, and you get something I think of as a suit, which gives you reach rather than intelligence. The first version I built was the equivalent of banging rocks together in a cave. It's still early and held together by discipline more than cleverness. The direction isn't in doubt, and the cost of the parts is falling for everybody at the same rate.

For most of my career, when I hit a problem I couldn't think through alone, I'd hire a senior freelancer by the hour just to talk to. We'd walk through the problem together, they'd go off and work on a piece of it, come back, talk some more. That arrangement borrows somebody's time and expertise, part-time, only when it's valuable, which is a lot cheaper than hiring. It was still a couple hundred dollars for a session and thousands for a real engagement, and for most of my life that was money I didn't have. I always figured AI would eventually get good enough to fill that role, and it has.

If you're studying a book, you can now hand the book to a model and have it teach you the book. If there's a concept you can't crack, the model will break it down in your culture, your language, with your specific setup and your specific problems. Pretty soon it will build you a custom environment to practice in. I've spent a lot of time in Southeast Asia explaining highly technical ideas to smart people whose English was their second, third, or fourth language. The whole skill was finding their words for my concept. A model can do that translation for anybody who asks. That's a feasible-set move that used to cost money and now costs almost nothing.

The evidence on this is new, and it's better than I expected. In a randomized trial in a Harvard physics course, students learned from an AI tutor that had been designed around how people actually learn. They showed more than twice the median learning gains of students in an active-learning class, and they got there in less time (Kestin, Miller, Klales, Milbourne & Ponti, Scientific Reports, 2025). In Nigeria, a six-week after-school program built on GPT-4 put teachers in the room and ran a set curriculum. It raised students' scores on a combined test of English, AI knowledge, and digital skills by about 0.3 standard deviations (De Simone et al., World Bank, 2025). The authors compare that gain to one and a half to two years of ordinary schooling, which is their framing rather than something they measured directly, and it's still a lot for six weeks.

Keep your expectations calibrated, though. The legend in education research is Benjamin Bloom's 1984 finding that one-on-one tutoring puts a student two standard deviations ahead of a regular classroom, the famous "two sigma problem." When later researchers pooled the actual tutoring studies, the average effect came in around 0.37 standard deviations (Nickow, Oreopoulos & Quan, NBER, 2020). That's real and worth having. Nobody should expect a chatbot to beat the legend when human tutors don't.

A team led by Hamsa Bastani ran an experiment with high school math students in Turkey (published in PNAS, 2025). One group got plain GPT-4 to help with practice problems. Another got a version set up like a tutor, one that gave hints and refused to just hand over answers. A third group got nothing.

On practice, the plain GPT-4 group did 48% better. Then the AI was taken away for the exam, and that group did 17% worse than the kids who never had it at all. The tutor group did 127% better on practice, and the exam penalty largely disappeared. The model underneath was the same, and the only difference was whether it did the thinking for them or made them do it.

A small study out of the MIT Media Lab measured brain activity while people wrote essays with and without ChatGPT, and the ChatGPT group engaged less and remembered less of what they had written. It had 54 participants and hadn't been peer-reviewed when I checked, so I hold it loosely. It points the same direction as the Turkey experiment, which is the direction any strength coach would've predicted: if the machine lifts the weight, you don't get stronger. Letting a model do your reps is intellectual edging: you get all the sensation of progress, nothing gets delivered, and you have to go back tomorrow for the same feeling.

And the tools make things up. The citation with the wrong authors that I mentioned at the start was a real paper, correctly described, with a completely different set of authors attached. If I hadn't checked the journal myself, you'd be reading a confident, specific, wrong citation right now.

The rules I actually use cost nothing:

I said something once that I still believe: this is the dumbest we'll ever be. The tools only get better from here. Everyone rents them at the same price. I've argued in print that the models are the beta, the return everybody gets just for showing up: the kid at the private school and the kid with a cracked phone rent identical intelligence by the token. The alpha is whatever you do with it that other people won't, and in learning, that means the reps. The tools make the box bigger for everybody who uses them to train, and they quietly shrink it for everybody who uses them to skip the training. Access used to separate people a lot more than it does now. So what separates them now? The reps do.

The hardest round I lost with the research engine came when I told it to argue against me. Its job was to make the strongest possible case that this whole memo hurts the people it's written for. I expected a lecture and got a study I can't stop thinking about.

In 2019, Erin Godfrey, Carlos Santos, and Esther Burson followed 257 mostly low-income middle schoolers at an urban school from sixth grade through eighth (Child Development, 2019). They measured how strongly each kid believed the American system is fair and rewards hard work. In sixth grade, the kids who believed it most had higher self-esteem and better behavior. Over the next two years, those same beliefs predicted falling self-esteem and more delinquent behavior, especially among kids who also reported experiencing discrimination.

The researchers' explanation is the part that got me. A kid who believes the game is fair and then keeps losing it has only one conclusion left: I must be the problem. The belief that was supposed to protect them turned into the thing that convicted them. It's one school and it isn't an experiment, so I hold it with caution. It still describes, precisely, a kid I knew very well.

If "it's your fault" means "the system is fair, so your situation proves you're lazy or stupid," then in the best study I found on it, that belief predicted real damage in the exact kids it's aimed at. The bootstrap crowd has been handing out that belief for decades as if it were medicine. Medicine for whom?

The softer version doesn't do much better. Growth mindset, the idea that telling kids their abilities can grow will make them grow, turned into a whole industry. The effect sizes are small. One large review put the average effect of mindset interventions around d = 0.08 (Sisk et al., Psychological Science, 2018). A national randomized study found about d = 0.11 for lower-achieving students, bigger in schools where the peer culture actually supported taking on hard work (Yeager et al., Nature, 2019). And a more skeptical analysis that corrected for publication bias and study quality found the average effect shrank to about d = 0.02, a tiny estimate with no statistically detectable effect in that analysis (Macnamara & Burgoyne, Psychological Bulletin, 2023). Telling people to believe harder is weak medicine. The one place it did something was where the environment gave the belief somewhere to go.

So what does help? The evidence points somewhere I recognized immediately, because it's how I survived. There's a research literature on something called critical consciousness, built on the work of the Brazilian educator Paulo Freire and studied in American youth by researchers like Matthew Diemer. It measures three things together: seeing structural unfairness clearly, believing you can act on it anyway, and actually acting. In several studies, young people who hold that combination tend to show stronger career development and higher expectations for their working lives (Diemer & Blustein, Journal of Vocational Behavior, 2006). The limits are real: it's mostly correlational, mostly self-reported, and nobody has proven it's a recipe. One study even found that seeing injustice clearly went with more anxiety when it wasn't paired with a way to act. That fits too, because seeing the rigging without a move is just a more detailed way to feel hopeless.

Those two failure modes are cousins. The kid who believes the table is fair is avoiding the truth about the table, and the doomer who believes nothing matters because the table is rigged is avoiding the truth about his moves. Which one are you running? I've written before about the doomer's trade: you surrender your accountability and you receive relief in exchange. It's the best drug on the market, and it's free until the bill arrives about a decade later in one lump sum. Both beliefs let you stop looking at the actual situation.

When I was a kid, being autistic and Black in a small Texas town didn't leave me the option of either belief. I learned early that if I was going to stand out, I'd better stand out on purpose. I watched. I analyzed what was working for the kids who weren't getting threatened. And eventually I stopped projecting my judgments about what the world should do and started being ruthlessly grounded in what the world was actually going to do. Refusing to engage reality wasn't a stable option for me, because reality was going to engage me either way.

I lived in Ukraine for a couple of years, starting when my business collapsed in 2018 and friends there offered me a place to land. The Kyiv and Odesa I found were a culture rebuilding itself after destruction with almost nothing to work with, and their directness let my brain breathe for the first time in my life. Then in February 2022 I watched that country get invaded and keep building anyway: businesses running, weddings happening, people I love working through sirens. Nobody there thinks the table is fair. Almost nobody there has concluded that their moves don't matter. Watching that is how I learned that conditions don't dictate posture.

Back at the table, the dice are loaded, the dungeon master plays favorites, and some players showed up with gear you'll never be issued. All of that is true, and I won't ask you to pretend otherwise. The best players I've ever met, at any table, in any country, know exactly which dice are loaded and exactly how. They're irreverent about comfort and reverent about pain. They don't laugh at anybody's wound, including their own, and they laugh at every excuse, especially their own. And they keep playing, because nobody is coming to play for them.

Something I said near the top of this memo turned out to be the whole thesis, and I didn't know it when I said it.

I told you I rolled a household where calm was the scariest weather on the forecast, because calm is what comes right before it isn't.

A calm stretch never guaranteed safety in that house. Years later, when I studied volatility, I recognized the resemblance: quiet periods can persist, and they can end. I hadn't measured a household time series or discovered a financial law as a kid. I had learned to watch for a change in the conditions. Volatility clustering describes how the size of recent moves can help estimate the size of coming ones. Hyman Minsky made a different argument: prolonged prosperity can encourage financing choices that make the system less stable (The Financial Instability Hypothesis, Levy Economics Institute, 1992). That's why grown professionals with risk systems walk into the same instability every cycle, with money. I learned the shape of it at a kitchen table before I could drive.

So the resemblance was familiar about twenty years before I had the vocabulary.

Then life kept handing me more volatility: a community where the read on a room could change in a second depending on who walked in, a business that collapsed in 2018 and took the plan with it, raising money from people half a century older than me where the whole job is presenting stability you don't personally feel, a crypto project where the people holding the money left with it, and life in Kyiv and Odesa, then February 2022, watching people keep running businesses and holding weddings through sirens.

None of that was preparation. There's a genre of essay where the author's suffering turns out to have been an excellent investment. That genre is a lie, and it sells. The bad years weren't worth it, and nobody would buy them at the price. Several of them cost me things I haven't gotten back.

What's true is smaller and stranger. When I finally sat down in 2021 and started studying the mathematics of markets on purpose, parts of it were already familiar from what I'd lived inside since childhood. The recognition came quickly; checking where the resemblance held still took work.

So is that redemption? No. It's just the reason the material came easy to me and hard to people with better childhoods and better degrees.

Most people hear volatility and think risk, but the two are different, and the difference is the whole business.

Volatility is the size of the moves, and risk is the chance of a bad outcome you care about. A market that swings wildly in both directions is volatile, and whether it's risky depends entirely on your position and your leverage. You can lose money in a dead-quiet market and make money in a screaming one.

It took me years to believe that you can get paid for the size of the moves without knowing which way they go. Direction is hard and everybody wants it, while magnitude is a separate variable that behaves with far more regularity: it clusters, it reverts, it has a floor and a shape.

That's the finding an entire field was built on, not my observation. Robert Engle's 1982 paper introduced a way to model changing conditional variance from recent errors, demonstrated on UK inflation. That machinery became foundational to forecasting financial volatility (Econometrica). He took the 2003 Nobel for it. Calm begets calm, violence begets violence, and the tomorrow you can't call in direction you can often call in size. A strategy harvesting movement instead of direction is playing a different and considerably more forgiving game than one that has to name the destination.

I first touched Python around eighteen or nineteen and did almost nothing with it. In 2021 I started actually writing it myself instead of hiring people to write it, and by 2022 and 2023 I was deep in algorithmic trading. I went to a hackathon in 2023 and went full-time later that year. I've been watching one specific edge for about five years now, and it has held up and gotten more legible over that time rather than less.

One property keeps me in it, and it's a thesis, not a finding. In most markets, the thing you're harvesting gets consumed by the people harvesting it. Everybody crowds the trade, the edge compresses, and it dies. Volatility behaves oddly on that point. The act of large numbers of participants crowding in, levering up, and getting forced out is itself a generator of movement. Consumption partly regenerates the supply. We'll bump shoulders over the easy fruit, and somebody else doing well at scale doesn't subtract from me the way it would in a directional trade.

That's a thesis with five years of watching behind it and no proof in front of it. Reread the disclaimer I quoted earlier and point it at me. Past performance is no indication of future results, and I'm not excluded from that sentence because I wrote it down.

You can run a version of this on your business for a few minutes a week over eight weeks. Pick the number in your business that swings most: leads per week, revenue per week, hours you actually worked. Write it down every week without editing it. After eight weeks you'll have a first observed range: your lowest week, your highest, and what happened between them. Keep extending the record. Eight weekly observations won't establish a stable risk estimate, but they'll give you something to compare the next week against. Most people running a business can't tell you their worst week last quarter. They can tell you their best one, which is the number they quote and the number that has never once been true twice.

Reading how a stable regime turns unstable is the same problem as reading when an audience that has ignored you for a year is suddenly ready. Sizing a position so a bad run can't end you is the same problem as setting a floor on what you charge. The instruments in the bag were assembled for marketing and they work on markets, and I didn't plan that. I noticed it, the way I noticed the sins lining up, and by then I had stopped being surprised.

This memo isn't for people who want to hear that none of it is their fault and stop there. You're right about the dice. I agreed with you for a whole section with a table. If that's where you want to stay, stay, and nobody here will bother you.

It isn't for people who want to hear that all of it is their fault so they can hate themselves productively. That's the cycle of tolerance in a gym shirt. It feels like discipline and it runs on shame, and I spent enough years in it to know it doesn't pay.

If you came here for a guaranteed income or a guaranteed escape, I don't have one to sell you. Past performance, mine or yours, is no indication of future results.

It's for people who are willing to have the conversation: the one about what actually happened yesterday, last week, and last year, with the whole record on the table, the wins included. It's for people who can hold both halves of the line on my whiteboard at once, and for people who've bled somewhere, which is nearly everyone, and who've decided the bleeding is information about the terrain rather than a verdict about them.

That's the audience I'm building. This memo brings together the methods I've been publishing as I build out andydataguy and the companies I'm assembling under the Looikos name. Every one of them is aimed at the same job: making the box bigger for people who didn't get dealt a big one. That means teaching tools that meet people in their language, the way I learned to explain engineering to smart people for whom English was a third language; data and research tools that let a small operator see what used to take a whole strategy department; and free material that's enough for someone with nothing, because I've been that person. I know what the paid version of this advice costs. I also know how much of it gets sold to people who were never missing information in the first place.

I spent a section of this memo asking you what you actually did yesterday, and what the last year looks like judged by events rather than intentions. It would be cheap to ask that and not answer it. So here's mine, in the same units, with the numbers I'd want from somebody else.

Right now I'm poor, starting close to zero, and I need money coming in, so the plan is built around that rather than around what would look best in a bio.

The first rung is Upwork applications, run through the platform I built for it. The target is past $20,000 a month, which is close to what a well-paid senior data engineer makes at a large company. There's a joke in that and I'll tell it on myself: I'm trading one job as a senior data engineer for five jobs where each client pays me considerably less than that. The reason it isn't insane is the only reason it's ever not insane: my systems do the parts that used to cost hours, so the work per dollar is nothing like what it looks like from the outside. If I'm wrong about that, the whole structure is an elaborate way to get paid like shit, and I'll find out in months rather than years.

The second rung is my agencies, two or three of them, each with its own brand and its own niche. The budget plan is deliberately small: about $1,000 to $2,000 per round of cold-outreach testing, and I'm budgeting for more rounds than the three it takes to learn anything, so call it $5,000 to $10,000 each and $15,000 to $25,000 across all of them. The goal is for each to reach $100,000 a month inside a year. Write that down somewhere and hold me to it, because a number with a date on it is the only kind that means anything.

The third rung is the trading. The cash flow from the first two funds a treasury I manage myself, running the volatility work I described above. That one is the long game and the one I care about most, and it's also the one where I'm most likely to be wrong in an expensive way, which is why it's third and not first.

What I'm selling in the meantime is easy enough to say in a sentence: agency-quality output at freelance prices, with a clear scope and a fast start. That claim sounds like every other claim on the platform, and here's what makes it checkable rather than a slogan: I've personally spent tens of millions of dollars on advertising and analyzed hundreds of millions more, across dozens of industries. I buy across all of them at this point, which means Facebook, Google, TikTok, and YouTube are one job with different plumbing. The part of that résumé that actually matters is that I know where to find most things. The bag of instruments is the product, and the war stories are how I paid for it.

Clients don't care about any of that, and they're right not to. What does a client actually care about? Whether I understand their problem, the solutions available, the reasons those solutions usually fail, and what to do about those reasons. If I can show that, the background is a footnote, and if I can't, the background is a costume.

So what's the actual play? The method gets published and the edge doesn't. Every framework in this memo is free, including the ladder, including the instrument audit with its embarrassing rows. Nobody much cares about an operating system until they've seen what it produced, which is exactly why I want it on the record early. When somebody does start caring, I'd rather point at a document with a date on it than start explaining.

The ground moves. I watched it move under Frisco, from a dirt road to a city, while the adults insisted nothing was changing. It's moved under me more times than I can count, mostly in the wrong direction and a few times in the right one. It's moving under you right now, and some of the reason is you.

The dice were never yours. You didn't choose the box you started in. The next move is yours, and so is the one after it.

What did you do yesterday?

Andy