Evolutionary biology, information theory, statistics, and deep history are not four subjects. They are four views of a single question: how order emerges, persists, and dies under constraint. Each field compresses an enormous surface area of the world into a handful of essential ideas. This page contains the handful.
Every idea below is shown in compressed form — the shortest string that carries it. Click to decompress. If, after reading the full version, you can re-derive it from the short string alone, you've learned it. That test isn't a gimmick: compression is understanding. It's idea №7 on this page, applied to the page itself.
One mechanism — selection acting on variation — and design appears without a designer, everywhere: genes, firms, habits, ideas. If something varies, replicates, and competes, these five ideas apply to it.
Anything that varies, copies itself, and competes for limited resources will evolve. Not "can" — will. The appearance of purpose and design emerges from a mechanical filter: variants that copy better become more common. No foresight, no plan, no designer required.
This idea applies far outside biology. Companies vary, replicate practices, and compete for capital. Habits vary, repeat, and compete for your hours. Ideas vary, spread, and compete for attention. All of them are subject to selection, whether anyone intends it or not.
One honest limit: selection is only as powerful as the copying is faithful. Genes copy with near-perfect fidelity; business practices and ideas copy sloppily, mutate on every retelling, and get redesigned on purpose. So outside biology, "it evolved" is often a useful redescription rather than an explanation — a lens for asking "what filter produced this?", not a proof that no one chose it.
Why bureaucracies grow rules nobody designed. Why the apps that survive are the engaging ones, not necessarily the good ones. Why "nobody decided this" deserves the follow-up: "then what selected for it?"
Selection never asks "is this good?" It asks "is this better than the rivals, right now?" There is no absolute bar to clear, only a relative one — and because your rivals are also improving, the bar moves. This is the Red Queen: running constantly just to stay in place.
It explains why nothing in a competitive system ever feels finished. Cheetahs and gazelles got faster together for millions of years; neither gained an inch of advantage. Trees grew tall not because height helps a forest — it doesn't — but because each tree only needed to out-shade its neighbor.
Arms races in advertising, college admissions, military spending, and résumé credentials. Everyone spends more; relative positions barely move. The forest canopy is the example to remember.
Anyone can claim to be strong, rich, or committed — words are cheap, so claims are noise. A signal becomes believable only when faking it would be ruinously expensive. The peacock's tail is honest precisely because a weak peacock couldn't afford to drag one around.
This changes how to interpret apparent waste. Much of what looks irrational — luxury goods, lavish weddings, four-year degrees that teach little used on the job — is information made credible by its cost. The cost is not a flaw; the cost is the information.
Why "money-back guarantee" works (costly if the product is bad). Why elite credentials persist even when the curriculum doesn't matter. Why cheap talk is discounted everywhere, from dating to diplomacy.
Selfish replicators shouldn't cooperate — yet the world is full of cooperation. Evolution found exactly two stable routes to it. First, kin: helping relatives spreads your own genes (Haldane's joke: "I'd lay down my life for two brothers or eight cousins"). Second, reciprocity: in repeated interactions with memory and reputation, cooperation beats defection.
The corollary is the important part: cooperation is conditional, not natural. Remove repetition, memory, or reputation, and it decays on schedule. Anonymity removes the conditions trust depends on.
Why people are kinder in villages than in traffic. Why online anonymity produces cruelty. Why "one-shot" deals attract scammers and long-term relationships attract honesty — change the rules, not the people.
Evolution is slow; environments now change fast. Your cravings, fears, and status instincts were calibrated for small foraging bands where sugar was rare, gossip was survival-relevant, and you knew everyone you'd ever meet. That world no longer exists; the calibration remains.
Mismatch is the one-line explanation for a startling share of modern misery: obesity (rare-calorie instincts meet abundant calories), doomscrolling (gossip instincts meet infinite gossip), status anxiety (compare-with-the-village instincts meet comparison with eight billion people).
Every supermarket, casino, and feed-ranking algorithm is built on this idea. The industries that understand your mismatches best are the ones monetizing them.
Only three ideas. Shannon answered "what is information?" with math in 1948, and the answer turned out to define learning, communication, and arguably intelligence itself. The smallest field here, and the most general.
A message carries information only to the degree it wasn't predictable in advance. "The sun rose this morning" contains almost zero information; "the sun didn't" would contain an enormous amount. Information is measured in surprise, and surprise is measured against expectation.
This gives you a precise instrument for evaluating any communication: how much of it could you have written yourself before reading it? That portion is filler. The remainder — the part you couldn't have predicted — is the actual content.
Why clichés carry no information. Why markets move only on news, never on what's already priced in. Why a person who always agrees with you is, in the technical sense, transmitting nothing.
You can only shrink a description of something by finding its structure — the regularities that let you say less and imply more. A million digits of a random string can't be compressed at all. A million digits of pi compress to a one-line formula. The compression is the understanding.
The converse is a theory of learning: to predict well, you must have compressed the past into a model. Learning, prediction, and compression are the same operation described three ways. This is not a metaphor: it is how current language models are trained, and a reasonable working definition of what a brain does.
"Explain it simply or you don't understand it" — now with a proof. Why good theories are short. Why this page tests you with compressed strings: if you can decompress them, the model is in your head.
Every physical channel — a wire, a radio band, a strand of DNA — has a maximum rate at which it can carry information, and every channel has noise. Shannon's startling result: you can transmit almost perfectly through a noisy channel, but only by adding redundancy, and only below the channel's capacity. Push past capacity and errors become unavoidable, no matter how clever you are.
Redundancy, in other words, is not waste — it is the cost of reliability. Anything that must survive noise repeats itself. (Applied to meetings, chains of command, or friendships, this is analogy, not theorem — nobody has measured the capacity of a friendship. The analogy is still a good one; just don't cite Shannon for it.)
Why DNA has error correction and aviation checklists say everything twice. Why good communicators make one point three ways. Why a reorganization that "removes redundant communication" often removes reliability instead.
Applied epistemology. The core is one fact — variance is real and your mind refuses to believe it — plus three corollaries that mislead you until learned.
Small samples don't just carry less information — they produce false patterns. Flip ten coins and streaks appear; run ten small schools and the best and worst test scores will both come from the small ones, purely because small groups swing harder. The mind reads every swing as signal.
The discipline is a single question asked every time: how many observations is this claim standing on? An anecdote is a sample of one. A vivid anecdote is still a sample of one.
Why the "hot new manager" with one good quarter is probably noise. Why miracle cures have glowing testimonials. Why every list of "habits of successful people" built on twelve interviews is not evidence.
Any measurement is part skill, part luck. An extreme result usually means the luck component was extreme — so the next measurement, with average luck, lands closer to the mean. The hot streak cools. The disastrous quarter improves. Nothing caused it; the luck just ran out, in either direction.
The trap: we intervene at the extremes. We praise after great performance, punish after terrible ones, take the medicine when symptoms peak — and then the natural drift back to average masquerades as the effect of whatever we did. Regression to the mean is the most common source of false causal claims.
Why punishment "works" and praise "backfires" (it doesn't; it's drift). Why the second album disappoints. Why the turnaround CEO hired after the worst year looks like a genius by default.
The data that reaches you has been filtered by what survived long enough to be observed. The WWII statisticians who armored bombers where returning planes weren't shot understood this: the planes hit in those spots never came back to be counted. The missing data was the important data.
Before trusting any sample, ask what process selected it into view. Successful founders write memoirs; the failed ones with identical strategies don't. Old buildings look better-made because the badly made ones fell down. Every visible population is the output of a filter you did not see.
Why "drop out like Gates did" is bad advice. Why mutual funds' average returns overstate reality (dead funds vanish from the average). Why your feed shows you winners and calls it the world.
Bayes' rule is the rule for updating belief: a belief is a probability, evidence shifts it, and the size of the shift depends on how surprising the evidence would be under each hypothesis. Two consequences are underrated.
First, the prior matters more than anyone admits: a positive test for a rare disease usually means you're still probably fine, because the rarity (the prior) dominates the test. Second, "extraordinary claims require extraordinary evidence" is not a slogan — it's arithmetic. The lesson in practice: hold beliefs at the strength the evidence warrants, not at zero or one hundred percent, and move them when the evidence moves.
Why most positive screening tests are false alarms. Why "I saw it with my own eyes" loses to base rates. Why the strongest thinkers say "I've updated" instead of "I was right all along."
No clean laws here — history is the one run of the experiment we got to observe. But the long-run economic and institutional record yields four patterns with enormous reach, and a corrective to both pessimism and hype.
Draw an arbitrary line through a single people — Korea in 1945, Germany in 1949, Nogales at the US–Mexico border — and watch the two halves diverge within a generation. Same geography, same culture, same genes. The variable that changed was the rules: who can own, who can compete, whether contracts hold, whether power rotates.
Incentive structures are upstream of much of what we attribute to national character. When you see a "lazy" workforce or a "corrupt" society, look first for the rules that make laziness rational and corruption profitable.
Two cautions. This is the Acemoglu–Robinson position, and it is argued, not settled: geography (Diamond, Sachs), disease burden, and culture (Clark, Henrich) all have serious defenders, and the debate is largely about which comes first — institutions may themselves be downstream of geography and history. And the clean split-country cases are, by №11, the survivors of a much messier literature; they're the strongest evidence precisely because they're rare.
Why the same person performs differently in two companies. Why "fix the people" fails where "fix the rules" works. The fundamental attribution error, at civilizational scale.
For nearly all of recorded history, the average human lived on the equivalent of a few dollars a day, and gains in productivity were eaten by population — the Malthusian trap. Then, around 1800, in one part of the world, growth accelerated sharply. Of the roughly five thousand years of recorded history, the last two hundred — about 4% — account for nearly all the growth in living standards.
Two lessons. 2% growth is imperceptible over a year and dominant over a century. And growth is not the natural state of things; it is an anomaly with causes, which means it can have an end.
Why every "boring" compounding process — index funds, skills, trust — is systematically underrated. Why the great historical question isn't "why is anyone poor?" but "why is anyone rich?"
The Bronze Age collapse, Rome, the Maya — none fell to a single cause. Complexity means coupling: trade routes, supply chains, and institutions that depend on each other. Coupling means a shock in one place propagates, and the failures correlate. The drought hits while the trade routes fail while the legitimacy crisis peaks — not coincidence, but contagion.
Tainter's sharper point: societies add complexity to solve problems, and each layer yields less than the last while costing more to maintain. Collapse is often just the moment the maintenance bill exceeds the budget.
Why financial crises and supply-chain failures are correlation events, not bad luck. Why "efficient" systems with no slack are prone to cascade. Redundancy again — №8, at civilizational scale.
The strong version — "the old patterns no longer apply; bubbles can't pop now; this technology suspends the rules" — has been declared before every mania and disaster on record, and has lost every time. Human incentives, leverage cycles, and crowd dynamics are the constants.
But the weak version — "the details are genuinely novel" — is always true. Printing press, railways, electricity, internet: each really did rearrange the world, just not in the way and not on the schedule the manic phase promised. The skill is holding both: the pattern repeats; the details differ.
Every market bubble and every panic about new technology. The practical test for any claim of unprecedentedness: go find the precedent first. There almost always is one.
Most of the systems on this page have feedback loops: the winner of round one gets more resources for round two. Under positive feedback, a trivial early edge — a slightly earlier start, a lucky first customer, a keyboard layout chosen for typewriter mechanics — compounds into a lock-in that later, better alternatives cannot dislodge. The outcome is stable, but it was never selected as optimal; it was selected as first.
This is the mechanism that connects the other ideas. It's how order persists (lock-in resists change long after the original reason is gone) and, because locked-in systems accumulate hidden dependencies, it's why order dies by cascade (№15) rather than by orderly replacement. Selection (№1) explains what survives; path dependence explains why what survives is so often mediocre.
QWERTY, railway gauges, the English language, the dollar. Why network-effect markets tip to one winner regardless of quality. Why "if we were starting from scratch we'd never build it this way" describes almost every institution — and why starting from scratch is almost never an option.
This is the part Munger was really pointing at. These are not four separate toolkits; each depends on the others.
The models above are learnable in a year. Munger's actual advantage was the discipline of checking the list when excited or afraid — which no amount of reading installs, including reading this page. The only known fix is mechanical: put the questions below somewhere you are physically forced to look before committing money, time, or reputation — the top of a decision memo, a pre-commit template, a note taped to the screen. A checklist that lives only in memory is one you will skip precisely when it matters, because excitement and fear impair recall.
Is this an arms race? Am I spending to get ahead, or spending to stand still while everyone else spends too?
Is this signal costly? What would it cost them to say this if it were false? If nothing — it's noise.
How big is the sample, and is this an extreme? If it's a small sample at an extreme, expect drift back to average — and don't credit whatever I do in between.
What filtered this into my view? Who failed with the same strategy and isn't here to tell me?
Am I respecting compounding? Is there a boring 2%-a-year option I'm ignoring because it's invisible at the one-year scale?
Have I looked for the precedent? Before accepting "this time is different," did I actually go find the last three times someone said that?
Is this winning because it's best, or because it was first? If I'm betting on the incumbent, is its edge quality or lock-in — and if I'm betting against it, do I have a way past the lock-in?