In 1989 the investor Warren Buffett put a piece of grade-school arithmetic into his letter to shareholders. One dollar, doubling every year, for twenty years. Two ways to play it. Either you sell at the end of every year, hand over 34% of the gain in tax, and reinvest what is left; or you never touch it and settle the tax once, at the end of year twenty. The first route leaves you with roughly $25,200. The second leaves you with roughly $692,000.
Same asset, same rate of return, same tax rate. Twenty-seven times apart. The only difference is whether you reach for it once a year.
That table sits on the investing site's page about portfolio management, printed next to a line from Buffett's partner Charlie Munger: the first rule of compounding is never to interrupt it unnecessarily.
The same word is lying on several other tables. The leadership site says firefighting, repeated labour and cleaning up after other people are linear spend — done this week, back to zero next week — and that only deep skill, reputation, networks and reusable work actually compound. The writing site is blunter: individual pieces do not add, they multiply, and after ten years you do not own five hundred loose drafts but a body of work that earns while you sleep. The mental-models site files it as a card: growth acting on growth.
All of these sound right. But that 27× is a number you can get out of a calculator, and "reputation compounds" gets you no number at all. Is the relationship a metaphor, or is it one thing?
What makes the question worth asking is that the answer is not "real in finance, a figure of speech everywhere else." Finance has a clean counterexample of its own. Berkshire, Buffett's company, owns a candy business with a return on capital that is frankly ridiculous, which Buffett and Munger adore, and which neither of them has ever counted as a compounding machine.
The investing site's page on quality investing opens by swapping the question. The old cigar-butt school asks how cheap a company is. Quality investing asks instead: at what rate can this business put its own profits back to work, and for how long can it keep doing that?
The page carries an arithmetic table — one unit of retained capital rolled for thirty years at the business's own ROIC, the return on invested capital, meaning how much the business earns in a year for every unit put into it. At 6%, that unit becomes 5.7. At 12%, thirty. At 18%, a hundred and forty-three. At 25%, eight hundred and eight. This is the arithmetic behind Munger's well-known remark that over the long run a stock has a hard time earning much more than the business underneath it earns: buy a 6% business at a huge discount and hold it forty years, and you still end up with something close to 6%.
Everything turns on the small print under that table. The premise is that the profits can keep being reinvested at that same rate.
See's Candies is what happens when they cannot. Between 1972 and 2007 it handed Berkshire more than $1.35 billion in pre-tax profit, on under $40 million of additional capital put in across the whole period. By any standard, a superb business. But it cannot absorb much incremental capital — open enough candy shops and the return on the marginal shop collapses. So it pushes its cash out every year and Buffett allocates it somewhere else.
The source page lands on a hard sentence: a high ROIC tells you how good the business is; the reinvestment runway tells you how long compounding can run. You need both. See's has the first half only. It is an excellent business, and it does not compound — not because it isn't good enough, but because the money it makes has nowhere to go.
Which exposes the real definition. Compounding has nothing to do with how fast something grows, and nothing to do with the word "exponential." It asks one question: can this round's output become next round's input?
The personal-finance site comes at this from the mirror side. Its page on good debt versus bad debt puts it in one line: investing compounds money on your behalf; debt compounds in reverse, and puts you to work for the lender.
The yardstick it offers is not moral, it is that percentage. Compare a loan's annual rate against the after-tax return you can reliably earn over the long run. Paying off a 22% balance locks in a risk-free 22%, and no investment hands you that reliably. So the high-rate debt dies first, while debt cheap enough to sit near your long-run return — a mortgage locked in low, say — can comfortably run in parallel. The page makes one mechanical point worth keeping: once a credit card balance starts revolving, new purchases accrue from the day you swipe, and a rate north of 20% is compounded daily. That is not a figure of speech. Daily compounding is how the statement is actually computed.
Less obvious is the spending side. The site's first-principles page makes an argument I think is badly underrated: every recurring expense compounds against you, and it does so twice. Once as the decades of growth that money will now never have. And once as the way it raises your baseline cost of living, and therefore raises the total you need to accumulate before you are free of work. Under a common estimate — drawing 4% a year from your assets in retirement — every extra 10,000 of annual spending adds 250,000 to the target. The page follows this to the savings rate: the share of income you don't spend and do invest. It works from both ends at once, growing the pile faster while shrinking the pile you need, which is why it dominates investment returns. Lift the savings rate from 10% to 30% and the timeline to financial independence drops from roughly fifty-one years to roughly twenty-eight. The page flags, correctly, that this is a model and not a promise — returns wobble, lives change — but the direction is very stable.
Put these pages next to the previous section and the signature of the financial side is clear. Every quantity in the loop has a unit, the unit is money, and you can look all of them up. So it gives you the shape of the curve and the points on it.
The next three sites are the borrowers. What is interesting is that they do not borrow equally loosely.
The page on playing the long game uses compounding as an inventory exercise. It sorts time into two buckets. The linear one holds firefighting, repeated labour and clearing up after other people, all of which reset to zero when you finish. The compounding one holds deep skill, reputation, networks and reusable work. Then it supplies a test question, which I think is the closest thing on the site to naming a reinvestment channel: will this still be producing a return for me a year from now?
Its examples are concrete, and the return comes back around a bend. The documentation you write gets used repeatedly by other people; the mechanism you build makes the team ship faster; those become your reputation and your standing, and standing is what makes the next mechanism easier to push through. The channel has a name. The page also names the failure case: going deeper and deeper into a stack that is on its way out is leverage applied to a depreciating asset, which is another way of saying the marginal return has already collapsed. Then there is the middle stretch, which it calls the valley of disappointment, after a line from James Clear, author of Atomic Habits: habits often appear to make no difference until you cross some critical threshold. The page resists the obvious move here. It does not make persistence the answer. It insists you distinguish "not yet past the threshold" from "wrong bet, cut it," and gives you the test: does the core assumption you bet on still hold, and are the leading indicators moving at all?
The writing site describes the same loop most explicitly, and does it twice. Its page on delivery splits compounding into three stacked layers. Craft: the feel you have writing your three-hundredth piece cannot be bought at the first. Asset: a good piece keeps being searched, quoted and forwarded, working while you sleep. Relationships: the thicker the body of work, the more people arrive because of it.
The page on writing as a vocation takes the mechanism further, to the point where you could act on it. What is scarce in the new world, it argues, is no longer permission to publish but being found — and search and algorithms favour only what is continuous, indexable and linkable. So the strategy is to keep publishing small things, manufacturing a large searchable surface area, so that the right person collides with you at the right moment. It writes the causal chain out: publish steadily, get found, get linked, compound. The illustration on the page is the giveaway — six months of small posts leaves you twenty-five entrances, exactly one of which someone stumbles into, and it is the one you would never have bet on.
This is a real loop. Old posts get found, which brings readers and links, and links raise the odds the next post gets found. It is not "keep at it and good things happen." It names what turns into what at each step.
The mental-models site files compounding as a model card. It is the most abstract treatment of the five, and therefore the one closest to the edge.
Its example is the classic one: a dollar doubling daily passes a billion in thirty days — precisely, $1,073,741,824. On day twenty you have only just cleared a million. The last ten days supply 99.9% of the total growth. The page adds another data point: Buffett started investing at eleven, and 99% of his net worth was accumulated after fifty. Both are making the same claim. The secret of compounding is not the rate, it is the duration, and the fatal mistake is breaking the chain.
But the same page carries another example, and it is worth transcribing exactly, because it performs the characteristic accident of this whole line of thinking. Use AI tools to get 5% more efficient every day, it says; tiny, but compounded daily it implies a multiple after one year that is frankly absurd. Then, in parentheses: of course there is a real ceiling.
That parenthesis is the most honest thing on the card. It concedes that there is no reinvestment channel here — the 5% you gain today does not become the principal on which tomorrow's 5% is earned. Those gains are adding, not multiplying. And efficiency has a ceiling, while the mathematics of compounding does not recognise ceilings. The page discounts its own claim, and that discount has to travel with it.
The same site has a sentence I find more useful than the compounding card itself, over on the page about business models: the cost of unit N+1 determines the shape of your compounding curve. A high-marginal-cost business pays again for every customer it serves; a low-marginal-cost one, once built, serves unit N+1 for almost nothing. That is a statement about how wide the reinvestment channel is.
Stack the five sites on top of each other and the loop has three parts.
Together these explain something that otherwise looks psychological: why a compounding curve is always flat first and steep later. Round n's input is the accumulated output of all the rounds before it, so the steepness itself accumulates. The valley of disappointment is not a problem of willpower; it is an arithmetic consequence of ①. The leadership site and the writing site each ran into that plateau independently, and they are describing the same shape.
What matters more is that the three parts work as a test, because the verdicts they return are counterintuitive.
They rule a superb business non-compounding — and they are right, since that is Buffett's own conclusion rather than something we reverse-engineered. They simultaneously rule the writer's body of work compounding, even though not one number can be computed there. So the line does not run between finance and everything else. It runs between loop and no loop. This is the whole point: a test that only ever certifies money and dismisses everything else as metaphor is a territorial claim, not a mechanism. It earns the name mechanism precisely by catching a See's Candies.
If all this proved were that five sites are describing one loop, it would be a curiosity. The value is that once the loop matches, decades of hard-won results on the financial side can be collected rather than reinvented. Three are worth collecting.
The portfolio-management page carries the Kelly criterion, given by the Bell Labs researcher John Kelly in 1956. It maximises not the expected value of a single bet but the long-run geometric growth rate — which is to say, compounding. The reason is the loop: investing multiplies rather than adds, and a single −100% wipes out everything that came before. From which follows a genuinely counterintuitive result. Past a certain fraction, a bigger bet does not buy faster money, it buys a nearer zero: the risk of ruin climbs steeply and the long-run growth rate falls. The page notes the practical response — since your true odds are always an estimate, most practitioners use half Kelly, giving up a little growth for a large buffer against being wrong about the odds.
Once you read that as a property of multiplicative processes in general, it travels wherever the loop travels. The leadership site's advice to carve a milestone visible within the current review cycle out of any two-year bet is, mathematically, the shape of half Kelly: even if the two-year bet has positive expected value, putting all your time on it can get you eliminated before it pays. That is not a counsel of moderation. It is the same inequality.
What you usually get about plateaus is encouragement. The deferred-tax table offers something else: it prices the interruption, and shows that the price is not the loss on the day but every year that principal would have kept rolling. Not a cent of that 27× came from better stock picking.
Carry it to writing and the word "persist" acquires content. The post that gets found at number eighty is found because of the indexable surface area left by the seventy-nine before it; stop for two years and what you lose is not two years of posts, but every subsequent year in which those posts would have been bringing people in. The same accounting is why the leadership site's observation about glue work going unrewarded is a structural problem rather than a grievance: what the review cycle fails to count is not the favour, it is everything the favour would have come back as.
The idea of a reinvestment runway supplies a ready-made diagnostic move: don't ask about the average return, ask about the return on the marginal unit of input. The investing site uses it to catch a particular kind of fraud — management acquiring for the sake of scale, putting cash that should have been returned into mediocre assets, while the "high ROIC" label stays on the tin long after the marginal return has caved in.
Point the same move at a person and you get the diagnostic version of the leadership site's line about leveraging a depreciating asset. Going deeper into a stack that is dying still shows a high average return; the return on this hour is already negative. The question doesn't change: does the next unit you put in still earn what the first one did?
There is one more thing to collect, running the other way. Paying off a 22% balance locks in a risk-free 22% — move that yardstick out of finance and it becomes: find the loop that is running in reverse and kill it before you optimise anything else. A standing meeting that eats ten hours a week and decides nothing is, structurally, the same entry on the books as a revolving balance.
First, to be fair to the borrowers: none of the three sites above is talking nonsense. Leadership and writing both name their channels — documentation gets reused, old posts get found — and that is competent usage, not decoration. The mental-models site discounts its own 5% example in the parentheses, which puts the test right there on the page. What actually goes wrong is the sentence that cannot name a channel at all.
So the tests are the three parts turned into answerable questions, plus three about units and scale.
One more, which isn't on the list but is worth holding on to: ask who collects the interest. With relationships and reputation the beneficiary is never only you, which is exactly why calling them "investments" is both accurate and faintly unpleasant. The leadership site handles this well by talking about repeated games and giving first, rather than treating people as a portfolio. Describing a human loop entirely in the language of compounding is not technically wrong. It just leaves out more than it computes.
When the quantities have units, go and compute them. Savings rate, the APR on a debt, the tax realised by turnover, ROIC and the length of the runway: every term here can be looked up, which is why this tier produces answers rather than attitudes. The personal-finance site's four check-up ratios are the model for it — balances flatter you, ratios tell the truth. The conclusions tend to be boring once the arithmetic is done: kill the 22% balance, then talk about asset allocation.
When the channel can be named, use it as a model. A body of work, a reputation, the people you have trained, documentation others reuse. Two requirements: name the channel explicitly, and accept that you get the shape and not the numbers. A quick self-test is to ask where round two's input comes from. If you can replace the word "compounding" in your sentence with a specific causal chain and lose nothing, you have borrowed the loop. If you can't, you borrowed the word for its authority.
Everything else is rhetoric, and should be called that. There is nothing wrong with rhetoric; just watch what the word is doing in the sentence. If it explains — why the early stretch is flat, where the steep part shows up, which link in the chain carries the return — it is working. If it is there to make you endure ("keep going, it compounds"), it is borrowing the authority of mathematics for an encouragement with no content, and quietly covering up the question that should have been asked: is there a channel here at all?
Which brings it back to the candy business. Its margins are enviable, it threw off more than $1.35 billion over thirty-five years, and it needed under $40 million of new capital to do it. By any definition of "good" it is almost embarrassingly good. And it does not compound, for a reason flat enough to be deflating: the money it makes has nowhere to go.
Translated out of finance, that is the line worth taking away. Compounding does not reward effort and it does not reward quality. It rewards one thing only — whether output actually becomes input. So the question is never whether you are working hard enough. It is which door the thing you made this round comes back through tomorrow.