Judgment · 7 min read
Your Number Was Never a Number
A single growth rate in a deck feels like knowledge. It is almost always a cloud wearing the costume of a point — and the confident conclusions hanging off it survive in far fewer futures than they seem to.
The number is always suspiciously round. Twelve percent a year. It goes into the deck, into the model, into the sentence you say out loud to the board — revenue grows twelve percent a year — and from that one number a dozen conclusions hang: the hire you can afford in the third quarter, the round you won't need to raise, the date you cross a million. The number feels like knowledge. It is doing the work of knowledge. But ask the person who wrote it the only question that matters — how sure are you? — and the honest answer is never twelve. It is something closer to five to eighteen, most years, probably.
That honest answer is a different kind of object. Twelve percent is a point. Five to eighteen is a cloud. And the moment you admit the cloud, something uncomfortable follows: nearly every confident conclusion you drew was balancing on the point — and the point was the part of your assumption you had the least right to trust. The Fog Behind the Number lets you watch a single projection come apart into the ten thousand honest versions of itself, and count how many of them still agree with the call you made.
Say what you expect, then admit you’re not sure of it — and watch your one confident number fan out into every future it was always hiding. Works for anything that compounds: revenue, savings, users, a market.
Drag this wider the less sure you really are. It’s not a guess at your confidence — it’s the span you’d put your name to.
What kind of surprises?
surprises are small and even. Try Crash risk and watch a few lines plunge — the rare disaster you never put in the plan.
Each faint line is one possible future — your best guess, but rolled forward at a different rate from your range. The bold lineis the single number you’d have written in the plan.
The red line on the chart. Your plan lands at €343k — drag right if the decision needs you to beat it, left if just surviving is enough. Futures ending above the red line are the ones where your bet pays.
Your plan
€343k
the one number you'd write down
Where most end up
€230k–€499k
the middle 90% of futures
Your bet pays in
50%
of ten thousand futures
A coin flip. Your call sits right in the middle of the spray — about as likely to miss as to hit.
And the answer barely holds still. Nudge your best guess up a single point — 8% to 9% — and your bet’s odds move by +11 points. One percent you couldn’t really be sure of, and the conclusion swings — that’s how much was resting on a number picked out of a fog.
Ten thousand futures, each handed one fixed growth rate drawn from your range, then compounded to the horizon. This is the mirror image of The Monte Carlo Fan, which shakes the path around a return you claim to know — here the return itselfis the thing you don’t know. Illustrative, not a forecast of any real business.
Risk you can measure, uncertainty you can't
A century ago the economist Frank Knight drew a line still worth drawing: between risk, where you don't know the outcome but you do know the odds — a roulette wheel, an actuarial table — and uncertainty, where you don't even know the odds, because the situation is too novel, too tangled, or too one-off to have a stable frequency at all. Almost every decision that actually keeps you awake — this market, this hire, this launch — lives on the second side of that line. And the second side has no true "twelve percent" waiting to be found. There is only a range of belief, wider or narrower depending on how much you really know.
Nassim Taleb gave the mistake of confusing the two a name: the ludic fallacy, from ludus, Latin for game. It is the error of taking the clean, closed probabilities of the casino — where the dice have exactly six sides and the house edge is known to the decimal — as a working model for the open, messy world outside it. A plan written as a single growth rate commits a quiet version of it: it dresses an uncertainty in the tidy clothes of a risk and hands you a number precise enough to build on.
The stubbornest overconfidence is about precision
Psychologists sort overconfidence into three kinds, and they don't behave alike. There is overestimation — thinking you'll do better than you will. There is overplacement — the "I'm an above-average driver" illusion. And there is overprecision — being too sure your estimate is close to right, drawing your error bars too tight around whatever number you landed on. Of the three, overprecision is the most robust and the hardest to train away. It is also the one a point estimate is made of: the whole act of writing "twelve percent" instead of "five to eighteen" is overprecision turned into a slide.
You can feel it directly. Asked for a range you are ninety-eight percent sure contains some quantity you're unsure of — the length of the Nile, next year's churn — people hand back one so narrow that the truth falls outside it something like forty percent of the time, not two. The interval that would actually be right feels absurdly, uselessly wide. That gap, between the range you'd offer and the range you'd need, is overprecision with a number on it. The Calibration Game will show you yours.
Foxes give ranges. Hedgehogs give numbers.
When Philip Tetlock spent two decades scoring the predictions of experts, the ones who did best were not the ones with the boldest single calls. They were the thinkers he named foxes — who held many small, provisional ideas, hedged, and spoke in probabilities and ranges rather than confident points. The hedgehogs, who ran everything through one big idea, gave better television and worse forecasts. His later superforecasters sharpened the same habit: they quote granular odds, widen and narrow them on purpose, and update the moment the world gives them a reason.
The better forecasters aren't more certain. They're more honest about the width of the fog.
The honest doubt
Two ways this move can mislead, worth saying out loud. First, a distribution is only ever as good as the range you feed it: a made-up "five to eighteen" buys you a fancier false precision — a smooth curve of nonsense. The cloud does not manufacture knowledge you lack; it only stops you pretending the point was knowledge you had. Second, some argue that under deep uncertainty elaborate probability models are the wrong tool altogether — that simple rules, robust to being wrong, beat fragile optimizations. On this coin that critic is an ally, not an opponent: the aim is not a better number, it is less faith in any single one.
How to read your own number
None of this says stop forecasting. A point estimate is a fine summary and a poor foundation. The fix is small: whenever a number is about to carry a decision, make it show its range, and test the decision against the range instead of the point.
You can run your own number through the fog: The Fog Behind the Number. Put in the figure you'd stake something on, then the range you'd actually defend, and watch the single confident line come apart into the ten thousand versions you were always choosing among — most of which you'd never have said out loud. For the same lesson aimed at track records and reputations instead of projections, its companion is The Genius Trap; for the difference between a wild path and a wild average, The Monte Carlo Fan.