Foundations · Interactive
The Shape of Uncertainty
Business runs on numbers that haven't happened yet — next month's sales, tomorrow's
arrivals, this project's final cost. A single guess ("about 6 days") throws away the most
important part: how wrong could we be, and how often? A distribution is the full
answer — every outcome that could happen, and how likely each one is. This page is about
why we lean on a handful of standard shapes, and how to read them. Drag everything.
Part 1 · Why bother
An average can hide the whole story
Two suppliers both promise an average lead time of 6 days. On a spreadsheet they look
identical. But you don't ship "the average" — you ship real orders, and you've promised customers
delivery within 8 days. Drag Supplier B's unpredictability and watch what the average refused
to tell you.
Same average, wildly different businesses. The spread — not the mean — is what breaks the
delivery promise, blows the budget, or empties the shelf. That gap between "what we expect" and
"what could happen" is exactly what a distribution measures.
How bad is the bad case?
The left/right tail — your worst week, your biggest overrun, your deepest loss.
What's normal vs. surprising?
Where the bulk sits, so a real signal can be told apart from ordinary noise.
How often do we breach a limit?
The probability past a threshold — a stockout, an SLA breach, a margin call.
What should we plan for?
A buffer, a safety stock, a capital reserve sized to a risk, not a hope.
Part 2 · Why these few shapes
A short list does most of the work
There are infinitely many possible shapes, yet business reuses a handful. The reason is that most
uncertainty comes from one of a few generating stories — counting yes/no tries, counting events
in a window, adding up many small wobbles, multiplying growth rates, waiting for the next event.
Each story has a characteristic shape. Learn the story and the shape comes with it. Below are the six
that cover most day-to-day business data — each one is live.
BinomialDiscrete · counts
The story: a fixed number of independent yes/no tries. How many succeed?
Conversions out of leads, defects in a batch of 100, loans that default out of a book.
PoissonDiscrete · events
The story: independent events dropping into a window of time with no fixed
ceiling. Customers arriving per hour, support tickets per day, machine faults per shift.
UniformContinuous · flat
The story: every value in a range is equally likely — total ignorance, fairness
by design. A starting point when you know the bounds but nothing else; the raw material of simulations.
NormalContinuous · the bell
The story: many small, independent wobbles added together. Measurement error,
filling weights, and — roughly — aggregated daily returns. Drag the loss line below: you're computing a
Value-at-Risk.
Log-normalContinuous · skewed
The story: many small factors multiplied together, and the value can't go below
zero. Project costs, claim sizes, incomes, prices, firm sizes. A long right tail — the average sits
above the typical case.
ExponentialContinuous · waiting
The story: the time you wait until the next random event. Gaps between arrivals,
time-to-failure, service durations. Its odd, useful trait: it's memoryless.
Part 3 · The connection
Why the bell shows up everywhere
The shapes aren't strangers — they morph into one another. The most important case: pile up enough
independent pieces and the result drifts toward the normal, whatever the pieces looked like.
Crank n on the binomial below and watch a smooth bell rise out of the bars on its own.
Binomial → NormalWatch it morph
Same yes/no story as before, but tick the box to overlay the bell it becomes.
A few tries: lumpy and lopsided. Many tries: a clean normal. This is the Central Limit Theorem,
and it's why the normal is the default for sums and averages.
↳ See it built from scratch
You watched averages pile into a bell in The Sampling Machine —
that's the same idea, run on real-looking customer data. The normal isn't magic; it's what adding
and averaging always produces.
Part 4 · The testing toolkit
Three shapes you meet when you test a claim
The six above describe data. The next three describe test statistics — the numbers you
compute to ask "is this difference real, or could it be luck?" You rarely plot them, but they're the
rulers behind confidence intervals, A/B tests, and regression. Here's what they look like and what each
one is for.
Student's tContinuous · fat tails
The bell's cautious cousin. With a small sample you're unsure of the true spread,
so t adds fatter tails — a built-in safety margin. The dashed line is the normal for comparison.
Chi-squareContinuous · positive
Add up squared wobbles and you can only get positive numbers, piled to the left.
It powers tests of variance and of whether survey categories are independent
(does region relate to product choice?).
FContinuous · a ratio
A ratio of two variances: is the variation between groups bigger than the
variation within them? The engine behind ANOVA and the overall significance of a
regression.
Part 5 · Choosing
Which one fits your situation?
You don't memorise these — you match the generating story to your problem. Answer one question
at a time and this will point you to the shape, then jump you to its playground.
Keep building intuition
The companion tools