Free interactive learning tool

The Base Rate Machine

A test that's 99% accurate sounds close to infallible. So here's the puzzle that catches doctors, security analysts, and executives alike: when a 99%-accurate test for a rare condition comes back positive, what's the chance it's actually true? Guess first. Then run the machine and count for yourself.

Guess first — commit before you count

A disease affects 1 in 100 people. The test is 99% accurate (it's right 99% of the time, for both the sick and the healthy). Your result comes back positive. How likely is it you actually have the disease?

The machine — 1,000 people, recounted live
Has it, flagged – Has it, missed – False alarm – Healthy, cleared –
How rare is it? (base rate) 1%
Share of the 1,000 who really have the condition
How accurate is the test? 99%
Applies both ways: catching the sick and clearing the healthy
If your test says positive…
–
Where this bites — the same math, wearing different clothes
SOC / IT

Alert fatigue

Real intrusions are rare, so even a good detection rule drowns analysts in false alarms — and the team learns to ignore the console. Tuning the base rate (better scoping, suppressing known-benign) beats chasing another point of accuracy.

Medicine

Screening programs

This is why population-wide screening for rare cancers is controversial: most positives are false, and each one triggers anxiety, biopsies, and cost. Screening works best in higher-risk groups — where the base rate is higher.

Fraud & risk

Blocking good customers

Fraud is a fraction of a percent of transactions. Flag aggressively and nearly everyone you decline is a legitimate customer having a terrible day. The false-positive cost sits outside the fraud team's dashboard — that's what makes it dangerous.

Hiring

The brilliant-signal myth

"Our interview loop is really good at spotting top performers" — but if genuine great fits are rare in the applicant pool, most candidates your loop flags as stars are false positives with good interview skills. Work samples raise the base rate before the interview ever happens.

The machine's lesson compresses to one sentence: a test's usefulness depends on what you point it at, not just how good it is. When the thing you're hunting is rare, even an excellent detector mostly finds ghosts — so before trusting any positive, ask "how common is this to begin with?" And when someone quotes you an accuracy figure, ask the follow-up that separates statisticians from slide decks: "out of everyone flagged, how many are real?"