A 99% accurate test that is usually wrong
Ten thousand people, one test each. Set how common the condition is and how good the test is, then look at everyone who tested positive: how many of them really have it? Make the condition rarer and watch what a positive result is worth.
The rules, all of them. Each person has the condition with the chance you set, independently. Everyone is tested once. Someone who has it tests positive with the first percentage; someone who does not tests positive with the second. Nothing else is known about anyone.
The page shows the exact answer from Bayes' rule beside this crowd's own count; a single crowd of 10,000 wobbles, especially when the condition is rare. What it leaves out: real tests are rarely equally good in both directions, a repeat test is not fully independent of the first, and nobody is a random person: age, symptoms and history all change the starting figure. This is arithmetic, not medical advice.