Data and distributions
The statistical foundations everything else is built on: spread, the normal curve, why sample means behave, and what one extreme value can do to a test.
These are the terms the rest of the glossary quietly assumes. Every significance test, every confidence interval and every sample-size calculation on this site is built from a notion of spread and a claim about how sample means are distributed, and the entries here are those two things stated properly rather than taken on trust.
The load-bearing idea is the central limit theorem, and it is more useful than it is usually made to sound. It is not the claim that data are normally distributed — revenue per user emphatically is not — but that the average of enough draws is, whatever the shape underneath. That is what lets a t-test work on data that looks nothing like a bell curve, and knowing it is what tells you when the test stops being safe.
The practical entry in this group is outlier. One customer who spent forty thousand pounds can move a revenue metric further than the experiment did, and the decision about what to do with them has to be made before the data are seen or it is simply a choice about which result to get. Spread is where that shows up: variance is what a single extreme value inflates, and variance is what sample size is spent on.