Stationarity
A property of a series whose statistical behaviour does not drift over time — its mean and variance stay put.
Most classical time-series methods assume it, and most real series don't satisfy it. Anything with a trend isn't stationary, because its mean is moving; anything whose volatility grows with its level isn't either, because its variance is.
Two transforms get you closer. Differencing — modelling the change rather than the level — removes a trend, and a second difference removes a curved one. A log turns multiplicative growth into additive growth, which stabilises a variance that scales with the mean.
Worth separating from drift. Non-stationarity is a property of the series you already hold, visible in the data you have. Drift is the world changing after you fitted, so a stationary training window can still be followed by a shifted production one.