Introduced on day 21
Structural break
A date after which the process that generates the series changes, for example a level shift after a competitor adds capacity, so earlier history describes a different regime.
In detail
A break can change the level, the slope, the seasonal shape or the variance. Averaging across it blends two regimes, and a rolling mean or a fitted trend over the break window looks like a slope. Treat it as a modelling decision: date it, look for the operational cause, then compare a step or interaction regressor with training on the post-break window on origins after the break.