> Normalization increases gain of all frequencies at any given point-in-time while reducing gain of all frequencies at other points in time.

When you do that then the difference between the loudest and the quietest part of the audio gets reduced. That's dynamic range reduction.

True. But.

While normalization is usually one-way, if you're doing DSP normalization & have a record of the level offsets you've applied, it's reversible. This is never the case for compression - you can't increase dynamic range of a compressed file (short of AI-generating something that never was)