What Scaling Is and Is Not - Feature Scaling, Exactly

Close the boundary: feature scaling changes coordinates, not model claims.

highlighted = computed this step

Preprocessing changes what later methods see

The later distance or gradient-style method sees the scaled coordinates, not the raw coordinates. That is the whole mechanic shown here.

\text{raw coordinates}\rightarrow\text{scaled coordinates}

It is not model magic

Scaling does not train a model and does not prove accuracy, calibration, model quality, or probability truth.

\text{coordinate rewrite only}

The honest boundary

The book pins one exact preprocessing table. It is NOT training, NOT learning, NOT generalization, NOT accuracy, NOT calibration, NOT a guarantee of model quality, and NOT probability truth.

\text{exact preprocessing; deferred claims explicit}