This is pretty technical for right at the top? Maybe down around the new L26 would be better?

I've reduced to

The mapping is implemented through.Transform subclasses.

Because that's context needed to understand why the transform columns in the subsequent table exist.

I agree that the technical details are not relevant for users. They are basically only needed when you want to implement your own scales. That information is (although not very concise) available in theScaleBase docstring and theCustom scale example. Therefore we can leave it out of the introductory paragraph. I may later improve the custom scale example. But that's a separate topic.

I find this sentence pretty confusing on its own. Appreciate you are just changing the formatting here, but if we are improving things, we might mention that the previous scales act on one axis at a time, and are for Cartesian axes.

, and are for Cartesian axes.

This is not quite true, you can e.g set a log radial axis in a polar plot.

I'm tempted to leave out the paragraph. I think that scales and projections are mostly independent of each other (but am not completely sure).

Ok fair enough. Maybe "scales are typically applied individually to the axis of a separable axes. Non-separable axes are dealt with using projections". But maybe you are correct that you could do without it here.

I've removed that paragraph. Irrespective of the exact technical relationship. People looking at scales will typically have a certain projection (mostly cartesian axes) in mind and only consider the scaling behavior along the given axes. It's likely not helpful for them to bring up the topic of other projections here.