Right now the second argument of array-curry specifies the outer
dimension of the result, so if the first argument, an array, has a
$d$-dimensional interval as its domain and the second argument is $k$,
array-curry returns a $k$-dimensional array (what I call the outer
dimension), each entry of which is a $d-k$-dimensional array (what I
call the inner dimension).

It happens that for the two examples that use array-curry, it seems more
natural to specify the inner, rather than the outer, dimension.

When showing slices of a CT or MRI scan, you're always showing
two-dimensional slices, whether the underlying data has three or four
dimensions (volume data or time-dependent volume data).

Similarly, to construct a separable transform from a one-dimensional
transform, the inner dimension is always 1 (you're always applying a
one-dimensional transform) no matter the dimension of the "base" data
(images, volumes, ...).

Which leads me to wonder:

Should array-curry specify the outer dimension or the inner dimension?

Brad