So why is the Prophoto space not 1.0 gamma instead of 1.8 . . .
The ProPhoto space, formerly known as "Reference Output Medium Metric RGB (ROMM RGB)", was published by Kodak. They say:
This color space is intended to be used for performing manipulations on images that exist in a rendered image state.
. . and a rendered image needs a non-linear gamma and 1.8 was chosen so as to match the independent profile connection spaces XYZ or Lab. It follows therefore that there can not be such an animal such "Linear ProPhoto". That is to say, another space might have the same white reference and the same primaries as ProPhoto but, if it's gamma is not 1.8, it ain't ProPhoto.
You can read the whole paper here: kronometric.org/phot/xfer/4Chris/Kodak%20ROMM%20white%20paper.pdf
Kodak also published a matching un-rendered space - "RIMM RGB". They say:
A new color encoding known as Reference Input Medium Metric RGB (RIMM RGB)
is defined. This color encoding is intended to be used for manipulating images that exist
in a device-independent un-rendered image state. This color encoding was chosen to
provide a large enough color gamut to encompass most common input devices, and is
defined to be suitable as an input space to the ICC profile connection space (PCS).
Examples of manipulations that might be applied in this color encoding include scene
balance algorithms, manual color/density/contrast/ tone scale adjustments, red-eye
correction, and dust/scratch removal. The color encoding is also appropriate for archiving
and/or interchanging unrendered images. 8-bit, 12-bit and 16-bit versions of this color
encoding are defined.
Curiously, RIMM has a stated encoding gamma of 1/0.45 = approx 2.2 with a small linear portion.
So it may be that "image data" is always
non-linear but, when editing, is converted to linear
and back (I think I read that somewhere).