Thanks again for your explanations, Graeme. I think our interpretation differs in one point:You assume that clipping should be avoided in either direction for every kind of profile, I assume that clipping should be avoided only in the "native" direction (device to PCS for input profiles and PCS to device for output and display profiles).
In your view there is no clipping and only the saturation intent. The device gamuts will always be mapped to the PRMG.
In my interpretation the PCS side of an input profile is a subset of the PRMG and the PRMG is a subset of the PCS side of an output device: there is no clipping in input to output device transforms, but there may be clipping in output to output transforms (for colors outside the PRMG) and in output to input transforms (for colors outside the input gamut). If a output device is able to render colors outside the PRMG, they can only be addressed in output to output transforms.
I searched through the ICC publications but did not find any statement that conflicts with or validates one of the two interpretations.