Simplifying somewhat, the intuition is that once we are in a colorimetric color space like XYZ (or even sRGB) we can look at the span of possible tones in the space as being contained in a cube with the origin at the minimum possible value (zero, blocked black) and the opposite apex at the maximum possible value (clipped white, normalized to 1). Negative and clipped values cannot be rendered, so they are swept under the carpet.
To get to the final color space you need to project from cube to cube via matrix multiplication (WB, camera->XYZ->ProPhoto->XYZ->xRGB). Every projection can and often does result in some negative and clipped values that need to be discarded. There is very little one can do about this if one wants to keep things as linear as possible, and we often do (right Glenn?). What little one can do is often quite a subjective fudge. So I like to know what tones those are.
… which is however under the direct control of the user so easier to subjectively guide into final ‘gamut’. For instance some form of contrast, which is pretty well needed in every single image in order to better squeeze its tones into the smaller contrast ratio of the typical display medium today. If not done properly (no way is perfect nor necessarily more or less desirable) it will substantially shift chromaticity and increase saturation, with consequences for the final tone range.