Digital samples have a limit which you cannot exceed. If you perform math on those samples in the process of resampling, filtering, etc. you may wind up with math that results over max. Care must be taken to not allow this to happen.

Let's say you only have whole numbers to work with. I give you 9 and say to divide by 2. You get 4.5 but can't have decimal places so you round up to 5. Now I ask you to double that and you get 10, not 9! A proper system has higher internal resolution and uses proper dither to convert back and forth as to not create errors like this. In this example, your internal representation could carry 4.5 so can multiply by 2 and get 9.

Since music is loudness compressed, it is very easy to get near max PCM samples. So the excuse from companies like Arcam that these test tones are too close to max doesn't hold water. System better handle them.