Dither must be applied only once, at the very end of the mastering chain, when reducing word length

Dither is a low-level noise signal added to gently trim a large digital word into a smaller one (e.g. 24-bit to 16-bit for CD) instead of simply truncating, which degrades the audio. Because dither is noise, it has a cumulative effect if applied more than once, and dither introduced too early harms subsequent DSP. So it must be applied only once, at the end of the chain, just before cutting the Red Book CD or making a DAT.

Examples

A mastering chain processes at 24-bit internally; the dither plug-in is inserted last, after all EQ, compression, and limiting, only when rendering the final 16-bit master.

Assessment

At what point in a mastering chain should dither be applied, and what happens if it is applied at multiple stages?