Figure 3.13. Linear encoding of scene luminance in 16 bit code values.

The 90% white point is mapped to 1/16 of the maximum code value giving the headroom that is needed to represent the highlights.