TF with out of bound values
Floating point pixel formats can reach values outside of [0, 1]. Only specific transfer functions are defined outside those bounds (linear). The protocol should define what happens to the pixel values when they are out of bound for the specified TF.
My opinion is that values should be clamped to the well defined range of the selected TF. In practice this means if one wants to use values outside of [0,1] one has to use the linear TF.