Hi,
I have been looking at reducing the error between CTL and OCIO by testing various covered ranges for the Log2 48 nits Shaper in the config generator and here are some images and numbers.
On the left is the CTL render reference and next images are OCIO renders the Log2 48 nits Shaper domain varying. The second row is the difference between the CTL and OCIO multiplied by 10.
Over-Exposure
Base
The takeaway is that increasing the covered range increases the error of well-exposed image areas but it also reduces significantly the error of saturated and over-exposed areas. Ideally, we should test with more sample images to avoid bias that said it is quite a bit of work. In any case, I would probably be inclined to use [-6.5, +10] instead of the current [-6.5, +6.5] for the Log2 48 nits Shaper.
I might look at how an ACEScc shaper would perform too.
Cheers,
Thomas