This article should be pinned to the sidebar. It's so on point with excellent quotes:

oscars.org/aces/spotlight-aces-vfx

@troy_s So you mean, it's a usefull way to save linear display images that can be color grade or transferred to an other color space without to save it as 32bit exr?

@Mareck It is indeed useful on a number of fronts, not all are obvious at first. It is a very efficient encoding format in that it "compresses" the data in line with a perceptual / nonlinear way. So that's point one: Useful for encoding at limited bit depths.

@Mareck It is also useful to look at. Given that it isn't an aesthetic view of your data, it's sort of a "We are trying to give each value in your data the same amount of 'say' so you can see it." That is, it is useful to give a visual representation of your data without deviating from the relative look of the scene.

@Mareck Finally, it is a technological tool. When we think of 3D LUTs, we have a bounded range that they can operate on, and interpolate across the values. If we scale the data down linearly, we are cramming the "bulk" of our image into very little range, which means we are relying far too heavily on interpolated values across a very small range. The solution? Use a quantisation that more closely maps to what we are going to be staring at.

@Mareck So lastly, it is a useful technological weapon to perform colour manipulations within the constraints of contemporary computing.

if we only had proper DPX support in Blender...

@Mareck If you add a single ASC CDL node after your work, grade accordingly, you can take those values direct and make a "Look" using a CDLTransform. Process space would need to be set to Linear in order to match Blender's pipeline.

@Mareck Essentially you should be able to make a perfect 1:1 by chaining the CDL into a look. I can explain further if you manage to get a demo CDL done. I think the most ideal way would be to make a view and use the + to add Look transforms.

@Mareck Pure primaries mean primaries missing one or two compliments. That is X,0,0 or X,X,0, etc.

@Mareck Example is to make say, an albedo diffuse value of 0.7, 0, 0, and smash it with light. Then more light. Then more light. Many folks experience cognitive dissonance when they do this based on what they expect to happen.

@Mareck That expectation is actually purely learned and psychological, and is why the desaturation aspect of a view transform is important.

@Mareck It applies to all "pure" primary combinations, and all will exhibit that brokenness.

@Mareck Night!

Greetings @JordanSzubert