I’m not really sure if we are talking about the same thing. I should have quoted earlier what my response all was about:
Yes, the tone curve is baked in, as are many other transformations (most easily seen, I think, are lens-corrections, sharpening, white-balance and color-science, and of course the tone curve we are talking about). But there is no reason why scene-referred modules will not be applicable because of this.
Consider this counterexample: I have one instance of tone equalizer, compressing shadows and highlights, improving contrast in midtones, just doing so by hand.
From now on, for all later modules in the pixel-pipe, we are working on data with a tone-curve baked in.
But I still can use another tone equalizer instance, or exposure, or color balance rgb. The existence of a non-linear tone-mapping step does not make it “technically inadvisable” to later do further scene-referred operations.
Of course I can perform steps early in the pixelpipe that will cause artifacts later. And in the case of jpegs, such transforms have been done, and in such a way that they cannot be undone (I cannot uncompress the highlights, for instance, and as such S-curve-like adjustments like filmic, sigmoid, AgX are usually not helpful).
But whether a module works or not is less to do with “scene-referred” or “display-referred”, and more whether the specific transformation can be applied as intended, with little enough artifacts.