Shortly after publishing my last post, I started thinking about the similarity between the scene-to-screen transforms that I showed there, and the shape of the “toe” of filmic tonemapping operators, and I realized that they’re both doing the same thing: compensating for the dim or dark surround of the viewing environment. A filmic tonemapping operator is simulating the response of film that will then be projected in a dark theatre, so it has the “scene-to-screen” transform already baked in.
Looking at the source code of the ACES Rec. 709 ODT, for example, it clearly mentions that it’s intended for a display with a dim surround, indicating that it’s already taken the scene-to-screen transform into account. This is further confirmed by the fact that the output is encoded with the inverse BT.1886 EOTF.
The upshot is that if you want to faithfully apply a filmic tonemapping operator such as ACES, you should modify it to cancel out any scene-to-screen transform that is applied on your final signal.