There are multiple answers to this one. For myself,
As it stands now, a flat pass EXR merged in IMF with sound, sub-titles, and captions is a deliverable that
could be expected out of current standards. It is also true that many shows need both the ‘texted’ version (which can include the VFX versions of road signs ) and the ‘non-texted’ version.
Your ideas though about a per shot basis also are sound. Having handles later in the process could help in several situations. Having more metadata in the archive could ease repurposing. For example, having
the mattes tracked by the colorist , on say the sky, could assist in latter mapping to a different HDR range.
I would think of the pre-grade archive as most appropriate for the per shot approach. In theory this should hold all of the camera range that was available. All of the sidecar elements would apply.
It would be nice if there was a fully metadata driven color mastering process, but so far, that isn’t a sharable thing.
The post-grade version could be the IMF assembly mentioned above.
In either, I think keeping the original images in AP0 is a good idea.