Oops. big plate of humble pie for me. This is not working how I thought it was, how i was told it would, and I'd not checked it properly divy that I am. Thanks for the warning, very timely as it happens. I usually use essentially this workflow but exporting to DNxHD, which did seem to work. I could pull real data back into highlits or shadows in Resolve.
Not to stray into the minefield of encoding and decompressing video and 'limited' vs ' full range' handling by applications but if you used a rec709 (16-235) workspace in AE to DNxHD then there'd be no supers. If you used the non limited rec709 space you'd very possibly get a DNxHD encode with full range levels and therefore supers. But you can have a situation where a receiving codec like QT may scale levels say 16 - 235 to 0 - 255 giving you a brighter washed out appearance lifting shadows and pulling down highlights or a codec crushing the dark supers to RGB 0 (black) and compressing the highlight supers to RGB 255 (white) and stretching the levels out in between. Possibly giving the appearance of more latitude to play with.
The DPXs definitly don't work like this though. Is there a way to get a proper broadcast levels signal into a DPX from AE, or am I flogging a dead horse?
I'd have thought your rec709 (16-235) work space dictated proper broadcast equivalent levels in DPX output. Are you not seeing that then, the part I questioned previously was outputting limited range into DPX full range, seemed like levels scaling might occur?
However is rec709 (16-235) really what you want to develop your raws into though, yes it's your final output space but you're intermediate is DPX and your outputting through Resolve?
Have you tried a wider working space in AE, exporting as full range DPX and importing into Resolve as full range, monitoring in Resolve via a rec709 display lut / calibrated monitor and encoding to rec709 video formats from there? You'll be monitoring / previewing as rec709 (16-235) so things may appear to be clipping and crushing in preview although your scopes should tell you otherwise as the 32bit work space should ensure no clipping whilst you grade and making your choices on how you compress your DR into rec709 for final encode?
Canon raw files can contain upto 11 stops, and with some cross channel highlight reconstruction you may get a little more.
Well I was being conservative when I said hold more than 9 'stops', linear encoded 16bit tif output will hold theoretically 16 stops with decent gradation which is what really matters rather than theoretical 'stops', 10bit log I'd assume much the same but it's subjective really, first how many stops are really usable re: noise, shooting ISO will affect that, then with regard to mixing from other channels that's a bit subjective too depending on the scene DR, exposure choices and color of light source(s), which channel clips first and how quickly followed by the increased noise from the multiplied weaker channels to get the white balance 'accurate'.
In my experience a default conversion via ARC to a standard RGB space will clip some of this. And likewise if you grade for output in ARC you are likely to clip some data. But then I much prefer to do this rather than do a low contrast conversion at this point. This makes sense for us as our stuff often goes straight off for stock from AE.
There's no such thing as 'standard' RGB space :-) and it can be unbounded creating values massively bigger and smaller values than 0.0 - 1.0 display space without clipping, take ACES for example, but Canon raw is neither HDR or ACES. :-) Going back to the 32bit work space, we preview in display space 0.0 - 1.0 so we see clipping 'by eye', visually because our displays can't handle the values but the scopes will show no clipping, so it can be misleading to think data is getting clipped when in reality you can store a wide dynamic range in an intermediate file format and in a 32bit workspace but have to make choices on what to display in a limited DR output like rec709 (16-235) generally at 8bit on 6 or 8bit monitors with typically poor calibration :-)